Murray Campbell lecture at University of Washington

[brucemo]

I just attended a lecture by Murray Campbell, one of the folks
responsible for Deep Blue.

I will not attempt to provide a summary of the entire lecture, or of my
questions to Campbell after the lecture, but I will provide some
information that I think will be of interest to those who read this
group. I took notes, and will try very hard to be accurate, but there
may be mistakes, so I apologize for any of these.

I asked about the famous 10-game match and go the following information
back.

Ten games total were played between "Deep Blue Jr" and two PC programs,
Rebel 7 and Genius 5.

The configuration of DB Jr that was used in this instance was a
single-processor version searching 2 million nodes per second, with some
search extensions turned off.

The PC programs were running on P5/90's. The time controls were
anywhere between 10 0 and 30 0.

The score of this "just so happened to be" 10-0. I didn't get the
impression that Campbell was suggesting that this would happen every
time.

There were 30 other games played between DB Jr with between 1 and 16
processors. The score of these games was 28-2.

I may have made a mistake. I might be reporting the NPS results for the
16-processor variant, but I don't think so.

In any case, I conclude from this that DB Jr had a significant hardware
advantage in this match, so that those who want to believe that their
favorite PC program can't be stomped 10-0 by something remotely similar
can breathe a little easier. The time controls were fast and the
machine had some advantages. So perhaps we can stop fighting about
this.

But on the other hand, it sounds to me like those of you who say that
"Deep Blue's eval sucks" are wrong.

Deep Thought was a machine with a great search and an "impoverished"
evaluation function. It doesn't sound like they've changed the search
much between Deep Thought and Deep Blue. In my opinion their search
sounds pretty simple, other than for the addition of singular extension.

In the first Kasparov match I do not know if they had improved the
evaluation function much either.

But they spent the year between the two matches doing evaluation
function work. It sounds like they left the search pretty much alone,
and went at the evaluation function like crazy.

The Deep Blue evaluation function has 8000 "feature recognizers".
Exactly what these do wasn't described very thoroughly, although the
example Campbell gave involved detecting a doubled pawn. Obviously
there would be much more complicated ones as well. I don't know exactly
how this compares to the number of terms in a micro program, which can
get very large if you start adding multi-dimensional piece-square
tables, but in my opinion their evaluation function does not sound at
all impoverished anymore.

In response to a question, Campbell stated that perhaps half of these
features come up often, and the rest were more special-purpose (rare).

In order to add a new feature, they have to produce a new chip. But
they can affect the values assigned to these features at will.

The values from features are summed in non-linear fashion. This was not
elaborated upon, but my sense is that this means that the value of one
feature can have an impact upon the values of other features.

The DB book is hand-tuned by grandmasters, is checked by Deep Blue, and
doesn't sound like it was very large. There is a backup book including
600,000 distilled games, which is accessed if the other book is exited.

They were girding for a big opening book battle with Kasparov, but it
sounds like he evaded their book pretty well.

The opening in game 6 was part of the hand-tuned book. They added that
sac thinking they would never get into this part of the book, and DB
liked the resulting position anyway, so they kept it.

The DB that played Kasparov searches 200,000,000 nps, up to almost
double this upon occasion.

They have 20 gigabytes of endgame databases courtesy of Thompson and
Stiller, including Thompson's 5-man databases, and some unknown number
of 6-man databases from Stiller (probably not many (just one?) in only
20 gigabytes, IMHO).

They have no plans to go to the WCCC next year, but anything is
possible.

They will be publishing some stuff in a couple of months.

Someone asked about changing the program between rounds, and Campbell
said that they had made changes after the first game loss, which were
installed "before game 3", which leaves something interesting unsaid --
perhaps the version that played game 2 was identical with the one that
played game 1?

In later rounds they increased their king safety values a bit.

It sounds like Campbell has moved on to other things at IBM.

These were the salient points as far as I am concerned.

bruce

[Rolf Tueschen]

brucemo <bru...@seanet.com> wrote:

>I just attended a lecture by Murray Campbell, one of the folks
>responsible for Deep Blue.

Ed Schroder, please do comment on this post.

I hereby promise NOT to give any commentary on whtever you may write
about this.

But we all are interested in what you might think now after we go these
news...

Thanks.

Rolf Tueschen

[Robert Hyatt]

brucemo (bru...@seanet.com) wrote:
: I just attended a lecture by Murray Campbell, one of the folks
: responsible for Deep Blue.

: I will not attempt to provide a summary of the entire lecture, or of my

: questions to Campbell after the lecture, but I will provide some
: information that I think will be of interest to those who read this
: group. I took notes, and will try very hard to be accurate, but there
: may be mistakes, so I apologize for any of these.

: I asked about the famous 10-game match and go the following information
: back.

: Ten games total were played between "Deep Blue Jr" and two PC programs,
: Rebel 7 and Genius 5.

: The configuration of DB Jr that was used in this instance was a
: single-processor version searching 2 million nodes per second, with some
: search extensions turned off.

: The PC programs were running on P5/90's. The time controls were
: anywhere between 10 0 and 30 0.

: The score of this "just so happened to be" 10-0. I didn't get the
: impression that Campbell was suggesting that this would happen every
: time.

: There were 30 other games played between DB Jr with between 1 and 16
: processors. The score of these games was 28-2.

: I may have made a mistake. I might be reporting the NPS results for the
: 16-processor variant, but I don't think so.

I believe your last assumption is true, because a single deep blue processor
doesn't search at 2M nodes per second... they are running around 1M the last
I heard, which is why 256 nodes was hitting around 200M nps against Kasparov.

However, as I originally said, I was told directly by Hsu, who also told
this same thing to at least one reporter, that the DB hardware processor was
slowed down by a factor of 10X, to get it "close" to the micro speed. If
you ignore their hardware eval I've mentioned many times.

I'll try to email Murray and ask him about this specifically to see if (a) he
recalls something different from what Hsu revealed, or (b) he wasn't clear when
he answered. I suspect more of the latter, because of the 2M number that
doesn't match anything but some *very* old hardware they once had...

Also, in the eval explanation, it has been my interpretation that their
8000+ terms are *not* implemented as any sort of piece square tables. Unless
you take the idea that *one* of the terms could be weighted over 64 squares and
put in a piece square table, which I believe they do for some things.

Glad they are releasing details. Now maybe we can see discussions here that
deal more in facts and not in religious opinions and prejudices. That'd be a
nice change for the better.

However, I'll try to get a definitive and precise answer about the nps of
the hardware used for the 10-0 match to make sure there's no misunderstanding.

[Robert Hyatt]

Rolf Tueschen (TUESCHEN.MEDIZ...@t-online.de) wrote:
: hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:

: >brucemo (bru...@seanet.com) wrote:
: >: I just attended a lecture by Murray Campbell, one of the folks
: >: responsible for Deep Blue.

: >Glad they are releasing details. Now maybe we can see discussions here that

: >deal more in facts and not in religious opinions and prejudices. That'd be a
: >nice change for the better.

: >However, I'll try to get a definitive and precise answer about the nps of
: >the hardware used for the 10-0 match to make sure there's no misunderstanding.

: Oh Lord, have mercy with Prof Hyatt, Alabama.

: The problem you seem to hide, Bob, is simply, WHY the hell you didn't do
: this long before, because you saw the trouble we had after your first
: infos.

I do not *hide* anything. I gave a precise explanation of details that
Hsu relayed to me. I asked him if it was ok to post them... he said
sure, he had given the same info to a couple of reporters, although he
had not seen it in print.

: As scientist -- not cheer-leader of course of DB/IBM - you had the dutry
: to fight for further releases.

They have *always* been doing this Rolf, it is simply your ignorance that
hasn't seen it. They have been giving such presentations for many years,
I have been to some, and have heard about others. This is *not* something
"new" they have started.

: You are also on the cold war ticket against some Botvinnikal attempts,
: so why do you behave now the same/ or help upright the same / or dont
: share the critics of the like -- when we still have a very warm summer,
: at least in Europe??

: Also after the friendly publication of Bruce which is the FIRST about
: the whole event at all, I see it as follows, and I write this also
: knowing that Murray was very close to you or vice versa....

: For me it seems as if they made such a spooky ballyhoo because thay
: wanted to make money on the american market. This tiny slip of
: tongue/full mouth of Bruce openend my eyes.
: MC is in another field with IBM.... .... makes the world go round, the
: world go round...... (wasn't it the singer of Cabaret?) :)

: Ok, I know all this will be censored in future. But, well, we've still
: the freedom of usenet...

: Also, let me say this. It's a scandal if Murray's memory was really so
: full of holes that he didn't remember exactly what Bruce's questions
: were all about...

: You know, it reminds me of certain secrecies of the cold war too.

: But the late Botvinnik, one of the best chess guys of all times (taken
: each one in relation to his times; he was the best from 1938 to 1960!!)
: is spoiled by unproven insults, which were pushed weeks before he died
: with over 80 y.

: But Murray. well, you never saw him providing us with fishy data...

not at all...

: How do you call the lectural/Bruce quest-answers?

I have no idea what you are asking. I assume Murray did the same thing at
his talk that I did last year at the University of Maryland when I was invited
up there to talk about Computer Chess, or when I visited the Univesity of
Waterloo this year to talk about computer chess, or when any other active
computer chess type gives a presentation.

: I think we have another *political* dimension....

: And I also think that the time schedule can also make data fishy.
: If I need months to publish game notations...
: I mean what will people like MC tell us if *they* are way beyond 60,
: 70... y. old? I know it's rhetorical because first we're further in new
: dimensions and the I hope we will *solve* the real clue about DB without
: any data from IBM at all.

: I will keep safe in my memory that *you* for one did never answer when I

: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
: asked /pointed at the question of this strange delay in publishing
: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
: promissed data. If we the see that something was really fishy somehow,
: ~~~~~~~~~~~~
: you can stop posting about such questions because everybody will no
: longer take you as scientifically independent in these fields.

They don't owe anyone anything, contrary to what you seem to believe.
They owe it to themselves to tell us how they did what they did. I
always believed that they would do so. Nothing has changed that. Whether
they release Kasparov game logs to the public or not is their decision. If
I were them, and I was going to play another match with him, I wouldn't let
'em go either. Otherwise I probably would...

: My God, you claimed close contacts to these guys, you publish half
: cooked data snips of Hsu, and suddenly when Bruce comes up with another
: snips you are prepared to ask them by email (hear hear, you can do
: that?) about details. I'm repeating myself, but it's so sad.
: You know I love you but now I feel like being cheated. Bob, science

It is very sad that you continue to post here, and try to run everyone
that has anything technical to say away. In my case, you aren't going to
succeed, because you aren't that important. I simply said I was going to
ask Hsu to be *certain* that I didn't mis-interpret anything he told me.

IE I believe that he used a one processor machine. Bruce said he thought
they used something that searched 2M nodes per second, but he wasn't sure
this was the machine used in the 10-0 match. 2M doesn't ring a bell with
me at all, so it deserves clarification. But I didn't see anything in his
report of Murray's talk that contradicted anything I posted after getting the
info from Hsu...

: means pushing and dealing and pushing, but not staying mute if a friend
: is involved. Either you phone him, like you phone Roman, either you
: email him, "MC, that's not ok, rgcc is burning, you've to provide us
: with data...", or you ask them here in public.

: But what you did. You were like the chief of their press center who ran
: around with this baseball stick to mute all speculations about spooky
: data you yourself published here in rgcc................... :((((

: Now it's *your* job to give the micro heroes back what they deserved....

If I was wrong, I'll post the corrected version. If I was not wrong, I
will post that again as well. But the micros got what they deserved, which
was a sound thumping, just like they (we) have been getting since 1987
when that project was first unveiled...

: As Pope I condemn you for a two weeks non-god status here in rgcc.

In this regard, I am 100% anti-religion...

[Rolf Tueschen]

hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:

>brucemo (bru...@seanet.com) wrote:
>: I just attended a lecture by Murray Campbell, one of the folks
>: responsible for Deep Blue.

>Glad they are releasing details. Now maybe we can see discussions here that
>deal more in facts and not in religious opinions and prejudices. That'd be a
>nice change for the better.

>However, I'll try to get a definitive and precise answer about the nps of
>the hardware used for the 10-0 match to make sure there's no misunderstanding.

Oh Lord, have mercy with Prof Hyatt, Alabama.

The problem you seem to hide, Bob, is simply, WHY the hell you didn't do
this long before, because you saw the trouble we had after your first
infos.

As scientist -- not cheer-leader of course of DB/IBM - you had the dutry

to fight for further releases.

You are also on the cold war ticket against some Botvinnikal attempts,

so why do you behave now the same/ or help upright the same / or dont
share the critics of the like -- when we still have a very warm summer,
at least in Europe??

Also after the friendly publication of Bruce which is the FIRST about
the whole event at all, I see it as follows, and I write this also
knowing that Murray was very close to you or vice versa....

For me it seems as if they made such a spooky ballyhoo because thay
wanted to make money on the american market. This tiny slip of
tongue/full mouth of Bruce openend my eyes.
MC is in another field with IBM.... .... makes the world go round, the
world go round...... (wasn't it the singer of Cabaret?) :)

Ok, I know all this will be censored in future. But, well, we've still
the freedom of usenet...

Also, let me say this. It's a scandal if Murray's memory was really so
full of holes that he didn't remember exactly what Bruce's questions
were all about...

You know, it reminds me of certain secrecies of the cold war too.

But the late Botvinnik, one of the best chess guys of all times (taken
each one in relation to his times; he was the best from 1938 to 1960!!)
is spoiled by unproven insults, which were pushed weeks before he died
with over 80 y.

But Murray. well, you never saw him providing us with fishy data...

How do you call the lectural/Bruce quest-answers?

I think we have another *political* dimension....

And I also think that the time schedule can also make data fishy.
If I need months to publish game notations...
I mean what will people like MC tell us if *they* are way beyond 60,
70... y. old? I know it's rhetorical because first we're further in new
dimensions and the I hope we will *solve* the real clue about DB without
any data from IBM at all.

I will keep safe in my memory that *you* for one did never answer when I

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
asked /pointed at the question of this strange delay in publishing
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
promissed data. If we the see that something was really fishy somehow,
~~~~~~~~~~~~
you can stop posting about such questions because everybody will no
longer take you as scientifically independent in these fields.

My God, you claimed close contacts to these guys, you publish half

cooked data snips of Hsu, and suddenly when Bruce comes up with another
snips you are prepared to ask them by email (hear hear, you can do
that?) about details. I'm repeating myself, but it's so sad.
You know I love you but now I feel like being cheated. Bob, science

means pushing and dealing and pushing, but not staying mute if a friend
is involved. Either you phone him, like you phone Roman, either you
email him, "MC, that's not ok, rgcc is burning, you've to provide us
with data...", or you ask them here in public.

But what you did. You were like the chief of their press center who ran
around with this baseball stick to mute all speculations about spooky
data you yourself published here in rgcc................... :((((

Now it's *your* job to give the micro heroes back what they deserved....

As Pope I condemn you for a two weeks non-god status here in rgcc.

The Chief Pope of all US states' press orgs and peace and justice

[SMO CFI]

all,

heard on the radio that DB is retiring as the heavyweight champion, and
will play no more...

Will

[David Hanley]

brucemo wrote:

> The opening in game 6 was part of the hand-tuned book. They added
> that
> sac thinking they would never get into this part of the book, and DB
> liked the resulting position anyway, so they kept it.

That's quite a statement. This implies that DB may even play thesac
without having it in the book. That is incredible.

dave

[Robert Hyatt]

David Hanley (da...@nospan.netright.com) wrote:
: brucemo wrote:

: dave

It really isn't... If you remember the Crafty vs ChessMaster 5000 game
in the Korrespondence Kup? Crafty sacrificed a knight in a wild position,
but only did so because it was evaluating the position at -.5, and the
knight sac led to a draw by repetition. However, after several moves
went by, Crafty realized that the sac didn't just lead to a draw, it led
to a forced win... just a very deep one. Deep Blue Junior saw this when
the knight sac was played, with an eval over +2...

I've been on the wrong end of their search depths too many times. :)

It is formidable in such positions. It likely has no equal in positions
like that, either human or computer...

[Robert Hyatt]

Robert Hyatt (hy...@crafty.cis.uab.edu) wrote:

: : dave

Also, I decided to backtrack to the original note from Hsu about the
10-0 match result, after reading Bruce's post and being left somewhat
confused. Here is a copy of the relevant part:

"I told the following to the Newsweek reporter already."

"We played 10 games against Genius 5 and Rebel 8 (about 5 games each)"
"with a single chess chip, running at about 1/10th speed, or roughly"
"the same speed as the PC programs. I was expecting to be slightlyi"
"ahead. The result was "a 10-0 slaughter. The process of patching"
"all the holes that Gary could drive a truck thru paid off much better"
"than I expected. "

There you have the important part of his comments. Other parts I have left
out as they were observations I said I wouldn't mention, about what the PC
programs seemed to have serious trouble with in those games. However, I don't
see *any* bit of ambiguity in the above. one chess processor (single chip)
running at about 1/10th the normal speed. That is clear, and really not open
to any other interpretation. So either Murray was not clear in his talk, or
(as Bruce mentioned) perhaps Bruce mixed the speeds up since he mentioned
several different DB results against the micro programs. But I think you'd
agree, the above is quite clear. No, it doesn't say anything about the time
controls. Bruce said Murray said they were 30 0 games and faster. That is
possible, and it is also not all blitz games either. Other than the above,
all else would be speculation by me...

One comment: Hsu mentioned that at one of the prematch conferences, Kasparov
stated that DB should beat the PC programs by 8-2. At that point, the DB
guys "knew" Kasparov was in for a major shock, because 8-2 wasn't even a good
guess for a single chip machine running at 1/10th the normal speed, much less
the full-blown fire-breathing whole thing...

He also mentioned that at an Icelandic web site he found a post where an
IM played genius 5 from the game 6 position after Nxe6, and the IM won quite
easily. Note that the IM was playing *black*. Still lends credence to my
original idea that he had tried this as black against Fritz or Hiarcs or
Whatever and won easily, and decided to "go for it" in this game, knowing he
could always claim "finger slip" or "memory slip" if he lost... Who knows of
course, but it's a thought...

[brucemo]

David Hanley wrote:
>
> brucemo wrote:
>
> > The opening in game 6 was part of the hand-tuned book. They added
> > that
> > sac thinking they would never get into this part of the book, and DB
> > liked the resulting position anyway, so they kept it.
>
> That's quite a statement. This implies that DB may even play thesac
> without having it in the book. That is incredible.

Not necessarily. I don't know how long they searched on the sac itself,
or at what point they had DB examine a position to verify it. It could
have been five moves later. That's ten plies, and ten plies is a lot.

It is interesting that this is something that they voluntarily allowed. I
pointedly asked Campbell if this was an opening book disaster that
coincidentally worked, and he said "no".

bruce

[brucemo]

Eugene Nalimov wrote:
>
> Today Murray Campbell gave another lecture, this time at Microsoft. Here
> are some interesting details:

I thought about trying to sneak in, but I figured that I might end up dealing
with the Redmond cops or an ex-boss or something :-)

> There are 30 general purpose CPUs – one master, and 29 slave. Each has 8
> specialized chess processors.
> Each chess processor calculates 5 full plies in hardware. 1.3mln
> transistors, 2mln nodes/sec

Ok, then I heard right about the 2M nps.

In that impromptu debugging match that Campbell did I thought we had the idea
that the thing had been slowed down even further for for purposes of
handicapping, but what Campbell says tends to contradict this, so obviously
there is a major misunderstanding.

> Each general purpose CPU calculates 4-5 full plies in software (w/o
> extensions) (in my opinion – and as were written by Bob and others - that
> means that they have a lot of extensions, otherwise they can easily search
> 2-3 plies deeper)

Sounds like mainly singular extension, although Campbell told me that they
have "influence" extensions, which are extensions based upon the influence
that a particular move has, the example being the idea of extending when you
push a passed pawn.

> Parallel search effectiveness is 25-30%
> Important lines typically 20-30 plies deep; Murray sometimes saw 70
> 8000 evaluation factors
> Some evaluating factors are known by any grandmaster, but not written in
> any textbook
> non-linear evaluating function
> they think that singular extension heuristic is very good – that
> contradicts with what they published 5-6 years ago

They originally (1980's) published some huge rating point increase due to
using this, then they scaled it back, but I thought they have always thought
this was "good"? It's always been their trademark.

> grandmaster-created opening database of 200.000 positions
> secondary database of 600.000 GM games; interesting selection criteria –
> not only based on score statistics (as in Crafty), but also on GM and year
> during one of the match games, there were endgame databases hits (but
> anyway, it was too late – position was drawn)

Did anyone ask any good questions?

Was he still wearing the blue suit? Did everyone hiss when they saw "Lotus"
on his screen? I thought about warning him. :-)

bruce

[Nouveau Jeff]

Rolf Tueschen <TUESCHEN.MEDIZ...@t-online.de> a écrit dans
l'article <6089co$qhb$1...@news00.btx.dtag.de>...

> brucemo <bru...@seanet.com> wrote:
>
> >I just attended a lecture by Murray Campbell, one of the folks
> >responsible for Deep Blue.
>

> Ed Schroder, please do comment on this post.
>
> I hereby promise NOT to give any commentary on whtever you may write
> about this.

Please Ed, I think it's worth a try, Rolf just gives his word.
Please, pleaaaaaaaase :))))

>
> But we all are interested in what you might think now after we go these
> news...

For once, I agree with Rolf, I'd like to hear Bob too : is it possible to
read
about computer chess again ?

Maybe Chris has some ideas ?

:)

Pleaaaaaaaaaaase

Jeff

--
Celui qui sait qu'il ne sait pas, éduque le.
Celui qui sait qu'il sait, écoute le.
Celui qui ne sait pas qu'il sait, éveille le.
Celui qui ne sait pas qu'il ne sait pas, fuis le.
Proverbe Chinois
>
> Thanks.
>
>
> Rolf Tueschen
>
>

[Eugene Nalimov]

Today Murray Campbell gave another lecture, this time at Microsoft. Here
are some interesting details:

There are 30 general purpose CPUs – one master, and 29 slave. Each has 8

specialized chess processors.
Each chess processor calculates 5 full plies in hardware. 1.3mln
transistors, 2mln nodes/sec

Each general purpose CPU calculates 4-5 full plies in software (w/o
extensions) (in my opinion – and as were written by Bob and others - that
means that they have a lot of extensions, otherwise they can easily search
2-3 plies deeper)

Parallel search effectiveness is 25-30%
Important lines typically 20-30 plies deep; Murray sometimes saw 70
8000 evaluation factors
Some evaluating factors are known by any grandmaster, but not written in
any textbook
non-linear evaluating function
they think that singular extension heuristic is very good – that
contradicts with what they published 5-6 years ago

grandmaster-created opening database of 200.000 positions
secondary database of 600.000 GM games; interesting selection criteria –
not only based on score statistics (as in Crafty), but also on GM and year
during one of the match games, there were endgame databases hits (but
anyway, it was too late – position was drawn)

Eugene

brucemo <bru...@seanet.com> wrote in article <34275C...@seanet.com>...

> I just attended a lecture by Murray Campbell, one of the folks
> responsible for Deep Blue.
>

> I will not attempt to provide a summary of the entire lecture, or of my
> questions to Campbell after the lecture, but I will provide some
> information that I think will be of interest to those who read this
> group. I took notes, and will try very hard to be accurate, but there
> may be mistakes, so I apologize for any of these.
>
> I asked about the famous 10-game match and go the following information
> back.
>
> Ten games total were played between "Deep Blue Jr" and two PC programs,
> Rebel 7 and Genius 5.
>
> The configuration of DB Jr that was used in this instance was a
> single-processor version searching 2 million nodes per second, with some
> search extensions turned off.
>
> The PC programs were running on P5/90's. The time controls were
> anywhere between 10 0 and 30 0.
>
> The score of this "just so happened to be" 10-0. I didn't get the
> impression that Campbell was suggesting that this would happen every
> time.
>
> There were 30 other games played between DB Jr with between 1 and 16
> processors. The score of these games was 28-2.
>
> I may have made a mistake. I might be reporting the NPS results for the
> 16-processor variant, but I don't think so.
>

> The opening in game 6 was part of the hand-tuned book. They added that
> sac thinking they would never get into this part of the book, and DB
> liked the resulting position anyway, so they kept it.
>

[Don Getkey]

In article <6090cf$f...@juniper.cis.uab.edu>, hy...@crafty.cis.uab.edu
(Robert Hyatt) writes:

>If I was wrong, I'll post the corrected version. If I was not wrong, I
>will post that again as well. But the micros got what they deserved, which
>was a sound thumping, just like they (we) have been getting since 1987
>when that project was first unveiled...

Ditto on the Rolf replies BTW.

The micros did get "thumped", but I wonder why they used slow P90s for
fast games?

If they wanted a closer approximation in speed to test out their eval (or
whatever) why did they not use PP200s? They should have been still faster,
by a bunch, but Deep Blue Jr. would have had a much better set of sparing
partners, capable of throwing a few punches, rather than just taking them.

Score might have been something like 7-3 instead of 10 zip. The Deep Blue
team might have learned a bit more about their eval too.

yours in chess,
Don

Ramsey MN USA

[brucemo]

Robert Hyatt wrote:

> Also, I decided to backtrack to the original note from Hsu about the
> 10-0 match result, after reading Bruce's post and being left somewhat
> confused. Here is a copy of the relevant part:
>
> "I told the following to the Newsweek reporter already."
>
> "We played 10 games against Genius 5 and Rebel 8 (about 5 games each)"
> "with a single chess chip, running at about 1/10th speed, or roughly"
> "the same speed as the PC programs. I was expecting to be slightlyi"
> "ahead. The result was "a 10-0 slaughter. The process of patching"
> "all the holes that Gary could drive a truck thru paid off much better"
> "than I expected. "

You have email and I have my recollection about what Campbell said yesterday. I
am pretty sure I heard right, and another poster has independently posted 2M
nps, so I'm convinced that that is what Campbell meant.

He did not mention dumbing it down further. Hsu does. Campbell did say that
they'd left out some extensions. We have a contradiction, possibly caused by a
misunderstanding between these two guys.

In any case, 2M nps, as reported by Campbell twice in two days, divided by ten
is 200K, which is way faster than Rebel, and way way faster than Genius, on a
P5/90.

But this is still weird. I have a hard time dealing with the notion of an equal
NPS, or even a 2X NPS match going 10-0. DB doesn't prune much, it doesn't even
use null-move, and in fact it extends. So it is going to get absolutely hashed
on general search depth.

You brought this up in order to indicate that their eval function must be great,
but I don't think that any eval function could overcome this. They almost have
to be getting killed on search speed, and how do you kill someone on search
speed if you don't prune?

25-1 speed difference, with a good eval function, with no pruning, but with some
extensions you've been messing with for years, 10-0, I can believe that.

I am willing to stipulate that their eval function is perfectly fine. So I
don't think I'm arguing against your original point, I'm just going on about the
details of this match.

I do not want to start a firestorm. I don't want to accuse anyone of clouding
this issue for personal benefit, certainly. I am not a DB hater, I respect what
they have done. I am not an "Aw, those micros could whack it" kind of guy. I
was not asking questions in order to find holes in arguments or in any way hash
on DB, certainly. I simply tried to get "on the record" comments about some
stuff that has caused trouble in here, since you've been constrained by the
source of your account. When I asked my question about this, I mentioned you.
I told Campbell that I wanted to get this resolved because you had been "hung
out to dry". And he said some stuff in front of a hundred people and a video
camera.

If there is a contradiction between Campbell and Hsu, I am sorry that I did not
get it resolved yesterday. I also should have spoken with you before posting,
probably.

> There you have the important part of his comments. Other parts I have left
> out as they were observations I said I wouldn't mention, about what the PC
> programs seemed to have serious trouble with in those games. However, I don't
> see *any* bit of ambiguity in the above. one chess processor (single chip)
> running at about 1/10th the normal speed. That is clear, and really not open
> to any other interpretation. So either Murray was not clear in his talk, or
> (as Bruce mentioned) perhaps Bruce mixed the speeds up since he mentioned
> several different DB results against the micro programs. But I think you'd
> agree, the above is quite clear. No, it doesn't say anything about the time
> controls. Bruce said Murray said they were 30 0 games and faster. That is
> possible, and it is also not all blitz games either. Other than the above,
> all else would be speculation by me...

The time controls were unclear in your version. They were either standard or
unspecified, but I got the idea that they were long. I'd never heard about 10 0
or 30 0, from you, I am pretty sure.

Maybe those 10 games were all 10 0. I don't know, but it sounds like best case
they were 30 0. At the time this big brouhaha was happening in here, I am sure
that Ed was either shrieking about, or wanted to shriek about, the time controls
used. If you'd have said 30 0 on a P5/90, he would have gone completely nuts,
and if you'd said 10 0, they'd still be finding pieces of Ed in Finland.

Rebel was on a P5/90 doing 30 0 games. I'm not sure exactly how good it is
going to do at that time control, but that certainly isn't 40/2 on a better
machine, which Ed seems to think shows Rebel in best light. And 10 0 on a P5/90
is way less than 5 0 on a P6/200, and neither of us will bet on Rebel under
those conditions, against even another micro with a hardware advantage, right?

> One comment: Hsu mentioned that at one of the prematch conferences, Kasparov
> stated that DB should beat the PC programs by 8-2. At that point, the DB
> guys "knew" Kasparov was in for a major shock, because 8-2 wasn't even a good
> guess for a single chip machine running at 1/10th the normal speed, much less
> the full-blown fire-breathing whole thing...

I don't think that 10-0 would be much different from 8-2 in the mind of a strong
human player. Who knows. In any case, it's not like he didn't respect it, he
treated it like a toxic nuclear biohazard bucket from Mars. He approached it in
this amazing crouch posture, expecting it to spew live cats and used condoms at
him at any moment. In game 6 it did. In game 2 he just thought it did.

Back to the micro match.

I think DB Jr was lucky to win 10-0, since it did drop a couple of points over
40 games.

And Campbell isn't gloating about this either. He said, quoting from memory,
"the score of those ten games happened to be 10-0". You don't say "happened to
be", with a straight face, unless you think that was just one possible outcome.
And other possible outcomes might include, for instance, 9-1.

> He also mentioned that at an Icelandic web site he found a post where an
> IM played genius 5 from the game 6 position after Nxe6, and the IM won quite
> easily. Note that the IM was playing *black*. Still lends credence to my
> original idea that he had tried this as black against Fritz or Hiarcs or
> Whatever and won easily, and decided to "go for it" in this game, knowing he
> could always claim "finger slip" or "memory slip" if he lost... Who knows of
> course, but it's a thought...

Yes, I couldn't ask any questions about this, since Campbell isn't Kasparov. He
may have tried this, but if he did, it says a lot about his mental state, that
he'd played five new games with something that he claimed to have wonderful
knowledge of after the first match, and still confused it with Fritz.

But I could verify that DB had the sac in its book *on purpose*, and I did this.

bruce

[Tim Mirabile]

brucemo <bru...@seanet.com> wrote:

>It is interesting that this is something that they voluntarily allowed. I
>pointedly asked Campbell if this was an opening book disaster that
>coincidentally worked, and he said "no".

I can't imagine they ever expected, hoped, or dreamed for this book line to ever
come into play in this match. But when preparing a book you put in lots of
stuff you never expect to happen, and if the sac is sound, and you have
confidence in your machine, why not...

--
Long Island chess -> http://www.webcom.com/timm/ TimM on ICC and A-FICS
Webmaster, tech support - Your Move Chess & Games: http://www.icdchess.com/

[Robert Hyatt]

Don Getkey (dong...@aol.com) wrote:
: In article <6090cf$f...@juniper.cis.uab.edu>, hy...@crafty.cis.uab.edu
: (Robert Hyatt) writes:

: >If I was wrong, I'll post the corrected version. If I was not wrong, I

: >will post that again as well. But the micros got what they deserved, which
: >was a sound thumping, just like they (we) have been getting since 1987
: >when that project was first unveiled...

: Ditto on the Rolf replies BTW.

: The micros did get "thumped", but I wonder why they used slow P90s for
: fast games?

: If they wanted a closer approximation in speed to test out their eval (or
: whatever) why did they not use PP200s? They should have been still faster,
: by a bunch, but Deep Blue Jr. would have had a much better set of sparing
: partners, capable of throwing a few punches, rather than just taking them.

What is this P90 stuff? I'm not certain, but I thought that when I asked
what machine they used, the response was a P6/200... I never gave it much
thought since this was IBM that was doing the testing. :)

: Score might have been something like 7-3 instead of 10 zip. The Deep Blue

[Robert Hyatt]

brucemo (bru...@seanet.com) wrote:
: Robert Hyatt wrote:

: > Also, I decided to backtrack to the original note from Hsu about the
: > 10-0 match result, after reading Bruce's post and being left somewhat
: > confused. Here is a copy of the relevant part:
: >
: > "I told the following to the Newsweek reporter already."
: >
: > "We played 10 games against Genius 5 and Rebel 8 (about 5 games each)"
: > "with a single chess chip, running at about 1/10th speed, or roughly"
: > "the same speed as the PC programs. I was expecting to be slightlyi"
: > "ahead. The result was "a 10-0 slaughter. The process of patching"
: > "all the holes that Gary could drive a truck thru paid off much better"
: > "than I expected. "

: You have email and I have my recollection about what Campbell said yesterday. I
: am pretty sure I heard right, and another poster has independently posted 2M
: nps, so I'm convinced that that is what Campbell meant.

: He did not mention dumbing it down further. Hsu does. Campbell did say that
: they'd left out some extensions. We have a contradiction, possibly caused by a
: misunderstanding between these two guys.

: In any case, 2M nps, as reported by Campbell twice in two days, divided by ten
: is 200K, which is way faster than Rebel, and way way faster than Genius, on a
: P5/90.

: But this is still weird. I have a hard time dealing with the notion of an equal
: NPS, or even a 2X NPS match going 10-0. DB doesn't prune much, it doesn't even
: use null-move, and in fact it extends. So it is going to get absolutely hashed
: on general search depth.

Not particularly. Remember that Rebel and Genius also don't seem to use
null-move. So it becomes a battle on two fronts: (1) eval vs eval, which
is a front where everyone (almost everyone) has assumed that DB was horribly
inferior to the commercial micros, but this has been pretty well shown to be
totally incorrect now; (2) search vs search. I agree they lose depth with
the extensions. But *if* the extensions are correct, this is an advantage,
not a problem, because the rest of us are searching a lot of garbage to a
deep depth, while it is possible that their extensions are done well enough
that the paths they search to shallow depths don't mean as much as the ones
they extend deeply...

: You brought this up in order to indicate that their eval function must be great,

: but I don't think that any eval function could overcome this. They almost have
: to be getting killed on search speed, and how do you kill someone on search
: speed if you don't prune?

See above... but there's nothing to say they are getting killed on depth in
the "right" positions..

: 25-1 speed difference, with a good eval function, with no pruning, but with some

: extensions you've been messing with for years, 10-0, I can believe that.

: I am willing to stipulate that their eval function is perfectly fine. So I
: don't think I'm arguing against your original point, I'm just going on about the
: details of this match.

: I do not want to start a firestorm. I don't want to accuse anyone of clouding
: this issue for personal benefit, certainly. I am not a DB hater, I respect what
: they have done. I am not an "Aw, those micros could whack it" kind of guy. I
: was not asking questions in order to find holes in arguments or in any way hash
: on DB, certainly. I simply tried to get "on the record" comments about some
: stuff that has caused trouble in here, since you've been constrained by the
: source of your account. When I asked my question about this, I mentioned you.
: I told Campbell that I wanted to get this resolved because you had been "hung
: out to dry". And he said some stuff in front of a hundred people and a video
: camera.

: If there is a contradiction between Campbell and Hsu, I am sorry that I did not
: get it resolved yesterday. I also should have spoken with you before posting,
: probably.

Not necessary at all, of course. I'm not a "moderator" here. :) But you saw
from what I posted that there really is no room for misinterpretation. It was
quite clear from Hsu that they used 1 chip, at 1/10th its normal speed, which
dropped it to 100K nodes per sec. I still believe 1M nodes per second for the
hardware, because that matches everything we've seen. IE DB claimed to search
200M nodes per second against Kasparov. 2M nodes per sec/chip x 240 chips
doesn't add up. It might be that the 2M number came from the older hardware
(ie pre-DB-II, where Hsu redesigned the hardware one more time for GK). But
that is speculation too, of course (not about the hardware redesign, that's
fact, but where the 2M number came from).

: > There you have the important part of his comments. Other parts I have left

: > out as they were observations I said I wouldn't mention, about what the PC
: > programs seemed to have serious trouble with in those games. However, I don't
: > see *any* bit of ambiguity in the above. one chess processor (single chip)
: > running at about 1/10th the normal speed. That is clear, and really not open
: > to any other interpretation. So either Murray was not clear in his talk, or
: > (as Bruce mentioned) perhaps Bruce mixed the speeds up since he mentioned
: > several different DB results against the micro programs. But I think you'd
: > agree, the above is quite clear. No, it doesn't say anything about the time
: > controls. Bruce said Murray said they were 30 0 games and faster. That is
: > possible, and it is also not all blitz games either. Other than the above,
: > all else would be speculation by me...

: The time controls were unclear in your version. They were either standard or
: unspecified, but I got the idea that they were long. I'd never heard about 10 0
: or 30 0, from you, I am pretty sure.

I hadn't heard that either I don't believe. I think I asked "were they blitz
games?" and got a "no"...

: Maybe those 10 games were all 10 0. I don't know, but it sounds like best case

: they were 30 0. At the time this big brouhaha was happening in here, I am sure
: that Ed was either shrieking about, or wanted to shriek about, the time controls
: used. If you'd have said 30 0 on a P5/90, he would have gone completely nuts,
: and if you'd said 10 0, they'd still be finding pieces of Ed in Finland.

: Rebel was on a P5/90 doing 30 0 games. I'm not sure exactly how good it is
: going to do at that time control, but that certainly isn't 40/2 on a better
: machine, which Ed seems to think shows Rebel in best light. And 10 0 on a P5/90
: is way less than 5 0 on a P6/200, and neither of us will bet on Rebel under
: those conditions, against even another micro with a hardware advantage, right?

I didn't hear any mention of a P90. I really thought the machine was a P6/200,
so there's a bit of inconsistency there. One reason for thinking that is to re-
read Hsu's comments, "approximately equal to the search speed of the PC
programs". Rebel displays its NPS value clearly on the screen. And on a
P5/90, Rebel wouldn't be anywhere near 100K. It wouldn't be anywhere near 100K
on anything but a P6/200, so I'm not quite sure what this means. Again, it is
simply a contradiction that needs clearing up. But based on the note from Hsu,
it certainly sounds like a P6... I probably assumed this from the numbers, but
I could not imagine anyone considering that 100K on a DB chip would be equal
to Rebel/Genius on a P5/90, which is probably going to search around 25K nodes
per second or so with Rebel...

: > One comment: Hsu mentioned that at one of the prematch conferences, Kasparov

: > stated that DB should beat the PC programs by 8-2. At that point, the DB
: > guys "knew" Kasparov was in for a major shock, because 8-2 wasn't even a good
: > guess for a single chip machine running at 1/10th the normal speed, much less
: > the full-blown fire-breathing whole thing...

: I don't think that 10-0 would be much different from 8-2 in the mind of a strong
: human player. Who knows. In any case, it's not like he didn't respect it, he
: treated it like a toxic nuclear biohazard bucket from Mars. He approached it in
: this amazing crouch posture, expecting it to spew live cats and used condoms at
: him at any moment. In game 6 it did. In game 2 he just thought it did.

: Back to the micro match.

No, but it tends to lead me to think that he was training against a micro, and
thought that he was only facing a program that was about 200 (8-2 result)
better than the micros he was using for training. He was *way* off in that
estimate...

: I think DB Jr was lucky to win 10-0, since it did drop a couple of points over
: 40 games.

: And Campbell isn't gloating about this either. He said, quoting from memory,
: "the score of those ten games happened to be 10-0". You don't say "happened to
: be", with a straight face, unless you think that was just one possible outcome.
: And other possible outcomes might include, for instance, 9-1.

That doesn't seem improbable at all. IE, given a machine like DB, full-blown,
against a 1980-era program like Sargon II on a radio shack machine, would you
bet your house that sargon wouldn't win a single game? I wouldn't, because
I have seen plenty of games where search had nothing to do with the outcome.
Neither did knowledge. Some games are simply won or lost, and even a weak
player can win them and a strong player can lose them...

So a non-10-0 result is obviously possible. But it is also *very* revealing,
when you look at it, because certainly the odds of a weaker program thrashing
a stronger program 10-0 is so close to zero that calling it zero is not too
far-fetched..

: > He also mentioned that at an Icelandic web site he found a post where an

: > IM played genius 5 from the game 6 position after Nxe6, and the IM won quite
: > easily. Note that the IM was playing *black*. Still lends credence to my
: > original idea that he had tried this as black against Fritz or Hiarcs or
: > Whatever and won easily, and decided to "go for it" in this game, knowing he
: > could always claim "finger slip" or "memory slip" if he lost... Who knows of
: > course, but it's a thought...

: Yes, I couldn't ask any questions about this, since Campbell isn't Kasparov. He
: may have tried this, but if he did, it says a lot about his mental state, that
: he'd played five new games with something that he claimed to have wonderful
: knowledge of after the first match, and still confused it with Fritz.

: But I could verify that DB had the sac in its book *on purpose*, and I did this.

: bruce

I have it in my book "not on purpose". One other discrepancy I noticed is that
GM Joel Benjamin said in one of the post-game interviews that DB's opening
database included *every* game they had found in electronic form. The approach
I use in Crafty (and used in Cray Blitz) was well-discussed, a big book to
cover everything, a small hand-tailored book to guide the programs down the
lines that best suite them. Murray's comments seems to say this wasn't done.
It would be nice to know what they used... :)

[Robert Hyatt]

Rolf Tueschen (TUESCHEN.MEDIZ...@t-online.de) wrote:

: Hyatt>I do not *hide* anything.

: Sorry, but you did. But it's surely not of the kind you now oppose with
: determination. It's a bit more complicated. Let's see also your next two
: sentences before I'll explain what I mean.

: >I gave a precise explanation of details that

: >Hsu relayed to me. I asked him if it was ok to post them... he said
: >sure, he had given the same info to a couple of reporters, although he
: >had not seen it in print.

: First let me politely make clear that I don't think you personally hid
: things you already knew. Still you hid things. Why?

You made the claim. You offer the evidence to support that claim. I
later posted the relevant part of the note from Hsu that *clearly* showed
what he said took place. It doesn't match exactly with what Murray said.
Why, I don't know. One of them is clearly at odds with the other over
what happened, or, possibly, Bruce misunderstood/misinterpreted/mis-remembered
what was said. however, the quote I posted leaves *zero* room for any sort
of interpretation that I can see..

: Bob, I always come back positively to your status, to your age and to
: your education and titles. You once explained to Enrique what it was for
: all the titles in rgcc. *Not* the final title but the education in
: basical science related methods for instance...

: Look, if you were a sophomore or what it's called in USA, I would have
: taken your same post about Hsu's snips in that situation in rgcc with
: joy and had given you positive feedback without longer digging deaper...

: What makes the whole story fishy is that ***you*** with your
: image/status want to pretent/play the role of an open mouthed praying
: server in iew of his bishop... Your commentary of the conditions of
: this info exchange in itself is a scandal, Bob. You're not getting tired
: repeating how you asked him if you could post... but you surely knew
: before that he had already talked to the press... But this is only the
: beginning. Then you had these info snips. ----------

No I did not. Hsu has corresponded with me on many occasions. I once
quoted something he said and he asked me to not quote him here, for what
is now obvious reasons, since everything gets so twisted by "some" that
it isn't worth the trouble. When he sent me this note, which was in the
middle of the Rebel NPS discussion with Ed, I simply asked him if it was
O.K. to post this. He responded with the clip I posted, which included
the remark "I have already told this to some reporters." So don't try to
stick ulterior motives into what I do. Until you produce evidence that you
have some sort of telepathic gift, please write about what I *say*. *or*
about what you *think*. But *not* about what I think...

so *no* I did not know beforehand that he had leaked this to the press.
However, at the time I posted the match (10-0) result, I did post the
fact that it had been given to the press already. Go dig up my original
post on the subject and you'll see that clearly indicated this. So I have
no idea of where you are coming from, nor of where you are going.

: Now, Bob, even Briuce, who's 20 years younger asked Marray better
: questions than you to Hsu. How come? You heard these snips. You *knew*
: immediately what these scores could be worth for, but you forgot to ask
: about some further details???????????????????????

You saw the quote I posted. *what* is missing? He explained *exactly*
how the DB hardware was configured, the only detail that I didn't ask about
was the time control. Which doesn't matter since both programs were playing
on equal time. 10-0 is a good result whether it be one second per move or
1 hour per move...

: Bob, is this a talk among scientists in USA of 1997? Well, then good
: night. Then we still have a chance in Europe to come close to the
: american standard in the future. If this is the way of religiously
: filled atmosphere of a talk between cc heroes both in their fields...

: Bob, if all this was true till here, and I wont doubt it, then you made
: an absolute master-clown out of yourself. Then you wrote in springtime
: that you planned however to email Hsu and ask him.... and so on.

: Bob!!! This is absolutely unneccessary to inform us about your plans of
: emailing someone. What I didn't see till today is simply, what were the
: results of this email exchange?

I responded to what Bruce wrote. The data he got from Murray and the data
I got from Hsu don't match as well as I'd like. The 10-0 result was a point
of agreement. But the speed of the hardware was not. If you saw the clip
from Hsu's email I posted, it was *very* clear as to the details of how DB
was running, and this was significantly different from what Bruce reported
from the questions he asked Murray. I added that I'd simply ask Hsu again
and make sure that the information he sent to me was correct, or else was
the info that Murray released more accurate. One must be incorrect. For
the moment, I'll stand by a concise statement from Hsu, as there is no other
information to cloud it or get mixed up...

: Let's talk a little bit of meta stuff.

: I saw you post stuff that must by force lead to a big trouble amoung Ed
: and the followers. With reason, as I could understand. All this you
: could have had forseen. As a scientist, no?

I have no idea what you mean. Ed has been getting clobbered by Chiptest/
Deep Thought/DB prototype for over 10 years. Just like I and every other
program author in the world have been getting clobbered. So I see no
impact on Ed or Rebel in any fashion. He made a rather odd claim that he
thought Rebel was only 50 Elo points below DB, which I *know* is incorrect.
And I suspect he knows that too, but that comment was simply the
marketeering side of him coming out...

: But thereafter you used every a little bit too far projected critic as
: sort of proof for an unallowed anti-DB "basher", I recall that notion,
: and whatelse. That was unallowed too. Because on the base of your
: data/evidence even stupid ignorants like me were able to detect that you
: had given a very blue balloon a lift off just to make a little
: propaganda for DB and *team*. To back them in their negociations about
: new contracts with IBM... in out-of chess regions as we now know.

Again, your telepathic prowess is coming up with a big fat zero. I don't
do *anything* to help DB obtain any sort of good PR, or support, or
anything. I report *facts* about them, to conteract all the rumors and
speculation that a few others like to produce.

: With that discussion you could break the big anti-DB feelings due to the
: behavior against Kasparov. This is called war-side-scene-creating. To
: shift the attention.

: >: As scientist -- not cheer-leader of course of DB/IBM - you had the dutry

: >not at all...

: After some other posts it should be clear by now.

: >I assume Murray did the same thing at

: >his talk that I did last year at the University of Maryland when I was invited
: >up there to talk about Computer Chess, or when I visited the Univesity of
: >Waterloo this year to talk about computer chess, or when any other active
: >computer chess type gives a presentation.

: >: I think we have another *political* dimension....

: >: And I also think that the time schedule can also make data fishy.
: >: If I need months to publish game notations...
: >: I mean what will people like MC tell us if *they* are way beyond 60,
: >: 70... y. old? I know it's rhetorical because first we're further in new
: >: dimensions and the I hope we will *solve* the real clue about DB without
: >: any data from IBM at all.

: >: I will keep safe in my memory that *you* for one did never answer when I

: >: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
: >: asked /pointed at the question of this strange delay in publishing
: >: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
: >: promissed data. If we the see that something was really fishy somehow,
: >: ~~~~~~~~~~~~
: >: you can stop posting about such questions because everybody will no
: >: longer take you as scientifically independent in these fields.

: >They don't owe anyone anything, contrary to what you seem to believe.

: Oh, yes, Bob, they do. You know why? Because of the big manking vs
: machine ballyhoo before the show. *Or* the impression/propaganda of a
: total/clear win in a *match* should be corrected as fast as possible.

Hmmm... then Ed owes us a detailed explanation of what goes on inside
Rebel, Lang owes us the same about genius, etc. Funny how it is ok for
one group to hide in secrecy, but not ok for Deep Blue. You'd think they
were paid for by taxpayer's money or something. They aren't. They are
a *private* company venture by IBM.

: >They owe it to themselves to tell us how they did what they did. I

: >always believed that they would do so.

: Did you *see* something yet?

I have seen *lots* of somethings. Murry giving talks is one example. Hsu
has done the same. The JICCA has had papers written by them although not in
the past two-three years.

: >Nothing has changed that. Whether

: >they release Kasparov game logs to the public or not is their decision. If
: >I were them, and I was going to play another match with him, I wouldn't let
: >'em go either. Otherwise I probably would...

: What's this? I thought they left chess for the big money transaction,
: known as beaming in StarTrek if I got this right. :)

???

: >: My God, you claimed close contacts to these guys, you publish half

: >: cooked data snips of Hsu, and suddenly when Bruce comes up with another
: >: snips you are prepared to ask them by email (hear hear, you can do
: >: that?) about details. I'm repeating myself, but it's so sad.
: >: You know I love you but now I feel like being cheated.

: Pity that you didn't comment on this. This was meant as the expression
: of a deep depression, Bob. Because you made as scientist a real torture
: for your fans. How could you crawl so deep on your knees. You, our heroe
: on rgcc?

here's the comment. You missed the post (apparently) from yesterday where
I included the exact response Hsu sent to me. It was un-edited, and un-
retouched. And it left *nothing* out, except for the time controls. I
now assume the games were game/30 type games and quicker.
As to feeling cheated, I have no idea why. I report what I know, and *only*
what I know. If more followed that, this would be a much more pleasant place
to be.

: >:Bob, science

: >It is very sad that you continue to post here, and try to run everyone
: >that has anything technical to say away.

: Bob, I never expected that retro from yours. This is the absolute gamma
: ray explosion. I was the only one who opposed Ed's cheaty private emails
: about you putting terror on people, and now, after Ed put anpther
: scandalous thing on a member of this group, and sure, his party soldiers
: left the scene with him, and he lost his face in Zwolle, you attack *me*
: for having thrown experts out of rgcc?????

I don't care about the war between you and Ed. You continue to bring that
up in nearly every post you write. It does *nothing* for the credibility of
the rest of what you write when you continue to harp on that. It is past
history, Ed has left, so it is time to move on.

: Bob, this is now history. It's in Dejas. And you will feel sorry in
: future about that quote of yours. I didn't expulse an expert from rgcc,
: I didn't intend to do so, and surely I *never* attacked someone who just
: posted technicasl stuff here in rgcc. These are all lies, Bob. You have
: no evidence at all. Or show it here right now.

: I claim here right now, that I did never attack a poster if he had not
: posted something strange here against me or against scientifical
: standards. You wont find a single counterexample, Bob.

: Give a correction on this point.

Simply read thru *this* post. you made several statements that are
incorrect...

: >In my case, you aren't going to

: >succeed, because you aren't that important.

: Wait a minute, you just left your usual logic. It's true I'm not that
: important and I never claimed expert status. I'm just the Pope. But how
: could I succeed when I never tried to achieve that goal? Bob?
: If you now also start sissying around where will it all end? If I were
: the reason for your departure I would fly immediately to Alabama,
: despite all the flies and snakes over there and I put you back into
: rgcc. Truely. Either I stayed mute for two years than loosing you and
: your posts here in the group, Bob. Uuuuch, those knees are so swollen
: hurt by my creeping on the floor, Bob......

I simply said that *you* are *not* going to run *me* off from here.
Nothing more. Nothing less. I'll leave when I want, or when the group
as a whole asks me to leave. But not when one person rants and raves.

: >I simply said I was going to

: >ask Hsu to be *certain* that I didn't mis-interpret anything he told me.

: >IE I believe that he used a one processor machine. Bruce said he thought
: >they used something that searched 2M nodes per second, but he wasn't sure
: >this was the machine used in the 10-0 match. 2M doesn't ring a bell with
: >me at all, so it deserves clarification. But I didn't see anything in his
: >report of Murray's talk that contradicted anything I posted after getting the
: >info from Hsu...

: >: means pushing and dealing and pushing, but not staying mute if a friend
: >: is involved. Either you phone him, like you phone Roman, either you
: >: email him, "MC, that's not ok, rgcc is burning, you've to provide us
: >: with data...", or you ask them here in public.

: >: But what you did. You were like the chief of their press center who ran
: >: around with this baseball stick to mute all speculations about spooky
: >: data you yourself published here in rgcc................... :((((

: >: Now it's *your* job to give the micro heroes back what they deserved....

: >If I was wrong, I'll post the corrected version. If I was not wrong, I
: >will post that again as well. But the micros got what they deserved, which
: >was a sound thumping, just like they (we) have been getting since 1987
: >when that project was first unveiled...

: No, you will correct this one too, you Popeye- schmonzes- Hulk. :)

: Now you put terror on me!

: >: As Pope I condemn you for a two weeks non-god status here in rgcc.

: >In this regard, I am 100% anti-religion...

: Doesn't help you if you're still in the boiling pot...

Takes a big person to put *me* in a boiling pot. He'd better come prepared
for a struggle... And he'd better be serious about trying, too...

[brucemo]

Robert Hyatt wrote:

> What is this P90 stuff? I'm not certain, but I thought that when I asked
> what machine they used, the response was a P6/200... I never gave it much
> thought since this was IBM that was doing the testing. :)

I asked Campbell, and he said Pentium 90.

bruce

[brucemo]

Don Getkey wrote:

> If they wanted a closer approximation in speed to test out their eval (or
> whatever) why did they not use PP200s? They should have been still faster,
> by a bunch, but Deep Blue Jr. would have had a much better set of sparing
> partners, capable of throwing a few punches, rather than just taking them.

This match was not meant to measure the degree to which they are better than
micros. They were trying to get a grip on their eval function by grabbing
what they had at hand and running it against another program on what they had
at hand.

I wish I could remember how Campbell put it, but it was something like, the
chess experts they had weren't giving them feedback yet, so he wanted to get
something back right away, with which he could begin to do basic work tuning
the eval function.

The only reason we found out about this at all is that we weren't getting
anything from them right after the match, Bob managed to get some off the
record stuff from Hsu, but Hsu let him publish this scrap because he'd
already told a Newsweek reporter the same thing.

Bob was just trying to fill in gaps, but there wasn't enough supporting
information for anyone to take the result seriously, and there was no way to
ask questions, so this backfired.

bruce

[brucemo]

Robert Hyatt wrote:

> Not particularly. Remember that Rebel and Genius also don't seem to use
> null-move. So it becomes a battle on two fronts: (1) eval vs eval, which
> is a front where everyone (almost everyone) has assumed that DB was horribly
> inferior to the commercial micros, but this has been pretty well shown to be
> totally incorrect now; (2) search vs search. I agree they lose depth with
> the extensions. But *if* the extensions are correct, this is an advantage,
> not a problem, because the rest of us are searching a lot of garbage to a
> deep depth, while it is possible that their extensions are done well enough
> that the paths they search to shallow depths don't mean as much as the ones
> they extend deeply...

Yes, the whole point of your bringing this up at all was to show that their eval
function isn't awful anymore. Actually, your point was to say that some combination
of eval and search is downright god-like.

I don't think you proved this, but I think it can be stipulated that the eval
function shouldn't be a problem anymore.

The Deep Thought eval was notoriously bad, they admit this in papers they publish.
This gave them the reputation of being fast but stupid. When they built DB, there
was no published info, so people assumed (for their own reasons) that this trend had
continued.

I don't know what they had done with that version that played Kasparov last year. I
don't know if that was a DT eval or whether they had attempted to build something
better as of that point. But they did spend an entire year focusing on it after
that match. So of course it got better.

I don't deify them for this though. I respect these guys, and I think that if they
spilled their guts we all could learn lots of stuff. I learned some stuff just from
watching Campbell's lecture to non- chess programmers. But it sounds like in the
end they build their program the same way the rest of us do, by cobbling it
together, eyeballing it, and crossing their fingers.

> : If there is a contradiction between Campbell and Hsu, I am sorry that I did not
> : get it resolved yesterday. I also should have spoken with you before posting,
> : probably.
>
> Not necessary at all, of course. I'm not a "moderator" here. :) But you saw
> from what I posted that there really is no room for misinterpretation. It was
> quite clear from Hsu that they used 1 chip, at 1/10th its normal speed, which
> dropped it to 100K nodes per sec. I still believe 1M nodes per second for the
> hardware, because that matches everything we've seen. IE DB claimed to search
> 200M nodes per second against Kasparov. 2M nodes per sec/chip x 240 chips
> doesn't add up. It might be that the 2M number came from the older hardware
> (ie pre-DB-II, where Hsu redesigned the hardware one more time for GK). But
> that is speculation too, of course (not about the hardware redesign, that's
> fact, but where the 2M number came from).

There was something mentioned about this not adding up, and that brought on the
comment, "Our search efficiency is 25%". I don't know what this means, I'm not a
parallel search guy.

But on two consecutive days Campbell said that it was 2M nps per processor. I have
that in my notes, he said it in his talk. The other reference to 2M nps was after
the talk. And he went to Microsoft the next day and we have another report in here
of 2M nps per processor.

I know what I heard, and in the one case where I detect an inconsistency between
what I heard and what I heard from you, I point this out as a place where I could
have failed to ask the right question, or there could have been some other mistake.

But please don't attack what I report because what I am reporting wasn't distributed
in printed form. I was awake, tried to take good notes, and had some understanding
of what it was I was hearing.

> : The time controls were unclear in your version. They were either standard or
> : unspecified, but I got the idea that they were long. I'd never heard about 10 0
> : or 30 0, from you, I am pretty sure.
>
> I hadn't heard that either I don't believe. I think I asked "were they blitz
> games?" and got a "no"...

Campbell said 30 0 down to 10 0 and that he'd been the one who did this match.

> No, but it tends to lead me to think that he was training against a micro, and
> thought that he was only facing a program that was about 200 (8-2 result)
> better than the micros he was using for training. He was *way* off in that
> estimate...

8-2 is 240. Everybody but you is reporting 2400-2500 for micros, by the way,
including Campbell, who said this as one of the first things in his talk.

I think DB would beat micros worse than it would beat people, and I think that a lot
of the problems involved in rating it would have to do with trying to deal with
result percentage inconsistencies produced by this, but that is my own speculation.

> : I think DB Jr was lucky to win 10-0, since it did drop a couple of points over
> : 40 games.
>
> : And Campbell isn't gloating about this either. He said, quoting from memory,
> : "the score of those ten games happened to be 10-0". You don't say "happened to
> : be", with a straight face, unless you think that was just one possible outcome.
> : And other possible outcomes might include, for instance, 9-1.
>
> That doesn't seem improbable at all. IE, given a machine like DB, full-blown,
> against a 1980-era program like Sargon II on a radio shack machine, would you
> bet your house that sargon wouldn't win a single game? I wouldn't, because
> I have seen plenty of games where search had nothing to do with the outcome.
> Neither did knowledge. Some games are simply won or lost, and even a weak
> player can win them and a strong player can lose them...
>
> So a non-10-0 result is obviously possible. But it is also *very* revealing,
> when you look at it, because certainly the odds of a weaker program thrashing
> a stronger program 10-0 is so close to zero that calling it zero is not too
> far-fetched..

Obviously it was very much stronger. But I think people reacted in here
approximately the way I would have reacted if someone had reported 10-0 in a match
between Crafty and a commercial program -- they questioned the experiment. 10-0 is
a massive brutal drubbing that indicates tremendous superiority or lots of good
luck. Either possibility introduces *questions*.

> I have it in my book "not on purpose". One other discrepancy I noticed is that
> GM Joel Benjamin said in one of the post-game interviews that DB's opening
> database included *every* game they had found in electronic form. The approach
> I use in Crafty (and used in Cray Blitz) was well-discussed, a big book to
> cover everything, a small hand-tailored book to guide the programs down the
> lines that best suite them. Murray's comments seems to say this wasn't done.
> It would be nice to know what they used... :)

They have two books. They have 200,000 positions (as reported from the Microsoft
lecture) that is hand-tuned and checked by Deep Blue (as I heard Campbell say).

The other book is the first 25 moves or so (Campbell was uncertain) pulled from
600,000 games. It doesn't sound like they distill this down to anonymous positions.
It sounds like they keep track of who played the game, when it was played, and what
the result was, and look at that data at the time they want to select a move.
Perhaps they also do a verification search, but this was not mentioned. I thought
this was very interesting.

Campbell said, "I was there when they added that line." That is plain enough for
me.

bruce

[Robert Hyatt]

brucemo (bru...@seanet.com) wrote:
: Eugene Nalimov wrote:
: >
: > Today Murray Campbell gave another lecture, this time at Microsoft. Here
: > are some interesting details:

: I thought about trying to sneak in, but I figured that I might end up dealing

: with the Redmond cops or an ex-boss or something :-)

: > There are 30 general purpose CPUs – one master, and 29 slave. Each has 8

: > specialized chess processors.
: > Each chess processor calculates 5 full plies in hardware. 1.3mln
: > transistors, 2mln nodes/sec

: Ok, then I heard right about the 2M nps.

If so, something doesn't add up somewhere. IE 30 SP proors, X 8 chess
processors = 240 total, X 2M nodes per second = 480M nodeds per second,
which is way over the figures they were quoting last year. IE the blurbs
I saw on the IBM web site claimed 200M nps in the middlegame, which is
closer to 1M nodes per sec / chess processor...

: In that impromptu debugging match that Campbell did I thought we had the idea

: that the thing had been slowed down even further for for purposes of
: handicapping, but what Campbell says tends to contradict this, so obviously
: there is a major misunderstanding.

: > Each general purpose CPU calculates 4-5 full plies in software (w/o

: > extensions) (in my opinion – and as were written by Bob and others - that
: > means that they have a lot of extensions, otherwise they can easily search
: > 2-3 plies deeper)

: Sounds like mainly singular extension, although Campbell told me that they

: have "influence" extensions, which are extensions based upon the influence
: that a particular move has, the example being the idea of extending when you
: push a passed pawn.

: > Parallel search effectiveness is 25-30%

: > Important lines typically 20-30 plies deep; Murray sometimes saw 70
: > 8000 evaluation factors
: > Some evaluating factors are known by any grandmaster, but not written in
: > any textbook
: > non-linear evaluating function
: > they think that singular extension heuristic is very good – that
: > contradicts with what they published 5-6 years ago

: They originally (1980's) published some huge rating point increase due to

: using this, then they scaled it back, but I thought they have always thought
: this was "good"? It's always been their trademark.

the first report claimed a very large rating increase. The later tests and
paper published in the JICCA attributed something like 8 rating points to
singular extensions. But this was calibrated using problem suites rather
than OTB play, which is likely why it turned out to be low..

: > grandmaster-created opening database of 200.000 positions

: > secondary database of 600.000 GM games; interesting selection criteria –
: > not only based on score statistics (as in Crafty), but also on GM and year
: > during one of the match games, there were endgame databases hits (but
: > anyway, it was too late – position was drawn)

: Did anyone ask any good questions?

[Robert Hyatt]

brucemo (bru...@seanet.com) wrote:
: Robert Hyatt wrote:

: > Not particularly. Remember that Rebel and Genius also don't seem to use
: > null-move. So it becomes a battle on two fronts: (1) eval vs eval, which
: > is a front where everyone (almost everyone) has assumed that DB was horribly
: > inferior to the commercial micros, but this has been pretty well shown to be
: > totally incorrect now; (2) search vs search. I agree they lose depth with
: > the extensions. But *if* the extensions are correct, this is an advantage,
: > not a problem, because the rest of us are searching a lot of garbage to a
: > deep depth, while it is possible that their extensions are done well enough
: > that the paths they search to shallow depths don't mean as much as the ones
: > they extend deeply...

: Yes, the whole point of your bringing this up at all was to show that their eval
: function isn't awful anymore. Actually, your point was to say that some combination
: of eval and search is downright god-like.

: I don't think you proved this, but I think it can be stipulated that the eval
: function shouldn't be a problem anymore.

: The Deep Thought eval was notoriously bad, they admit this in papers they publish.
: This gave them the reputation of being fast but stupid. When they built DB, there
: was no published info, so people assumed (for their own reasons) that this trend had
: continued.

: I don't know what they had done with that version that played Kasparov last year. I
: don't know if that was a DT eval or whether they had attempted to build something
: better as of that point. But they did spend an entire year focusing on it after
: that match. So of course it got better.

Actually, there was DT, then DB, then DBII at least. Which are three separate
and distinct chess chip implementations. I remember interesting discussions
about their debugging the new chip early this year before the match... And,
as Murray mentioned at the talk you heard, the "feature recognizers" are
flexible in that they have controls over the values they produce, but if they
want to add a new one, back to the hardware design table they go, because these
"gadgets" operate in parallel within the chess chip and they have to build in
more to do the new things...

: I don't deify them for this though. I respect these guys, and I think that if they

[Robert Hyatt]

Bo Persson (bo.pe...@mbox3.swipnet.se) wrote:
: Robert Hyatt wrote:
: > =

: > brucemo (bru...@seanet.com) wrote:
: > : Eugene Nalimov wrote:

: =2E.. =

: > : > There are 30 general purpose CPUs =96 one master, and 29 slave. Eac=
: h has 8

: > : > specialized chess processors.
: > : > Each chess processor calculates 5 full plies in hardware. 1.3mln
: > : > transistors, 2mln nodes/sec

: > =

: > : Ok, then I heard right about the 2M nps.
: > =

: > If so, something doesn't add up somewhere. IE 30 SP proors, X 8 chess
: > processors =3D 240 total, X 2M nodes per second =3D 480M nodeds per sec=
: ond,
: > which is way over the figures they were quoting last year. IE the blur=
: bs
: > I saw on the IBM web site claimed 200M nps in the middlegame, which is

: > closer to 1M nodes per sec / chess processor...

: > =

: =2E..
: > : > Parallel search effectiveness is 25-30%
: =2E..

: Just guessing,but...

: Could it be that they are counting some sort of "effective nps"?
: 30% of 480M is not too far from 200M.

No. At least not in the 10-0 match, because there was only one chess
processor used, if you read the "Hsu clip" I posted. This 30% figure
is a measure of how much work the serial/sequential search does compared
to the number of nodes searched by the parallel architecture to reach the
same depth. In general, 30% means that the parallel search is searching
a little over 3X the number of nodes required to search the same tree,
but not in parallal, which is a reasonable number. Of course, they are
searching 240 nodes at a time in parallel, which still means they are
running 80X faster than they would with one processor.

: Could it be that the chess processors are idle part of the
: time, while the SP2 does its search?

A bit of the time, yes, because *both* the SP and the chess chips don't
search in parallel... the SP handles the shallow parts of the tree and
hands off deeper pieces to the chess processors. There's a fine balance
in DB where the chess processors on a single SP processor can just keep up
with the SP...

But in the 10-0 match, again, none of this applies as there was no
parallel search at all... The SP was searching part of the tree, the
hardware (single chip) was searching the tips, but they were never
searching at the same time...

: Bo Persson
: bo.pe...@mbox3.swipnet.se

[Bo Persson]

Robert Hyatt wrote:
> =

> brucemo (bru...@seanet.com) wrote:
> : Eugene Nalimov wrote:
=2E.. =

> : > There are 30 general purpose CPUs =96 one master, and 29 slave. Eac=

h has 8
> : > specialized chess processors.
> : > Each chess processor calculates 5 full plies in hardware. 1.3mln
> : > transistors, 2mln nodes/sec

> =

> : Ok, then I heard right about the 2M nps.
> =

> If so, something doesn't add up somewhere. IE 30 SP proors, X 8 chess
> processors =3D 240 total, X 2M nodes per second =3D 480M nodeds per sec=
ond,
> which is way over the figures they were quoting last year. IE the blur=
bs
> I saw on the IBM web site claimed 200M nps in the middlegame, which is
> closer to 1M nodes per sec / chess processor...
> =

=2E..
> : > Parallel search effectiveness is 25-30%
=2E..

Just guessing,but...

Could it be that they are counting some sort of "effective nps"?
30% of 480M is not too far from 200M.

Could it be that the chess processors are idle part of the

time, while the SP2 does its search?

Bo Persson
bo.pe...@mbox3.swipnet.se

[Rolf Tueschen]

Because we have in the meantime a very intense debate with most of my
"points" om the record, I prefere answering only a few interesting
theories. Please try to correct my impression.

But what's concering the details of the architecture of DB or DB Junior
I would rather learn more to be able to judge...

hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:

>Rolf Tueschen (TUESCHEN.MEDIZ...@t-online.de) wrote:
>: hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:

>: >brucemo (bru...@seanet.com) wrote:
>: >: I just attended a lecture by Murray Campbell, one of the folks
>: >: responsible for Deep Blue.

>: >Glad they are releasing details. Now maybe we can see discussions here that
>: >deal more in facts and not in religious opinions and prejudices. That'd be a
>: >nice change for the better.

>: >However, I'll try to get a definitive and precise answer about the nps of
>: >the hardware used for the 10-0 match to make sure there's no misunderstanding.

>: Oh Lord, have mercy with Prof Hyatt, Alabama.

>: The problem you seem to hide, Bob, is simply, WHY the hell you didn't do
>: this long before, because you saw the trouble we had after your first
>: infos.

>I do not *hide* anything.

Sorry, but you did. But it's surely not of the kind you now oppose with

determination. It's a bit more complicated. Let's see also your next two
sentences before I'll explain what I mean.

>I gave a precise explanation of details that

>Hsu relayed to me. I asked him if it was ok to post them... he said
>sure, he had given the same info to a couple of reporters, although he
>had not seen it in print.

First let me politely make clear that I don't think you personally hid

things you already knew. Still you hid things. Why?

Bob, I always come back positively to your status, to your age and to

your education and titles. You once explained to Enrique what it was for
all the titles in rgcc. *Not* the final title but the education in
basical science related methods for instance...

Look, if you were a sophomore or what it's called in USA, I would have
taken your same post about Hsu's snips in that situation in rgcc with
joy and had given you positive feedback without longer digging deaper...

What makes the whole story fishy is that ***you*** with your
image/status want to pretent/play the role of an open mouthed praying
server in iew of his bishop... Your commentary of the conditions of
this info exchange in itself is a scandal, Bob. You're not getting tired
repeating how you asked him if you could post... but you surely knew
before that he had already talked to the press... But this is only the
beginning. Then you had these info snips. ----------

Now, Bob, even Briuce, who's 20 years younger asked Marray better

questions than you to Hsu. How come? You heard these snips. You *knew*
immediately what these scores could be worth for, but you forgot to ask
about some further details???????????????????????

Bob, is this a talk among scientists in USA of 1997? Well, then good

night. Then we still have a chance in Europe to come close to the
american standard in the future. If this is the way of religiously
filled atmosphere of a talk between cc heroes both in their fields...

Bob, if all this was true till here, and I wont doubt it, then you made
an absolute master-clown out of yourself. Then you wrote in springtime
that you planned however to email Hsu and ask him.... and so on.

Bob!!! This is absolutely unneccessary to inform us about your plans of
emailing someone. What I didn't see till today is simply, what were the
results of this email exchange?

Let's talk a little bit of meta stuff.

I saw you post stuff that must by force lead to a big trouble amoung Ed
and the followers. With reason, as I could understand. All this you
could have had forseen. As a scientist, no?

But thereafter you used every a little bit too far projected critic as

sort of proof for an unallowed anti-DB "basher", I recall that notion,
and whatelse. That was unallowed too. Because on the base of your
data/evidence even stupid ignorants like me were able to detect that you
had given a very blue balloon a lift off just to make a little
propaganda for DB and *team*. To back them in their negociations about
new contracts with IBM... in out-of chess regions as we now know.

With that discussion you could break the big anti-DB feelings due to the

behavior against Kasparov. This is called war-side-scene-creating. To
shift the attention.

>not at all...

After some other posts it should be clear by now.

>I assume Murray did the same thing at

>his talk that I did last year at the University of Maryland when I was invited
>up there to talk about Computer Chess, or when I visited the Univesity of
>Waterloo this year to talk about computer chess, or when any other active
>computer chess type gives a presentation.

>: I think we have another *political* dimension....

>: And I also think that the time schedule can also make data fishy.
>: If I need months to publish game notations...
>: I mean what will people like MC tell us if *they* are way beyond 60,
>: 70... y. old? I know it's rhetorical because first we're further in new
>: dimensions and the I hope we will *solve* the real clue about DB without
>: any data from IBM at all.

>: I will keep safe in my memory that *you* for one did never answer when I

>: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>: asked /pointed at the question of this strange delay in publishing
>: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>: promissed data. If we the see that something was really fishy somehow,
>: ~~~~~~~~~~~~
>: you can stop posting about such questions because everybody will no
>: longer take you as scientifically independent in these fields.

>They don't owe anyone anything, contrary to what you seem to believe.

Oh, yes, Bob, they do. You know why? Because of the big manking vs

machine ballyhoo before the show. *Or* the impression/propaganda of a
total/clear win in a *match* should be corrected as fast as possible.

>They owe it to themselves to tell us how they did what they did. I

>always believed that they would do so.

Did you *see* something yet?

>Nothing has changed that. Whether

>they release Kasparov game logs to the public or not is their decision. If
>I were them, and I was going to play another match with him, I wouldn't let
>'em go either. Otherwise I probably would...

What's this? I thought they left chess for the big money transaction,

known as beaming in StarTrek if I got this right. :)

>: My God, you claimed close contacts to these guys, you publish half

>: cooked data snips of Hsu, and suddenly when Bruce comes up with another
>: snips you are prepared to ask them by email (hear hear, you can do
>: that?) about details. I'm repeating myself, but it's so sad.
>: You know I love you but now I feel like being cheated.

Pity that you didn't comment on this. This was meant as the expression

of a deep depression, Bob. Because you made as scientist a real torture
for your fans. How could you crawl so deep on your knees. You, our heroe
on rgcc?

>:Bob, science

>It is very sad that you continue to post here, and try to run everyone
>that has anything technical to say away.

Bob, I never expected that retro from yours. This is the absolute gamma

ray explosion. I was the only one who opposed Ed's cheaty private emails
about you putting terror on people, and now, after Ed put anpther
scandalous thing on a member of this group, and sure, his party soldiers
left the scene with him, and he lost his face in Zwolle, you attack *me*
for having thrown experts out of rgcc?????

Bob, this is now history. It's in Dejas. And you will feel sorry in

future about that quote of yours. I didn't expulse an expert from rgcc,
I didn't intend to do so, and surely I *never* attacked someone who just
posted technicasl stuff here in rgcc. These are all lies, Bob. You have
no evidence at all. Or show it here right now.

I claim here right now, that I did never attack a poster if he had not
posted something strange here against me or against scientifical
standards. You wont find a single counterexample, Bob.

Give a correction on this point.

>In my case, you aren't going to
>succeed, because you aren't that important.

Wait a minute, you just left your usual logic. It's true I'm not that

important and I never claimed expert status. I'm just the Pope. But how
could I succeed when I never tried to achieve that goal? Bob?
If you now also start sissying around where will it all end? If I were
the reason for your departure I would fly immediately to Alabama,
despite all the flies and snakes over there and I put you back into
rgcc. Truely. Either I stayed mute for two years than loosing you and
your posts here in the group, Bob. Uuuuch, those knees are so swollen
hurt by my creeping on the floor, Bob......

>I simply said I was going to

>ask Hsu to be *certain* that I didn't mis-interpret anything he told me.

>IE I believe that he used a one processor machine. Bruce said he thought
>they used something that searched 2M nodes per second, but he wasn't sure
>this was the machine used in the 10-0 match. 2M doesn't ring a bell with
>me at all, so it deserves clarification. But I didn't see anything in his
>report of Murray's talk that contradicted anything I posted after getting the
>info from Hsu...

>: means pushing and dealing and pushing, but not staying mute if a friend
>: is involved. Either you phone him, like you phone Roman, either you
>: email him, "MC, that's not ok, rgcc is burning, you've to provide us
>: with data...", or you ask them here in public.

>: But what you did. You were like the chief of their press center who ran
>: around with this baseball stick to mute all speculations about spooky
>: data you yourself published here in rgcc................... :((((

>: Now it's *your* job to give the micro heroes back what they deserved....

>If I was wrong, I'll post the corrected version. If I was not wrong, I
>will post that again as well. But the micros got what they deserved, which
>was a sound thumping, just like they (we) have been getting since 1987
>when that project was first unveiled...

No, you will correct this one too, you Popeye- schmonzes- Hulk. :)

Now you put terror on me!

>: As Pope I condemn you for a two weeks non-god status here in rgcc.

>In this regard, I am 100% anti-religion...

Doesn't help you if you're still in the boiling pot...

[Eugene Nalimov]

Murray said: "it cannot find that move at a tournament conditions. Maybe
once I'll run it for a week, and we'll see..."

Eugene

David Hanley <da...@nospan.netright.com> wrote in article
<34284E5B...@nospan.netright.com>...

> brucemo wrote:
>
> > The opening in game 6 was part of the hand-tuned book. They added
> > that
> > sac thinking they would never get into this part of the book, and DB
> > liked the resulting position anyway, so they kept it.
>

> That's quite a statement. This implies that DB may even play thesac
> without having it in the book. That is incredible.
>

> dave
>
>

[Robert Hyatt]

brucemo (bru...@seanet.com) wrote:
: Robert Hyatt wrote:
:

: > What is this P90 stuff? I'm not certain, but I thought that when I asked

: bruce

yet another interesting contradiction, if you read the clip from Hsu, because
as you noticed, he clearly said searching at 1/10th the normal speed of the
DB hardware, which was pretty close to the commercial chess program speed.

None of the numbers add up. IE 2M nps that Murray quoted, gives 200K nps
for Rebel, which is faster than any PC I know of to date. the 1M figure
I have seen many times converts to 100K, which is pretty close to a P6/200
speed for Rebel (I have not seen genius produce NPS figures, so don't know
about it, but Rebel is pretty "chatty" like Crafty and does display NPS
all the time).

Can't see any way a P5/90 running Rebel could do anything over 20-25K max.
<sigh> Need more data. :)

[Robert Hyatt]

Ed Schroder (rebc...@xs4all.nl) wrote:
: Bob, I respect you too much to let this pass...

: : You saw the quote I posted. *what* is missing? He explained *exactly*

: : how the DB hardware was configured, the only detail that I didn't ask about
: : was the time control. Which doesn't matter since both programs were playing
: : on equal time. 10-0 is a good result whether it be one second per move or
: : 1 hour per move...

: I can remember very precisely I asked you twice about the used time control.
: Answer-1: "normal time control"

: Then I (and Chris) asked you "Is normal time control 40/2:00"?
: Answer-2: Yes, for most of the games.

If I responded thusly, I believed it at the time. I have not had any
information about the time controls, other than at one point I asked
Hsu if they were blitz games and he said "no".

in light of Bruce's comments, I was obviously wrong about "standard"
being 40/2...

: You painted the following picture...
: - Deep Blue NPS 100,000 (Campbell 2,000,000)

I still believe my NPS figure is correct, as you can see from the
quote I posted directly from Hsu. Bruce said he was not clear on
the speed used for the 10-0 match. Taking that with a direct quote
from Hsu that the DB hardware was running at 1/10th it's normal speed,
and since that "normal speed" has been quoted as 1M nodes per second for
about 2 years, I still stand by this number, as there is no contrary
evidence yet, and the statement from Hsu (which I posted) was very
precise on the conditions...

: - Programs Genius5/Rebel8 (Campbell Genius5/Rebel7)

That's what Murray said (genius/rebel7). Hsu said genius 5/
rebel 8. I am again enclosing the direct quote from Hsu here
for clarity:

We played 10 games against Genius 5 and Rebel 8 (about
5 games each) with a single chess chip, running at about
1/10th speed, or roughly the same speed as the PC programs.

I was expecting to be slightly ahead. The result was a

10-0 slaughter. The process of patching all the holes
that Gary could drive a truck thru paid off much better
than I expected.

As you can see, Hsu *clearly* said Rebel 8. Rebel 7 could well have
been a slip of the tongue by Murray. Or Hsu could have been mistaken.
*I* was not mistaken, as you can see above.

: - Time control mostly 40/2:00 (Campbell 10-30 seconds)

That was a mistake on my part I believe. But not 10-30 seconds.
He said game/30 minutes (Murray). Hsu only said "not blitz games"
So I took that to mean "standard" type games. Game/30 is not
very fast when you factor in thinking on the opponent's time...

: - Used Pc's Pentium Pro 200 (Campbell P90)

If you look at Hsu's post above, cutting DB's single chip by 1/10th
produced an equivalent NPS. DB's chip is at least 1M nodes per
second. Therefore, cutting it by 1/10th is at *least* 100K. If
you take his "roughly the same speed" comment, that would mean a
P6/200 was used, because Rebel won't hit 100K on a P90. I assume
Murray was more forgetful here since the direct quote by Hsu is
somewhat different.

: According to Campbell I would say YES, the 10-0 is real.
: No big deal, I don't even need the games to believe that.

: But the challenge on 40/2:00, AMD 233 still stands.
: Quite a different story...

: NPS 2,000.000 against NPS 120,000 I take it.

what challenge are you talking about? DB? Single chip it sounds
like? Perhaps since DB Jr is going to continue to exist, we might
arrange something. *if* you really want to. I think the outcome
might be surprising to some...

: But you may have the last word about this as I am disappearing.
: Just couldn't resist... :)

: : I have no idea what you mean. Ed has been getting clobbered by Chiptest/

: : Deep Thought/DB prototype for over 10 years. Just like I and every other
: : program author in the world have been getting clobbered. So I see no
: : impact on Ed or Rebel in any fashion. He made a rather odd claim that he
: : thought Rebel was only 50 Elo points below DB, which I *know* is incorrect.

: : And I suspect he knows that too, but that comment was simply the
: : marketeering side of him coming out...

: I don't like such unfounded 'fishy' statements.
: Especially when they are not true.

what I am sure of, is that you are smart enough to know you have absolutely
no chance of beating DB with Rebel 8 or 9 or 10... And I am also absolutely
certain you know that if you played them a 20 game match, it would end up
at least 15-5 in their favor. I believe it would be worse, but that is a
different story. And 15-5 means they are at least 200 Elo rating points
better. *not* 50. So there is *nothing* fishy about my statement. If you
don't know that DB is better, after having seen them in tournaments and
perhaps even losing to them a couple of times, then maybe I'll change my
opinion about you. :) I can guarantee you that Cray Blitz can beat
rebel in a match, and I *know* that Cray Blitz won't beat DB. As I said
many times, NPS is important, if one program doesn't have a significant
advantage to offset it, like a far superior eval. In this case, DB has
both the NPS *and* the eval advantage. It's too much to overcome....

[Ed Schroder]

Bob, I respect you too much to let this pass...

: You saw the quote I posted. *what* is missing? He explained *exactly*

: how the DB hardware was configured, the only detail that I didn't ask about
: was the time control. Which doesn't matter since both programs were playing
: on equal time. 10-0 is a good result whether it be one second per move or
: 1 hour per move...

I can remember very precisely I asked you twice about the used time control.
Answer-1: "normal time control"

Then I (and Chris) asked you "Is normal time control 40/2:00"?
Answer-2: Yes, for most of the games.

You painted the following picture...

- Deep Blue NPS 100,000 (Campbell 2,000,000)

- Programs Genius5/Rebel8 (Campbell Genius5/Rebel7)

- Time control mostly 40/2:00 (Campbell 10-30 seconds)

- Used Pc's Pentium Pro 200 (Campbell P90)

According to Campbell I would say YES, the 10-0 is real.

No big deal, I don't even need the games to believe that.

But the challenge on 40/2:00, AMD 233 still stands.
Quite a different story...

NPS 2,000.000 against NPS 120,000 I take it.

But you may have the last word about this as I am disappearing.

Just couldn't resist... :)

: I have no idea what you mean. Ed has been getting clobbered by Chiptest/

: Deep Thought/DB prototype for over 10 years. Just like I and every other
: program author in the world have been getting clobbered. So I see no
: impact on Ed or Rebel in any fashion. He made a rather odd claim that he
: thought Rebel was only 50 Elo points below DB, which I *know* is incorrect.

: And I suspect he knows that too, but that comment was simply the
: marketeering side of him coming out...

I don't like such unfounded 'fishy' statements.

Especially when they are not true.

- Ed Schroder -

[David Hanley]

Rolf Tueschen wrote:

> hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>
> >brucemo (bru...@seanet.com) wrote:
> >: I just attended a lecture by Murray Campbell, one of the folks
> >: responsible for Deep Blue.
>
> >Glad they are releasing details. Now maybe we can see discussions
> here that
> >deal more in facts and not in religious opinions and prejudices.
> That'd be a
> >nice change for the better.
>
> >However, I'll try to get a definitive and precise answer about the
> nps of
> >the hardware used for the 10-0 match to make sure there's no
> misunderstanding.
>
> Oh Lord, have mercy with Prof Hyatt, Alabama.
>
> The problem you seem to hide, Bob, is simply, WHY the hell you didn't
> do
> this long before, because you saw the trouble we had after your first
> infos.

Probably because it was only *that* important to a fewknuckleheads
who deliberately misinterpret things anyways..

> As scientist -- not cheer-leader of course of DB/IBM - you had the
> dutry
> to fight for further releases.

Why?

dave

[David Hanley]

brucemo wrote:

> David Hanley wrote:
> >
> > brucemo wrote:
> >
> > > The opening in game 6 was part of the hand-tuned book. They added
>
> > > that
> > > sac thinking they would never get into this part of the book, and
> DB
> > > liked the resulting position anyway, so they kept it.
> >
> > That's quite a statement. This implies that DB may even play
> thesac
> > without having it in the book. That is incredible.
>

> Not necessarily. I don't know how long they searched on the sac
> itself,
> or at what point they had DB examine a position to verify it. It
> could
> have been five moves later. That's ten plies, and ten plies is a lot.

I remember reading that they had DB go over the opening book,to
verify it didn't have any 'holes' and that it had position it 'liked'.
Also, I think the whole DB system can, indeed, do ten plies--in a
couple seconds.

dave

[David Hanley]

Robert Hyatt wrote:

> David Hanley (da...@nospan.netright.com) wrote:
> : brucemo wrote:
>
> : > The opening in game 6 was part of the hand-tuned book. They added
>
> : > that
> : > sac thinking they would never get into this part of the book, and
> DB
> : > liked the resulting position anyway, so they kept it.
>
> : That's quite a statement. This implies that DB may even play
> thesac
> : without having it in the book. That is incredible.
>

> : dave
>
> It really isn't... If you remember the Crafty vs ChessMaster 5000
> game
> in the Korrespondence Kup? Crafty sacrificed a knight in a wild
> position,
> but only did so because it was evaluating the position at -.5, and the
>
> knight sac led to a draw by repetition. However, after several moves
> went by, Crafty realized that the sac didn't just lead to a draw, it
> led
> to a forced win... just a very deep one. Deep Blue Junior saw this
> when
> the knight sac was played, with an eval over +2...

Yes, but that's not my point. The original post was related to DB's

kight sac in game 6. It said that the sac was in the opening book, but
the DB liked the resulting position anyways. That means that DB 'sees'
the compensation for sacrificing a knight for a pawn.

Now, it could be that DB sees some short-term way of recovering
som material, a pawn or two, but at each move, plays the 'right'
thing for the 'wrong' reasons, because it thinks it can get back the
material later anyways--but I don't know if that's case here or not.

I have, incendentally, seen programs sac a knight for 2 pawns,
but not typically at the correct time.. :)

> I've been on the wrong end of their search depths too many times. :)
>
> It is formidable in such positions. It likely has no equal in
> positions
> like that, either human or computer...

No doubt. I'd hate to have it play the evans gambit on me!!

dave

[David Hanley]

brucemo wrote:

> In any case, 2M nps, as reported by Campbell twice in two days,
> divided by ten
> is 200K, which is way faster than Rebel, and way way faster than
> Genius, on a
> P5/90.

Hmm... At least rememeber that we have some confusion here that
needsclearing up. Hsu says same node rate, and this would be pretty
easy to
pull off, considering the hardware/sofware combo that was used.

> But this is still weird. I have a hard time dealing with the notion
> of an equal
> NPS, or even a 2X NPS match going 10-0. DB doesn't prune much, it
> doesn't even
> use null-move, and in fact it extends. So it is going to get
> absolutely hashed
> on general search depth.

Curious.. Are you sure that DB doesn't prune much? Remember, they
havecustom, so they can do things like sophisticated move ordering
very,very cheaply.

dave

[brucemo]

Tim Mirabile wrote:
>
> brucemo <bru...@seanet.com> wrote:
>
> >It is interesting that this is something that they voluntarily allowed. I
> >pointedly asked Campbell if this was an opening book disaster that
> >coincidentally worked, and he said "no".
>
> I can't imagine they ever expected, hoped, or dreamed for this book line to ever
> come into play in this match. But when preparing a book you put in lots of
> stuff you never expect to happen, and if the sac is sound, and you have
> confidence in your machine, why not...

No, they didn't expect to see it.

At the 1995 WMCCC in Paderborn, my program sacrificed a piece vs MChess, in book.
The game worked out alright, but I was mortified at the time.

I had thought that something like this might have happened here.

But this is not what happened. They knew about this line and were willing to play
into it.

bruce

[AMIR BAN]

Robert Hyatt wrote:

>
> No, but it tends to lead me to think that he was training against a micro, and
> thought that he was only facing a program that was about 200 (8-2 result)
> better than the micros he was using for training. He was *way* off in that
> estimate...
>

You have this point backwards I think. Kasparov played games with
Hiarcs6 before the match, and admitted to having problems. Based on
this, and maybe other things, he was not so very confident before the
match. Kasparov expressed praise for Hiarcs several times, specifically
stating that it is the program with the best positional understanding.

Amir

[Robert Hyatt]

David Hanley (da...@nospan.netright.com) wrote:
: brucemo wrote:

: > In any case, 2M nps, as reported by Campbell twice in two days,

: > divided by ten
: > is 200K, which is way faster than Rebel, and way way faster than
: > Genius, on a
: > P5/90.

: Hmm... At least rememeber that we have some confusion here that

: needsclearing up. Hsu says same node rate, and this would be pretty
: easy to
: pull off, considering the hardware/sofware combo that was used.

: > But this is still weird. I have a hard time dealing with the notion

: > of an equal
: > NPS, or even a 2X NPS match going 10-0. DB doesn't prune much, it
: > doesn't even
: > use null-move, and in fact it extends. So it is going to get
: > absolutely hashed
: > on general search depth.

: Curious.. Are you sure that DB doesn't prune much? Remember, they

: havecustom, so they can do things like sophisticated move ordering
: very,very cheaply.

: dave

If you knew Hsu, you wouldn't even ask this. :) He refuses to use null-move
because of the inaccuracies it introduces. We had the big discussion about
SEE pruning in the capture search, which he also did not like for the same
reason. Bottom line is *no* forward pruning in DB, although they do lots
of forward extensions...

[Robert Hyatt]

AMIR BAN (ms...@ibm.net) wrote:
: Robert Hyatt wrote:

: >
: > No, but it tends to lead me to think that he was training against a micro, and

: > thought that he was only facing a program that was about 200 (8-2 result)
: > better than the micros he was using for training. He was *way* off in that
: > estimate...
: >

: You have this point backwards I think. Kasparov played games with

: Hiarcs6 before the match, and admitted to having problems. Based on
: this, and maybe other things, he was not so very confident before the
: match. Kasparov expressed praise for Hiarcs several times, specifically
: stating that it is the program with the best positional understanding.

: Amir

Possibly, although I also think he grossly underestimated DB's skill
as well, basing his estimate on Hiarcs + N, where N was too low...

[Tim Mirabile]

hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:

>He also mentioned that at an Icelandic web site he found a post where an
>IM played genius 5 from the game 6 position after Nxe6, and the IM won quite
>easily. Note that the IM was playing *black*.

Yes, it only takes a slight inaccuracy by white to let black untangle and
completely turn the tables. Did the IM play Qe7 or fxe6?

>Still lends credence to my
>original idea that he had tried this as black against Fritz or Hiarcs or
>Whatever and won easily, and decided to "go for it" in this game, knowing he
>could always claim "finger slip" or "memory slip" if he lost... Who knows of
>course, but it's a thought...

I can't believe this because the subsequent defense seemed so weak. If this was
prepared, you'd think he'd have put up a much better fight. Do you think he was
betting it all on DB's first move out of book?

[r...@jojodyne.ruhr.de]

> Robert Hyatt wrote:
>
> >
> > No, but it tends to lead me to think that he was training against a micro,
> > and thought that he was only facing a program that was about 200 (8-2
> > result) better than the micros he was using for training. He was *way*
> > off in that estimate...
> >
>

> You have this point backwards I think. Kasparov played games with
> Hiarcs6 before the match, and admitted to having problems. Based on
> this, and maybe other things, he was not so very confident before the
> match. Kasparov expressed praise for Hiarcs several times, specifically
> stating that it is the program with the best positional understanding.
>
> Amir

SOURCE please !
I never heard that Kasparov admitted to having problems with Hiarcs6. It
was stated several times that Kasparov found Hiarcs6 to be a tougher
opponent then Fritz, but this is far cry from admitting problems !

QUOTE CSS 3/97 PAGE 10:
"Jeden Tag spielte Garry eine Reihe von Sparringspartien - meist auf der
Blitzstufe - gegen die Hiarcs 6.0 Engine. Es war interessant seine
Fortschritte im Spiel gegen die Maschine zu verfolgen. Anfangs, in den
ersten Partien, hat er gerade mal 50% erreicht. Danach waren es 60, dann
70 und zum Schluss ueber 80%. Ich sah wie er Hiarcs nach Strich und Faden
auseinandernahm."

Paraphrased: Every day Kasparov played training games against Hiarcs 6,
mostly on BLITZ timecontrols. In the first games he managed to score 50%
and then gradually got better until he scored more then 80%. Friedel saw
how Kasparov tore Hiarcs 6 apart.

Keep in mind that this happened at BLITZ timecontrols on a PentiumPro 200!

- ROB -

[r...@jojodyne.ruhr.de]

Hyatt wrote:
> Bo Persson (bo.pe...@mbox3.swipnet.se) wrote:
> : Robert Hyatt wrote:

> : > If so, something doesn't add up somewhere. IE 30 SP proors, X 8 chess

> : > processors =3D 240 total, X 2M nodes per second =3D 480M nodeds per sec=
> : ond,
> : > which is way over the figures they were quoting last year. IE the blur=
> : bs
> : > I saw on the IBM web site claimed 200M nps in the middlegame, which is
> : > closer to 1M nodes per sec / chess processor...
> : > =
>
> : =2E..
> : > : > Parallel search effectiveness is 25-30%
> : =2E..
>
> : Just guessing,but...
>
> : Could it be that they are counting some sort of "effective nps"?
> : 30% of 480M is not too far from 200M.
>

> No. At least not in the 10-0 match, because there was only one chess
> processor used, if you read the "Hsu clip" I posted. This 30% figure
> is a measure of how much work the serial/sequential search does compared
> to the number of nodes searched by the parallel architecture to reach the
> same depth. In general, 30% means that the parallel search is searching
> a little over 3X the number of nodes required to search the same tree,
> but not in parallal, which is a reasonable number. Of course, they are
> searching 240 nodes at a time in parallel, which still means they are
> running 80X faster than they would with one processor.

But your arguement about the numbers not adding up is not based on the 10-
0 Match. You simply claim that 30 SP2 * 8 ChessP * 2M = 480M nps where as
the IBM marketing blurb says 200M nps. This has nothing to do with the 10-
0 Match. Maybe they are counting the effective nps in the 200M figure and
this would fit into my and Bo's arguement. Whatever really happened, one
shouldnt construct any contradictions out of marketing data, it mostly is
to unreliable. Just ask Murray or Hsu what the deal is and we will know.
But i wouldnt be suprised if it turns out that they didnt find this match
as important as we obviously find it to be.

[r...@jojodyne.ruhr.de]

Hyatt worte:

> brucemo (bru...@seanet.com) wrote:
> : Eugene Nalimov wrote:

> : >
> : > Today Murray Campbell gave another lecture, this time at Microsoft. Here
> : > are some interesting details:
>
> : I thought about trying to sneak in, but I figured that I might end up
> dealing : with the Redmond cops or an ex-boss or something :-)
>

> : > There are 30 general purpose CPUs ū one master, and 29 slave. Each has 8

> : > specialized chess processors.
> : > Each chess processor calculates 5 full plies in hardware. 1.3mln
> : > transistors, 2mln nodes/sec
>

> : Ok, then I heard right about the 2M nps.
>

> If so, something doesn't add up somewhere. IE 30 SP proors, X 8 chess

> processors = 240 total, X 2M nodes per second = 480M nodeds per second,

> which is way over the figures they were quoting last year. IE the blurbs

> I saw on the IBM web site claimed 200M nps in the middlegame, which is
> closer to 1M nodes per sec / chess processor...
>

> : > Parallel search effectiveness is 25-30%

Did you (Bob) take the parallel search effectiveness into account ?
480M*30/100 = 144M This is probably an average case.
If you take a best case figure round it up you get easily to the 200M nps
figure for the marketing people.

- ROB -

[Rolf Tueschen]

David Hanley <da...@nospan.netright.com> wrote:

>Rolf Tueschen wrote:

>> hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>>
>> >brucemo (bru...@seanet.com) wrote:
>> >: I just attended a lecture by Murray Campbell, one of the folks
>> >: responsible for Deep Blue.
>>
>> >Glad they are releasing details. Now maybe we can see discussions
>> here that
>> >deal more in facts and not in religious opinions and prejudices.
>> That'd be a
>> >nice change for the better.
>>
>> >However, I'll try to get a definitive and precise answer about the
>> nps of
>> >the hardware used for the 10-0 match to make sure there's no
>> misunderstanding.
>>
>> Oh Lord, have mercy with Prof Hyatt, Alabama.
>>
>> The problem you seem to hide, Bob, is simply, WHY the hell you didn't
>> do
>> this long before, because you saw the trouble we had after your first
>> infos.

> Probably because it was only *that* important to a fewknuckleheads
>who deliberately misinterpret things anyways..

Note, that I'm really not in the basical details. But I observed the
struggle about definitions and so on....

>> As scientist -- not cheer-leader of course of DB/IBM - you had the
>> dutry
>> to fight for further releases.

> Why?

You a scientist? :)

Anyway. Please don't read into my critic to Bob more than needed.

I meant this: If I'm in that field such an expert like Bob, I knew'd
what questions I asked.

That's the crucial point. Maybe I get infos I cant publish. But Bob on
the contrary sweared not having got some of those.

This is now the spot.

This is simply unbelievable for me, because it proves that Bob left out
important questions. I dont want to create another flame war, but let me
explain it with a topic I can better understand, what definitions are
important and should be added...

I once mentioned the rather sloppy way of description Althoefer provided
us with his n-hirn experiments. He left out the most important
definitions. To be able to repeat or even criticise seriously his
results. See Popper and followers about challenging hypotheses.
Ok, we're in a newsgroup...

Hope you can understand what I was criticising Bob for. All the infos we
now got had to be asked by him directly. I mean if it's not important
for Bob if it was really 40/2 or 30 0 or 30'' per move..... ;-)

As scientist I have to keep the definitions straight...

Dont laugh about me, I cant afford another Junior to repeat the
experiment but do want to keep the standards alive.

And note further, **therefore** I also made my points against the event
in May. For me Kasparov made an important point. Perhaps most spectators
didn't quite understand it. He mentioned the "scientifical" question,
both sides should try to find an answer for. Excuse my sloppy quotes.

For GK and me too the question was *not* mainly could a human Wch lose
siome games to the machine, but more if the machine as it is right now,
note, as the *machine* is right now, could play such strong chess as to
be able to play high classed chess in front of a human champ? And
perhaps if it could do so even better than the human?

Now look at the conditions in NY. Was there one machine????
Did this machine play through the 6 games?

You know all this too.

Look, we didn't want to attend to a spectacular cheat where it was tried
to "win" by all allowed or not expressively forbidden tricks. That this
is possible for the tuning individuals in the team, this was clear
beforehand. Look again, sorry it may sound a bit arrogant, if Kasparov
has his fight against Karpov. And each third round another super GM is
seated behind a glass wall.... Would you bet that Kasparov wins a very
short match? Don't take me wrong. This twisting could be allowed. But
it should be made clear in advance. No, you may say. This is really the
question between machine and mankind. The human may never know what's
hidden behind the wall.....

Ok, but then Kasparov could not play a 6 games show-match because in
reality you'll have a 3 times a 2 games show or 6 times a 1 game
singular show....

Dave, to find a strategy as a human you must have either more time to
prepare in advance or simply more games in the match, no?

Sorry that I mixed again the Junior 10-0 with the GK/DB match. But in a
way it's all very closely related. Spooky fog wherever you take a look.

> dave

[Chris Carson]

A description of Deep Blue Jr (DB Jr) can be found at:

http://www.chess.ibm.com/press/html/g.6.4.html

Deep Blue DB Jr.
Host computer 32-node RS/6000
RS/6000 SP workstation
Chess
processors 256 16

Avg. speed (in chess
positions per second) 200 million 10 million

Computing
method parallel serial

[Rolf Tueschen]

r...@jojodyne.ruhr.de wrote:

> Whatever really happened, one
>shouldnt construct any contradictions out of marketing data, it mostly is
>to unreliable. Just ask Murray or Hsu what the deal is and we will know.
>But i wouldnt be suprised if it turns out that they didnt find this match
>as important as we obviously find it to be.

No, but the question remains why exactly this half cooked data appeared
when we were talking about such claims as rebel is only 50 points weaker
than DB....?

Taken your idea for granted it's the question why Bob worked this info
in? With unfounded assumptions about time scheduled and the like....

I dont get the technical points in detail. But the first game in may was
sufficient for me to *not* ridiculize Ed's claim... Period.

[Robert Hyatt]

Chris Carson (chris-car...@ti.com) wrote:
: A description of Deep Blue Jr (DB Jr) can be found at:

: http://www.chess.ibm.com/press/html/g.6.4.html

: Computing
: method parallel serial

Which seems to say that Murray and I are both off. I had thought 1M
nodes per sec, he mentioned 2M at the meeting Bruce attended. In
reality is seems closer to 750K if you divide 10M by 16...

[Robert Hyatt]

Chris Carson (chris-car...@ti.com) wrote:
: Bob,

: I agree with you 750K NPS (aprox 1M NPS) seems to be about right.

: Can I also reduce this by 1/10 (75K NPS, aprox 1M NPS) based on
: HSU?

Sounds right to me...

: Rebel on a Pentium 90 would be about 60K NPS based on data from
: Rebel homepage: http://www.xs4all.nl/~rebchess/bench.htm

This I don't know about. Ed will have to respond. I ran Rebel on a
P6/180 and never saw anything over 80K that I remember. I might be
badly mistaken. But if he is doing 60K on a P90, he is *way* faster
than Crafty, which is "supposedly" much faster and dumber. I am
getting around 80K on a P6/200 for reference...

: (20957080 Nodes / 350 sec) = 60K NPS for Rebel 8.0.

: Single processor DBjr = 75K NPS aprox.
: Rebel 8.0 on Pentium 90 = 60K NPS aprox.
: Genius 5.0 on Pentium 90 = ??

: Did I miss something? :)

nope. we are in sync, except perhaps for the Rebel 8 speed. I
will try to test this again tonite on my P6/200 at home....

other than that, you summarized my ideas well. :)

stand by for flames...

[Herbert Groot Jebbink]

Carl Tillotson wrote:

>Ed Schroder wrote:
>
>> But you may have the last word about this as I am disappearing.
>

>What again :-)

Ed will never stop reading rgcc, from time to time he stops posting to
it :-)

---
The Trans-Siberian Railroad Page, http://www.xs4all.nl/~hgj/

[Robert Hyatt]

r...@jojodyne.ruhr.de wrote:
: Hyatt worte:

Yes. Most of us don't consider this. IE, in the last comprehensive test
I ran with Cray Blitz on a 16 cpu machine, this "efficiency" issue went
from maybe 25% worst case, to 75% average, up to 100% best case. But the
NPS remains constant over that, in that a C90 can search N nodes per second
per processor, regardless. It just often searches extra nodes. But I don't
know of anyone that reports that "lower" NPS value you mention, because it
only confuses things. When I have quoted 5M nodes per sec for Cray Blitz on
a T90/32, that is the number of nodes per second it searched on all processors
summed. Efficiency is down to the 50% range on that machine, maybe a little
better...

: - ROB -

[brucemo]

Ed Schroder wrote:

> You painted the following picture...
> - Deep Blue NPS 100,000 (Campbell 2,000,000)
> - Programs Genius5/Rebel8 (Campbell Genius5/Rebel7)
> - Time control mostly 40/2:00 (Campbell 10-30 seconds)
> - Used Pc's Pentium Pro 200 (Campbell P90)

10 0 is "game in ten minutes". 30 0 is "game in 30
minutes".

bruce

[Robert Hyatt]

Rolf Tueschen (TUESCHEN.MEDIZ...@t-online.de) wrote:

: See? This for instance. You claim naively I simply asked... LOL
: In the middle of the NPS he sent me... Hehe.

I don't understand your comment. Hsu *did* send me this info, because
he was following the NPS discussion with a lot of "interest". When I
received it, I did exactly as I said above: I sent a one-liner back to
him asking if it was ok to post this. So what's the "LOL"???

: You asked if it was ok to post, but you forgot to ask for the time
: controls...? Ahem.

The data, as it stood, was interesting. Whether it was blitz, standard,
or correspondence. 10 0 is 10 0, just so there was nothing odd going on.
I did ask if it was blitz, and got a "no". I assumed it was standard. That
turned out to be incorrect. The truth is somewhere in between. But it is
still 10-0...

: But you also fail to realize where's the scandal? Wait another second
: and read below...

: >He responded with the clip I posted, which included

: >the remark "I have already told this to some reporters." So don't try to
: >stick ulterior motives into what I do. Until you produce evidence that you
: >have some sort of telepathic gift, please write about what I *say*. *or*
: >about what you *think*. But *not* about what I think...

: >so *no* I did not know beforehand that he had leaked this to the press.
: >However, at the time I posted the match (10-0) result, I did post the
: >fact that it had been given to the press already. Go dig up my original
: >post on the subject and you'll see that clearly indicated this. So I have
: >no idea of where you are coming from, nor of where you are going.

: >: Now, Bob, even Briuce, who's 20 years younger asked Marray better

: >: questions than you to Hsu. How come? You heard these snips. You *knew*
: >: immediately what these scores could be worth for, but you forgot to ask
: >: about some further details???????????????????????

: >You saw the quote I posted. *what* is missing? He explained *exactly*

: >how the DB hardware was configured, the only detail that I didn't ask about
: >was the time control. Which doesn't matter since both programs were playing
: >on equal time.

: While being not at all in the technical details, Bob, I accuse you of a
: cleasr twistr/cheat or what you want to define that.

: If apples and beans play on the same time control it's not always the
: same. Bob. Oh, my dear. How to explain this to GOD?

It's going to be *very* difficult to explain how two players using the
same control is "not the same."

: Look, Ed stated more than once that rebel is at his best at 40/2, so
: called tournament time control. Did you forget that? So what did you
: want to say with your statement that it doesn't matter because both were
: on the same? Maybe even this is not correct due to different power
: force?

Does this mean rebel plays lousy chess at 5-0? Seems to me Ed did better
against GM players at 5 0 than he did at 40/2 last time I looked at his
results. So I don't completely buy this "better at longer time controls"
as absolute fact...

: But another attack on you.

: You played a bad scientist those days, no?

Don't see where... as I said, the 10-0 result says a lot. The point
for discussion back then was "DB's evaluation sucks". I claimed it
did not. 10 - 0 at roughly equal search speeds says it does not. If
you want to nit-pick one small detail and say "the games weren't long
enough", I'm sure had the games been 1 hour per move, you'd be saying
"the games were too long"...

: You bought info fresh from a collegue and forgot to ask for the needed
: experimental conditions. Bob?

I got *most* of them. I didn't consider this important enough to spend
days checking every little detail, about the weather, Hsu's health, the
state of the moon, and so forth. The one paragraph from Hsu contains
data that we did not have prior to that point. It contains useful info
that we did not have prior to that point. Even if every game was 1 minute
per game, 10-0 says something about the two programs. A dumb program should
be beaten by a smart program any time they are equalized in search speed.
This shows that at least DB is not dumb...

[Robert Hyatt]

r...@jojodyne.ruhr.de wrote:

: Hyatt wrote:
: > Bo Persson (bo.pe...@mbox3.swipnet.se) wrote:
: > : Robert Hyatt wrote:

: > : > If so, something doesn't add up somewhere. IE 30 SP proors, X 8 chess

: > : > processors =3D 240 total, X 2M nodes per second =3D 480M nodeds per sec=
: > : ond,
: > : > which is way over the figures they were quoting last year. IE the blur=
: > : bs
: > : > I saw on the IBM web site claimed 200M nps in the middlegame, which is

: > : > closer to 1M nodes per sec / chess processor...

: > : > =
: >
: > : =2E..

: > : > : > Parallel search effectiveness is 25-30%

: > : =2E..

: >
: > : Just guessing,but...
: >
: > : Could it be that they are counting some sort of "effective nps"?
: > : 30% of 480M is not too far from 200M.
: >
: > No. At least not in the 10-0 match, because there was only one chess
: > processor used, if you read the "Hsu clip" I posted. This 30% figure
: > is a measure of how much work the serial/sequential search does compared
: > to the number of nodes searched by the parallel architecture to reach the
: > same depth. In general, 30% means that the parallel search is searching
: > a little over 3X the number of nodes required to search the same tree,
: > but not in parallal, which is a reasonable number. Of course, they are
: > searching 240 nodes at a time in parallel, which still means they are
: > running 80X faster than they would with one processor.

: But your arguement about the numbers not adding up is not based on the 10-
: 0 Match. You simply claim that 30 SP2 * 8 ChessP * 2M = 480M nps where as
: the IBM marketing blurb says 200M nps. This has nothing to do with the 10-
: 0 Match. Maybe they are counting the effective nps in the 200M figure and

: this would fit into my and Bo's arguement. Whatever really happened, one

: shouldnt construct any contradictions out of marketing data, it mostly is
: to unreliable. Just ask Murray or Hsu what the deal is and we will know.
: But i wouldnt be suprised if it turns out that they didnt find this match
: as important as we obviously find it to be.

But they are related. Because Hsu said "one chess chip at 1/10th of its
normal speed, which was approximately to the PC chess program."

First question, what is the "normal speed"??? We need to know that to
resolve the other questions. IE if "normal" is 1M nps, then 1/10th is
100K, which is what I'd claim rebel does on a P6, from my tests at
home. If normal is 2M, then 1/10th is 200K, which is double what Rebel
does on a P6/200. Either case, however, is *way* over what Rebel would
do on a P5/90, which is why I had originally assumed the P6/200 was
used. I find the P5/90 numbers suspicious, because that would mean that
the quote from Hsu was not as accurate as I'm used to seeing from him,
ie "approximately equal" and "5x-10x faster" are *not* the same...
If rebel ran on a P5/90, 200Knps is about 10x faster, which is not very
equal at all... and if we take my 1M nps for DB's single chip, it is
still 5x faster than the P5/90, which is still not very close...

[Rolf Tueschen]

hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:

>Rolf Tueschen (TUESCHEN.MEDIZ...@t-online.de) wrote:

>: Hyatt>I do not *hide* anything.

>: Sorry, but you did. But it's surely not of the kind you now oppose with
>: determination. It's a bit more complicated. Let's see also your next two
>: sentences before I'll explain what I mean.

>: >I gave a precise explanation of details that
>: >Hsu relayed to me. I asked him if it was ok to post them... he said
>: >sure, he had given the same info to a couple of reporters, although he
>: >had not seen it in print.

>: First let me politely make clear that I don't think you personally hid
>: things you already knew. Still you hid things. Why?

>You made the claim. You offer the evidence to support that claim.

I know I made a claim but not of that sort you assumed. And I also gave
evidence yet. For further details see below...

>I
>later posted the relevant part of the note from Hsu that *clearly* showed
>what he said took place. It doesn't match exactly with what Murray said.
>Why, I don't know. One of them is clearly at odds with the other over
>what happened, or, possibly, Bruce misunderstood/misinterpreted/mis-remembered
>what was said. however, the quote I posted leaves *zero* room for any sort
>of interpretation that I can see..

>: Bob, I always come back positively to your status, to your age and to
>: your education and titles. You once explained to Enrique what it was for
>: all the titles in rgcc. *Not* the final title but the education in
>: basical science related methods for instance...

>: Look, if you were a sophomore or what it's called in USA, I would have
>: taken your same post about Hsu's snips in that situation in rgcc with
>: joy and had given you positive feedback without longer digging deaper...

>: What makes the whole story fishy is that ***you*** with your
>: image/status want to pretent/play the role of an open mouthed praying
>: server in iew of his bishop... Your commentary of the conditions of
>: this info exchange in itself is a scandal, Bob. You're not getting tired
>: repeating how you asked him if you could post... but you surely knew
>: before that he had already talked to the press... But this is only the
>: beginning. Then you had these info snips. ----------

>No I did not. Hsu has corresponded with me on many occasions. I once
>quoted something he said and he asked me to not quote him here, for what
>is now obvious reasons, since everything gets so twisted by "some"

You said it. But this a normal happening in usenet. You also twist often
enough... I would never assume *intentionally* of course. Wait a second
we have an actual example...

>that
>it isn't worth the trouble. When he sent me this note, which was in the
>middle of the Rebel NPS discussion with Ed, I simply asked him if it was
>O.K. to post this.

See? This for instance. You claim naively I simply asked... LOL

In the middle of the NPS he sent me... Hehe.

You asked if it was ok to post, but you forgot to ask for the time
controls...? Ahem.

But you also fail to realize where's the scandal? Wait another second
and read below...

>He responded with the clip I posted, which included
>the remark "I have already told this to some reporters." So don't try to
>stick ulterior motives into what I do. Until you produce evidence that you
>have some sort of telepathic gift, please write about what I *say*. *or*
>about what you *think*. But *not* about what I think...

>so *no* I did not know beforehand that he had leaked this to the press.
>However, at the time I posted the match (10-0) result, I did post the
>fact that it had been given to the press already. Go dig up my original
>post on the subject and you'll see that clearly indicated this. So I have
>no idea of where you are coming from, nor of where you are going.

>: Now, Bob, even Briuce, who's 20 years younger asked Marray better
>: questions than you to Hsu. How come? You heard these snips. You *knew*
>: immediately what these scores could be worth for, but you forgot to ask
>: about some further details???????????????????????

>You saw the quote I posted. *what* is missing? He explained *exactly*
>how the DB hardware was configured, the only detail that I didn't ask about
>was the time control. Which doesn't matter since both programs were playing
>on equal time.

While being not at all in the technical details, Bob, I accuse you of a
cleasr twistr/cheat or what you want to define that.

If apples and beans play on the same time control it's not always the
same. Bob. Oh, my dear. How to explain this to GOD?

Look, Ed stated more than once that rebel is at his best at 40/2, so

called tournament time control. Did you forget that? So what did you
want to say with your statement that it doesn't matter because both were
on the same? Maybe even this is not correct due to different power
force?

But another attack on you.

You played a bad scientist those days, no?

You bought info fresh from a collegue and forgot to ask for the needed
experimental conditions. Bob?

Read please down below this post, what I wrote to Dave about your
performance in relation to the scientifical standards... :)

>: >: As scientist -- not cheer-leader of course of DB/IBM - you had the dutry

>: >: to fight for further releases.

>: >They have *always* been doing this Rolf, it is simply your ignorance that

>: >not at all...

Another dissence between us...

Bob, in a certain theoretical context you're right about this. But not
regarding the concrete situation.

As I have understood the situation for the pro's like Ed or Richard,
this guys make a living out of their hobby. They never had the chance to
fight against the human Wch for a big show with TV broadcast worldwide
and so on. Ok all micros did here and there a smaller blitz or whatever
exhibition. But I fear it did cost them more than it brought money
directly back...

But with DB its another story. This was announced as the final decisive
match / scientifical experiment for the question 'mankind or machine'.
hope you see that. And I hope you also agree that it's acrtually a bit
too early to claim that DBII should be really stronger than the best
humans... Again talk about game one if you have doubts... And read
again down below my new statement answering Dave.

In this case they *have* a responsibility first to provide kasparov of
some during the match info and the rest of us let's say in some weeks.
You ask why?? Well I explained it and first of all it was Kasparov
himself who was sure that these were the conditions of the *show*. Note,
show.

And another explanation for my claim of cheat.

If they continued it was another question as you pointed out, but I
heard they said good-bye to further challenges because they seem to
think -- falsely -- that the question mankind-machine was already
sufficiently answered... It was not of course.

>???

>: >:Bob, science

Yeah you unresponsible warrior, you claim being blind but on the other
side you dare to talk about group specific socio-psychological
conditions and the need to build another one for safer posting for elite
experts. LOL It is this blindness which seems to prove for me again
'Alekhine's trap' I gave a lecture about last year in "misc". :)
On one side the expert and suddenly the illogical average guy...
It's always funny to watch. Bob, ask Roman, but it's so contradictorious
(?) if you want to make safe your kingside and you want to open against
all standards with 1.f3. Well, it's up to you to decide for it. :)

>You continue to bring that
>up in nearly every post you write. It does *nothing* for the credibility of
>the rest of what you write when you continue to harp on that.

You're right on this one. But what led you to believe that I wanted
something like this? No, I repeat this for totally other reasons, Bob.
And your reaction too is a clear indicator for me to continue like this
because it really seems to be *very* effective... And I tell you my
anti-nazi politics is *very* effective talking about Ed and Chris who
both defended Czub's nazi-like stuff by attacking me like mad. :)

> It is past
>history, Ed has left, so it is time to move on.

You must be joking. Ed has left. Yesterday his newest post. And to tell
the truth I even wrote him to stay right here on this occasion... Ed and
leaving. Seems not fit together. This group is also Ed's group and I
hope he will come back after a good refreshment. But I expect a few
explanations, that's also very clear. But Bob, this is not your
business, no?

>: Bob, this is now history. It's in Dejas. And you will feel sorry in
>: future about that quote of yours. I didn't expulse an expert from rgcc,
>: I didn't intend to do so, and surely I *never* attacked someone who just
>: posted technicasl stuff here in rgcc. These are all lies, Bob. You have
>: no evidence at all. Or show it here right now.

>: I claim here right now, that I did never attack a poster if he had not
>: posted something strange here against me or against scientifical
>: standards. You wont find a single counterexample, Bob.

>: Give a correction on this point.

>Simply read thru *this* post. you made several statements that are
>incorrect...

But I beg you pardon, Bob. I wrote elsewhere that also the Pope had the
right to make a mistake. Only if he's speaking ex cathedra, he only
tells the truth. But this is another story...

But again, I said I did never attack someone at hominem by insults, but
that I only defended myself when I was attacked in a very unfair manner.

Incorrect statements could be corrected, Bob, but disgustful insults,
characterassassination has deeper and further consequences.

Therefore I attacked Ed so fierecely and also in the case where he did
this against you. But it seems you never saw what I had a fight about
this when you were on vacations...?

Again, give evidence where I did insult out of the blue or correct you
former statement.

>: >In my case, you aren't going to
>: >succeed, because you aren't that important.

>: Wait a minute, you just left your usual logic. It's true I'm not that
>: important and I never claimed expert status. I'm just the Pope. But how
>: could I succeed when I never tried to achieve that goal? Bob?
>: If you now also start sissying around where will it all end? If I were
>: the reason for your departure I would fly immediately to Alabama,
>: despite all the flies and snakes over there and I put you back into
>: rgcc. Truely. Either I stayed mute for two years than loosing you and
>: your posts here in the group, Bob. Uuuuch, those knees are so swollen
>: hurt by my creeping on the floor, Bob......

>I simply said that *you* are *not* going to run *me* off from here.

Gush, Bob, you're so simply today. All what you did was simply. :)

And again you failed to grasp the meaning of my post.

I claimed that you argued illogical. I could not throw you out because I
was too unimportant. Again, that's bull. Because in my case you see that
I myself could not be hrown out by such experts like CW ES TC and MB.
So my claim was that you unneccessarily brought this point that I'm not
important. Ok, Bob. I never felt importance. I have not the Edian
illness of status lurking. I'm firm, educated, adult enough to be able
to concentrate on my own standards. And dont forget, I'm young at 22.
Only if I did hurt those I would take a vacation. But not if you wrote
each day a letter that I'm not important, or Czub, that I'm mentally
ill, and so on. :)

You again missed the clue. You jumped in all by yourself. :)

>He'd better come prepared
>for a struggle... And he'd better be serious about trying, too...

==================================================

Here now my quote of the post to Dave where again the attack on your
unscientifical proceduring is explained and the scandal of may in NY.
______________________________________________________

[Rolf Tueschen]

Carl Tillotson <anti...@anti.spam.somewhere.com> wrote:

>In article <60bq04$1eo$1...@news2.xs4all.nl>, Ed Schroder wrote:

>> But you may have the last word about this as I am disappearing.

>What again :-)

Stop this, Carl. That is a real and honest job.

In Paris they call it frou frou, or so. Place de Pigalle. Each evening a
new girl on stage... And then you have to change the clothes for each
new appearance.

Well, it's not easy to be a primadonna assoluta. Cant help but LOL.
Or for our younger children, it's like Jack in the box. Hello, it's me
again.

Tell me someone that Ed doesn't have a good portion of humour behind his
ears...

>Carl Tillotson
>=============

>Lancashire Chess Association Homepage
>http://www.netcomuk.co.uk/~lca/index.htm

>For the latest news, reviews and events happening on the
>Lancashire Chess scene, visit the Lancashire Chess Association
>homepage.

[Rolf Tueschen]

h...@xs4all.nl (Herbert Groot Jebbink) wrote:

>Carl Tillotson wrote:

>>Ed Schroder wrote:
>>
>>> But you may have the last word about this as I am disappearing.
>>

>>What again :-)

>Ed will never stop reading rgcc, from time to time he stops posting to
>it :-)

But he stops reading while travelling to the court in Zwolle I hope?

[Chris Carson]

Bob,

I agree with you 750K NPS (aprox 1M NPS) seems to be about right.

Can I also reduce this by 1/10 (75K NPS, aprox 1M NPS) based on
HSU?

Rebel on a Pentium 90 would be about 60K NPS based on data from
Rebel homepage: http://www.xs4all.nl/~rebchess/bench.htm

(20957080 Nodes / 350 sec) = 60K NPS for Rebel 8.0.

Single processor DBjr = 75K NPS aprox.
Rebel 8.0 on Pentium 90 = 60K NPS aprox.
Genius 5.0 on Pentium 90 = ??

Did I miss something? :)

Best Regards,
Chris Carson

--
Best Regards,
Chris Carson email: chris-...@ti.com

[Rolf Tueschen]

brucemo <bru...@seanet.com> wrote:

>It is obvious who should clarify this, but I don't know if they want to be
>involved in this discussion. I could email Campbell to see if he'll say
>anything else, but he may still be out on the road, and I already feel like
>I asked a lot of questions.

And a few good ones, Bruce. Don't care about the future. You finally are
perceived as independent. Why did you ask if you should have contacted
Bob before? That was a fall back into former attitudes.

I really think that the DB team to Kasparov and Bob to the group did
play a very bad game. I thought they were scientists at first. Ok, if
the money comes, not bad at all. But cheating all of us with a cold war
like boycott for details ----

And what Bob now tries to explain with "simply" minor important
negligence is a scandal because I'm now understanding. Hsu smiles in the
background while the micro heroes struggle with Bob. Hsu had a nice
idea. Why not sending Bob a strong baseball stick? Thought and done.
Bob, almost paralysed, as he wants to make us believe, takes that stick
and doesn't even think of it for a moment to ask Hsu if this weapon is
registered or not. So he informed us about his new tool. No, he then
said, it's absolutely safe, it's from DiMaggio in person. Dont worry, I
checked it with Hsu... And almost all serves were paralyzed by all
these big names... When Bob threatened to use his stick if further
"some" sneaky b a s h e r s of the DB dared to open their lips
for an unvisible millimeter, the scene was "cleaned" of all
extremistical critics. Nuts, loners and mothersons.

And now. We heard the truth, no we heard some contradictions.
This stick was used by OJ I heard some whispering. No, it was in real a
tiny tooth sticker Hsu always used already in NY during the big fight...

Haha.

I thought about all that.

And here are my 2 cts.

Some of the "team" wanted to leave for the money. But others wanted to
continue the series of challenges as before. What we see is the typical
Chicago trouble between the lower east of Frankie Lamentino against the
south boys of Al The Face Di Angelo. It's a clear war for the power. And
I think we all hope that at least the "challenge" crew will win, no?
My bet is the *chinese* fraction of Hsu will profit of the man slaughter
and then announce their comeback.

For me Bob said enough. It's the same with Ed. **Leaving** chess for
another job? And reading rgcc in the closet.... Hehehehe.

And while being sure that Bob is on our side well wishing that "they"
will continue to play, I forgive him all his little sins. Because he did
it to stir our attention.

How about that? Would this be a good script for Hollywood? :)

BTW Best wishes for Peter Falk. And late congratulations for his
anniversary. In case he also reads the group.

It seems as if I learned a lot from Columbo. He also came always back
for another question.... Head slightly bent like a real Pope.

The Pope

>bruce

[Carl Tillotson]

In article <60bq04$1eo$1...@news2.xs4all.nl>, Ed Schroder wrote:

> But you may have the last word about this as I am disappearing.

What again :-)

[brucemo]

Robert Hyatt wrote:

> First question, what is the "normal speed"??? We need to know that to
> resolve the other questions. IE if "normal" is 1M nps, then 1/10th is
> 100K, which is what I'd claim rebel does on a P6, from my tests at
> home. If normal is 2M, then 1/10th is 200K, which is double what Rebel
> does on a P6/200. Either case, however, is *way* over what Rebel would
> do on a P5/90, which is why I had originally assumed the P6/200 was
> used. I find the P5/90 numbers suspicious, because that would mean that
> the quote from Hsu was not as accurate as I'm used to seeing from him,
> ie "approximately equal" and "5x-10x faster" are *not* the same...
> If rebel ran on a P5/90, 200Knps is about 10x faster, which is not very
> equal at all... and if we take my 1M nps for DB's single chip, it is
> still 5x faster than the P5/90, which is still not very close...

There is no way that you are going to argue that I mis-heard Campbell, or
that Campbell is wrong about what happened and Hsu is right, by doing this
kind of math.

It is obvious who should clarify this, but I don't know if they want to be
involved in this discussion. I could email Campbell to see if he'll say
anything else, but he may still be out on the road, and I already feel like
I asked a lot of questions.

bruce

[Harald]

quoting a mail from r...@jojodyne.ruhr.de

> > You have this point backwards I think. Kasparov played games with
> > Hiarcs6 before the match, and admitted to having problems. Based on
> > this, and maybe other things, he was not so very confident before the
> > match. Kasparov expressed praise for Hiarcs several times, specifically
> > stating that it is the program with the best positional understanding.
> >
> > Amir
>
> SOURCE please !
> I never heard that Kasparov admitted to having problems with Hiarcs6. It
> was stated several times that Kasparov found Hiarcs6 to be a tougher
> opponent then Fritz, but this is far cry from admitting problems !
>
> QUOTE CSS 3/97 PAGE 10:

> Paraphrased: Every day Kasparov played training games against Hiarcs 6,
> mostly on BLITZ timecontrols. In the first games he managed to score 50%
> and then gradually got better until he scored more then 80%. Friedel saw
> how Kasparov tore Hiarcs 6 apart.
>
> Keep in mind that this happened at BLITZ timecontrols on a PentiumPro 200!
> - ROB -

Right, and if he only had Fritz4 as comparison of course Hiarcs is
stronger. Like Rebel, MChessPro, Genius ....

Harald Faber

[Robert Hyatt]

brucemo (bru...@seanet.com) wrote:
: Robert Hyatt wrote:

: > First question, what is the "normal speed"??? We need to know that to
: > resolve the other questions. IE if "normal" is 1M nps, then 1/10th is
: > 100K, which is what I'd claim rebel does on a P6, from my tests at
: > home. If normal is 2M, then 1/10th is 200K, which is double what Rebel
: > does on a P6/200. Either case, however, is *way* over what Rebel would
: > do on a P5/90, which is why I had originally assumed the P6/200 was
: > used. I find the P5/90 numbers suspicious, because that would mean that
: > the quote from Hsu was not as accurate as I'm used to seeing from him,
: > ie "approximately equal" and "5x-10x faster" are *not* the same...
: > If rebel ran on a P5/90, 200Knps is about 10x faster, which is not very
: > equal at all... and if we take my 1M nps for DB's single chip, it is
: > still 5x faster than the P5/90, which is still not very close...

: There is no way that you are going to argue that I mis-heard Campbell, or
: that Campbell is wrong about what happened and Hsu is right, by doing this
: kind of math.

I wouldn't argue that either was right or wrong, not knowing more about who
did what. For the hardware questions, I'd suspect Hsu is dead on. He
designed the whole thing, and revised the design several iterations until
he reached the DB-II configuration. I strongly believe that 1M (or a little
less) is the right number for the current hardware, based on other
communication with Hsu and based on the IBM web site with DB info last year.

As far as Rebel's speed goes, anyone can figure that out... I just ran it
on my Gateway 2000 P6/200 with 64 megs of EDO RAM, and I got 80K average in
a complicated position with queens, two rooks each, and two minor pieces
each, plus 7 pawns each. So the P90 numbers of 60K are wrong, whereever
that came from. The very slowest DB/1chip could run at 1/10th normal speed
would be 750Knps, max (taking Murray's number) is 2M. 1/10th of that
would be 75K to 200K depending on the right number. That is right on top
of a P6/200. If Murray's right, and they used a P90, they were about 4x too
fast, best case, and 10X too fast worst case.

Murray's been the eval person there for a long time, so he was most likely
present when the games were played, as he would be interested in watching how
his tuning efforts had paid off. If he said P90, I'd assume P90 was right.
However, it is also possible that it wasn't important enough at the time and
he simply thought it was a P90. Maybe we'll find out...

[Don Getkey]

In article <342A72...@ti.com>, Chris Carson
<chris-car...@ti.com> writes:

>A description of Deep Blue Jr (DB Jr) can be found at:
>
>http://www.chess.ibm.com/press/html/g.6.4.html
>
> Deep Blue DB Jr.
>Host computer 32-node RS/6000
> RS/6000 SP workstation
>Chess
>processors 256 16
>
>Avg. speed (in chess
>positions per second) 200 million 10 million
>
>Computing
>method parallel serial
>
>

This "10 million" figure does not seem to jive with a recent article By
GRANT McCOOL that states DB Jr. searches at "20 million"! The article
below agrees that DBll does search at 200 million, so I would assume that
this article might also be correct about DB Jr.s search speed?

NEW YORK (Reuter) - IBM has retired Deep Blue, the supercomputer that made
history when it
defeated world champion Garry Kasparov in May, a company spokeswoman said
Tuesday.

Melinda McMullen, spokeswoman for IBM Research, which is operated by
International Business
Machines Corp, said a less powerful version of the massive calculating
machine called Deep Blue
Junior would continue to play demonstration games.

"What we told Garry Kasparov is that we do not anticipate a rematch in the
foreseeable future,"
McMullen said. "Our scientists said they really did want to move on to
other grand challenges and
the company wants to."

Russian grandmaster Kasparov, world champion since 1985 and considered the
strongest player in
the history of the ancient game, said in a statement from his home in
Moscow that he was "very
disappointed" by the IBM announcement. He had challenged IBM to another
match after his defeat
in New York in May.

Kasparov, disappointed and bitter about his stunning loss in the six-game
match, wanted a "winner
takes all" contest. The May match had a prize fund of $1.1 million with
$700,000 going to the
winner and $400,000 to the loser.

"This action has the public appearance of an investor cashing in their
chips on the stock market -
take your profits and run," Kasparov said.

Deep Blue is an RS/6000 SP parallel processor with specialized microchips
for chess and calculates
at a rate of 200 million moves per second. It was the first chess-playing
computer system to defeat
a reigning world champion when it won two games, drew three and lost one
in the six-game match.

In his statement, Kasparov said that Deep Blue team leader Dr. 'CJ' Tan
had "promised to publish
the computer's details for the scientific world. I trust this will happen
because 300 million chess
players worldwide await the answers."

Kasparov said that he believed it was not too late for IBM to announce a
"sporting match", which
would benefit the company and the world of chess.

McMullen said the Deep Blue team of scientists who created the system
would continue to tour the
United States and overseas to demonstrate Deep Blue Junior, which
calculates at a rate of 20
million moves per second.

"As long as the public is interested in the story of Deep Blue we will go
out and tell it," McMullen
said.

The New York match and a contest in Philadelphia in 1996 was part chess
match and part research
project to help build computers that can make complex, simultaneous
calculations for applications
such as weather forecasting, air traffic control and molecular dynamics.

yours in chess,
Don

Ramsey MN USA

[Rolf Tueschen]

dong...@aol.com (Don Getkey) wrote:

Garry, I'm the one for repeating this on and on to the djorks of RGCC...

>Kasparov said that he believed it was not too late for IBM to announce a
>"sporting match", which
>would benefit the company and the world of chess.

>McMullen said the Deep Blue team of scientists who created the system
>would continue to tour the
>United States and overseas to demonstrate Deep Blue Junior, which
>calculates at a rate of 20
>million moves per second.

>"As long as the public is interested in the story of Deep Blue we will go
>out and tell it," McMullen
>said.

I hate it but it must be stated. This woman's either not comprehensible
for these questions or she's not telling the truth. Because the truth is
they didn't provide the world with the material they had announced.

True is that they seem to travel around to get some money and public
attention. Like Muhammed Ali. But he's at least on the welfare tour for
the handicapped.

But this seems to be the american way. Kasparov had to know that...

>The New York match and a contest in Philadelphia in 1996 was part chess
>match and part research project

Hear. Hear.

>to help build computers that can make complex, simultaneous
>calculations for applications
>such as weather forecasting, air traffic control and molecular dynamics.

Hope that they don't get the Michael Johnson illness, that "forced" the
olympic winner to hide after the event in shame. At least in Europe he
didn't show probably due to the intense doping controls over here...
Only this year, but then was beaten in most races... seems clear *why*.

...iiiif I were a rich man, yabidaibidabiday, I would play the deepest
Blue, dibitdibit.....

The guys should be careful. In a land where even Carl Lewis had to crawl
to become popular, nothing could be guaranteed. But a good image is
quickly lost. If that's not the case already with the team....

Your advisor in *deep* philosophy, the Pope Rolf

[Chris Carson]

Bob,

I checked my numbers from the Rebel web page, I had the numbers for a
Pentium Pro 200 (not Pentium 90). This looks more valid now.

Rebel homepage: http://www.xs4all.nl/~rebchess/bench.htm

Rebel 8.0 Pentium 90 = 20K NPS
Rebel 8.0 Pentium Pro 200 = 61K NPS

Deep Blue JR: http://www.chess.ibm.com/press/html/g.6.4.html

1/10 speed Single Chess Processor Deep Blue Jr (DBjr) = 75K NPS

At least from my point of view, we are in sync and the
reports by HSU look reliable.

I would add that Rebel is a great program, this just
continues to show that Deep Blue has some remarkable
accomplishments (in my opinion).

Robert Hyatt wrote:
>

[SNIP]

> nope. we are in sync, except perhaps for the Rebel 8 speed. I
> will try to test this again tonite on my P6/200 at home....
>
> other than that, you summarized my ideas well. :)
>
> stand by for flames...

--

[Chris Carson]

Don,

Thanks for posting this article (could I get the URL?)

I do not know why the descrepancy. The IBM web site was
around when we started this discussion last spring and
represents an average speed (as stated on the web page).

see: http://www.chess.ibm.com/press/html/g.6.4.html

Don Getkey wrote:
>

[snip]

> This "10 million" figure does not seem to jive with a recent article By
> GRANT McCOOL that states DB Jr. searches at "20 million"! The article
> below agrees that DBll does search at 200 million, so I would assume that
> this article might also be correct about DB Jr.s search speed?
>

[snip]

[Robert Hyatt]

Don Getkey (dong...@aol.com) wrote:
: In article <342A72...@ti.com>, Chris Carson
: <chris-car...@ti.com> writes:

: >A description of Deep Blue Jr (DB Jr) can be found at:
: >
: >http://www.chess.ibm.com/press/html/g.6.4.html
: >
: > Deep Blue DB Jr.
: >Host computer 32-node RS/6000
: > RS/6000 SP workstation
: >Chess
: >processors 256 16
: >
: >Avg. speed (in chess
: >positions per second) 200 million 10 million
: >
: >Computing
: >method parallel serial

: >
: >

I think that this is actually fairly close. Supposedly the chess
processors can search up to 2.4M nodes per second each, when running
at max speed. Also, supposedly, all of the processors don't run at
this max speed. And then there is the issue of the SP having to
feed the chess processors positions to search. While I haven't seen
any recent numbers, once upon a time, at an ACM event, Hsu and Murray
said that they were running the chess processors at about 70% of their
max duty cycle. It's pretty complex to figure out exactly how fast
they are going based on the above info. But 70% of 2.4M is about 1.6M
or so. It might be that they are running below 70% now, I don't
know. Best case would be around 24M for junior based on this,
worst case is probably the 10M number above.

When they publish details, we'll find out. By the way, for Rolf,
who is always nagging about why they have published something, here
is an important question:

"Have *you* ever published something in a significant scientific
technical journal? (of course not) Do you have *any* idea of
the lead time from writing an article until it is published in
such a journal? (of course not) Here is a clue: the typical
lag from submission to publication is over one year. One year
is considered very quick turnaround, while many journals operate
on a 2-year delay. So back off and give them time."

[Robert Hyatt]

Rolf Tueschen (TUESCHEN.MEDIZ...@t-online.de) wrote:
: dong...@aol.com (Don Getkey) wrote:

: >This "10 million" figure does not seem to jive with a recent article By

: >GRANT McCOOL that states DB Jr. searches at "20 million"! The article
: >below agrees that DBll does search at 200 million, so I would assume that
: >this article might also be correct about DB Jr.s search speed?

: >NEW YORK (Reuter) - IBM has retired Deep Blue, the supercomputer that made

Read my last post, but here goes again:

Have *you* ever published anything in a scientific journal? of course
not. Do you have *any* idea of the delay between submission and
publication of an article? of course not. here's a clue: it typically
takes one year from submission to publication for a good journal that is
not backlogged. it can take two years for a top journal that has a big
backlog of submissions waiting for publication. So grow up. I wouldn't
write this up and submit it to the Journal of DC Comics for publication
either.

: True is that they seem to travel around to get some money and public

: attention. Like Muhammed Ali. But he's at least on the welfare tour for
: the handicapped.

Just keep those insults coming. If nothing else, it exposes your
supreme ignorance to the bright light of this newsgroup...

: But this seems to be the american way. Kasparov had to know that...

Ah, now we are on to "the American way". Fortunately I'm wise enough
to know that your way is not the "German Way." I can paint with a
much narrower brush than that, and only hit the people that fit the
description.

: >The New York match and a contest in Philadelphia in 1996 was part chess

[Robert Hyatt]

Chris Carson (chris-car...@ti.com) wrote:
: Bob,

: I checked my numbers from the Rebel web page, I had the numbers for a
: Pentium Pro 200 (not Pentium 90). This looks more valid now.

: Rebel homepage: http://www.xs4all.nl/~rebchess/bench.htm

: Rebel 8.0 Pentium 90 = 20K NPS
: Rebel 8.0 Pentium Pro 200 = 61K NPS

: Deep Blue JR: http://www.chess.ibm.com/press/html/g.6.4.html

: 1/10 speed Single Chess Processor Deep Blue Jr (DBjr) = 75K NPS

: At least from my point of view, we are in sync and the
: reports by HSU look reliable.

: I would add that Rebel is a great program, this just
: continues to show that Deep Blue has some remarkable
: accomplishments (in my opinion).

I followed this up last night after running Rebel on my P6/200 at
home... and I got 80K average. The lowest blip I saw was 60K, the
highest was 90K, the average was 80K, in a position from the hiarcs
hergott match that had lots of material on the board...

[Rolf Tueschen]

hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:

Yeah, I already answered there too. Bob, you lost your control. But
without getting a clue of.

I attacked you, and it was and is justified for bad scientifical
behavior. Note I didn't write about you as scientist or human being.
Perhaps you should also concentrate on the same standards.
Attack my postings, but dont lose your style and "face" by joining the
primitive "ad-hominem" fraction. I hope you got it?

What you wrote here below is so primitive, you know. And it's typical
for the elitarian bullshit you and others supported the last days here
in rgcc while discussing a new rgcp.

But I remarked it on your side time ago. When you had a slight mutual
masturbative chuckle with collegue impostor Ingo...
Chuckle about the idiots non-experts...
On the other hand you were never too tired to defend IA when I put my
**questions** about the exact description of his impostor-like n-brain
promotions. Already there you lost a lot of value as scientist, I tell
you.

You seem to miss that censorship and also selection of welcomed
participants by their personal level of submissiveness...

Disgustful.

But let's not build a second front. The scandal of the DB team is topic
number one. And I feel for you and am even jealous that I was never in
their nearness... But I took you for a scientist. And, sorry Bob, the
standards are universal and not dependent of personal relationships.
More so, they are stable in their definitions. They cant be "simply"
bent.

And then this. You made a primitive cheat. Note, I dont call you a
cheater. But you made a cheat.

I nowhere claimed that the guys had to lift off a rocket to build up a
slovenly report in the American Scientist....

Dont fool around, Bob.

What we wanted were the results and games of the pre-match skittles. Got
it? And I repeated it since end of May, that Benjamin, admitted, not a
scientist, had promissed to give them after the event. Also those of
Ilesscas.

Bob, what are you talking about scientifical invoices for the noble
prize commetee?? Your style is becoming more and more primitive.

And then the output of DB.

Bob, we want the data. Not how Hsu digested the weather turbulences of
last May. Moon ecclipses. Tooth stickers. We want the output.
And I'm sure you or others were ready to write a little program to be
able to *read* the allegedly too complicated computer language of the
Blue. We dont want a final analysis of all that in two years, Bob.

You know why?

Because it's not so interesting. Sorry, if I miss something, but Hsu
could not be the authority to talk about chess matters, no?

What we need is the output. And then there're enough here around who
could think about the interpretation of the data. First of all we all
want to see hgame 1, 2, and 6. Then for the gourmets the others too...

Bob, even myself with v e r y limited computer knowledge and the
same in chess, I did observe interesting stuff in the first mentioned
games. Moves with the need to know about the variations, depth and
values. Because it's so astonishing sometimes. I mean the difference
between a phyntastic Be4 and then some bull. And what I personly want to
know if the output makes sense to all these moves...

============

Let me come back to *your* bad scientifical behavior. Till now I didn't
read the reasons why you didn't search for detailed conditions of the
Junior matches. Again, you're not the average guy who receives data and
post it just so. But you had to ask for the conditions of their playing.
nd where were the games?? I thought a computer had a memory to keep the
notations??

I admit that you didn't cheat the data you gave us. Dont mix that up.
But you are guilty as scientist to having bought naively half cooked
data without further details... And all indication leads us to the
assumption that you played a little game with Hsu, who, as you told us,
lay in the background ROTFLying. Tell him, that I'm ready to tell him
some words about it at our next meeting... :)

============

Finally if you dont excuse for your primitive outing of unknown
ad-hominem tendencies in your self, I promiss I'll do *my* research to
learn more about your rather late carreer at the university of
Brimingham right in the middle of Alabama. Duh. Watch out. :)

We'll see if you're really a natural-born or more an adopted prof...
In Germany we have a tradition of famous cooks and other honorable
people who suddenly become Prof without all the embarrassing year-long
studies...

And then this. Do you really think that a scientist necessarily only is
a scientist if he worked in a university and published minimum 50
articles? And nobody from the outside whether scientist or not is
forbidden to criticise a member of a university? LOL

It's true, scientifical journals are very important. Mostly the articles
are a preliminary stage of a later publication. But that's it, Bob.

We have no longer the 19th century.

Let's come directly to the internet. Today dood ideas also could be
published by non-scientists. Ideas then could be tested by specialists
with the known scientifical standards. No reason to look in despair when
a postings doesn't come from a non-academic. Why throwing away possibly
useful data?

I'm convinced that a lot of good data could be provided by non-academics
in our special field of cc. Dont you think so? (Rhetorical question, I
know you in the meantime. But why then all the hustle?)

================

In awaiting a sound apology I want to repeat what you did wrong.

I criticised you because you forgot of the duties a scientist always has
regarding the data he gets from wherever.

You asked me (rhetorically) if I did ever publish in a famous
scientifical journal...

And this is unallowed ad-hominem, at least following the accepted
world-wide standards of science. You wanted to construct the impression
that I couldn't be sound with my critic against your bad behavior
because I "simply" was not playing in your league.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
hey Bob my comments are not yet finished... please read below...

>Have *you* ever published anything in a scientific journal? of course
>not. Do you have *any* idea of the delay between submission and
>publication of an article? of course not. here's a clue: it typically
>takes one year from submission to publication for a good journal that is
>not backlogged. it can take two years for a top journal that has a big
>backlog of submissions waiting for publication. So grow up. I wouldn't
>write this up and submit it to the Journal of DC Comics for publication
>either.

>: True is that they seem to travel around to get some money and public
>: attention. Like Muhammed Ali. But he's at least on the welfare tour for
>: the handicapped.

>Just keep those insults coming. If nothing else, it exposes your
>supreme ignorance to the bright light of this newsgroup...

Will you please be so friendly to explain why these words of mine could
be understood as insults? Thanks.

Note that you are wrong in assuming that I wanted to play in the new cc
experts elitarian league. I do what I can, but I'm not like an impostor.
My education is sufficient to judge about what's happening. At least on
the base it is written here in the group.

Will you please explain to the rest of the group where I showed my
ignorance? Please dont bother to take the strongest example... :)

>: But this seems to be the american way. Kasparov had to know that...

>Ah, now we are on to "the American way". Fortunately I'm wise enough
>to know that your way is not the "German Way." I can paint with a
>much narrower brush than that, and only hit the people that fit the
>description.

No, you misunderstood my words. The event took lace in USA, no? The DB
team plus sponsor were US-american, no?

I wrote what I claimed since end of May, that also from a sportive view
the event and its outcome was not satisfactory if one looked a bit over
the internal US-american spectrum. I mean we all know about the typical
narrow-minded US-american sports philosophy... and not just in that
topic.

Example: the winner of the american football internal championship is
directly World Champion too...

Same in baseball I assume.

This is absolutely unknown in the rest of the world outside the USA.

Then the tradition that the winner takes it all independent of how he
achieved to win. And that is the point for our event in May.

The result was treated as the winning of the Wch.... But in the rest of
the world especially among chess experts the result was more so regarded
as a bad joke by a badly prepared Kasparov, but not at all as anything
reliable for the qestion man or machine.

And here I know it you also agreed, Bob!

So with my quote you "simply" can't bash me down. It was not meant as an
anti american spam, Czub oftenly demonstrated.

Bob, WHY do you lose your style and your control in such a primitive
way?

We're friends, no? :))

You Batman Godfather and me the young "Robin"

[Chris Carson]

Bob,

Your numbers look to be right on target.

For what it is worth, I think any differences between Campbell
and HSU (or the IBM website and the Reuters report) would
be within expected results given the average speeds
we have been talking about (some positions may provide
more NPS, some less). I see no issues here.

Robert Hyatt wrote:
>
[snip]

>
> I followed this up last night after running Rebel on my P6/200 at
> home... and I got 80K average. The lowest blip I saw was 60K, the
> highest was 90K, the average was 80K, in a position from the hiarcs
> hergott match that had lots of material on the board...

--

[Robert Hyatt]

Rolf Tueschen (TUESCHEN.MEDIZ...@t-online.de) wrote:
: hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:

: > technical journal? (of course not) Do you have *any* idea of
: > the lead time from writing an article until it is published in

: > such a journal? (of course not) Here is a clue: the typical
: > lag from submission to publication is over one year. One year
: > is considered very quick turnaround, while many journals operate
: > on a 2-year delay. So back off and give them time."

: Just for the record, Bob.

: You just did something which is among scientists at least *very* frowned
: upon. You simply lost control and usual style. But apparently without
: knowing....

: For further details of this scandal read my reply to your other post.

There is no scandal. There is no misunderstanding. It turns out the
games were played on two machines, a P5/90 which one of the DB guys
has at home, and a P6/180 which they had at work. The hardware speed
for DB was exactly as I had quoted it from Hsu's quote. There is nothing
to misunderstand, except your continual use of the world "scandal"..

Maybe you are confused and think "scandal" is a type of footwear?

[Robert Hyatt]

Rolf Tueschen (TUESCHEN.MEDIZ...@t-online.de) wrote:

: Yeah, I already answered there too. Bob, you lost your control. But

: without getting a clue of.

: I attacked you, and it was and is justified for bad scientifical
: behavior. Note I didn't write about you as scientist or human being.
: Perhaps you should also concentrate on the same standards.
: Attack my postings, but dont lose your style and "face" by joining the
: primitive "ad-hominem" fraction. I hope you got it?

: What you wrote here below is so primitive, you know. And it's typical
: for the elitarian bullshit you and others supported the last days here
: in rgcc while discussing a new rgcp.

: But I remarked it on your side time ago. When you had a slight mutual
: masturbative chuckle with collegue impostor Ingo...
: Chuckle about the idiots non-experts...
: On the other hand you were never too tired to defend IA when I put my
: **questions** about the exact description of his impostor-like n-brain
: promotions. Already there you lost a lot of value as scientist, I tell
: you.

I don't defend Althofer, because he doesn't need defending. What he
does is interesting and useful, because it shows that a computer *can*
produce good analysis, but that it doesn't always find the best move.

If you don't like his experiments, don't pay any attention to 'em. But
you ought to leave your hatchet at home and not attack everything you
don't like...

: You seem to miss that censorship and also selection of welcomed

: participants by their personal level of submissiveness...

: Disgustful.

: But let's not build a second front. The scandal of the DB team is topic
: number one. And I feel for you and am even jealous that I was never in
: their nearness... But I took you for a scientist. And, sorry Bob, the
: standards are universal and not dependent of personal relationships.
: More so, they are stable in their definitions. They cant be "simply"
: bent.

: And then this. You made a primitive cheat. Note, I dont call you a
: cheater. But you made a cheat.

: I nowhere claimed that the guys had to lift off a rocket to build up a
: slovenly report in the American Scientist....

"cheat" is unambiguous. I either am or I am not. So far as I know, I
am not.

: Dont fool around, Bob.

: What we wanted were the results and games of the pre-match skittles. Got
: it? And I repeated it since end of May, that Benjamin, admitted, not a
: scientist, had promissed to give them after the event. Also those of
: Ilesscas.

They were given to Kasparov.

: Bob, what are you talking about scientifical invoices for the noble

: prize commetee?? Your style is becoming more and more primitive.

: And then the output of DB.

: Bob, we want the data. Not how Hsu digested the weather turbulences of
: last May. Moon ecclipses. Tooth stickers. We want the output.
: And I'm sure you or others were ready to write a little program to be
: able to *read* the allegedly too complicated computer language of the
: Blue. We dont want a final analysis of all that in two years, Bob.

: You know why?

: Because it's not so interesting. Sorry, if I miss something, but Hsu
: could not be the authority to talk about chess matters, no?

you produce way too much output already. We really don't need more.

: What we need is the output. And then there're enough here around who

: could think about the interpretation of the data. First of all we all
: want to see hgame 1, 2, and 6. Then for the gourmets the others too...

: Bob, even myself with v e r y limited computer knowledge and the
: same in chess, I did observe interesting stuff in the first mentioned
: games. Moves with the need to know about the variations, depth and
: values. Because it's so astonishing sometimes. I mean the difference
: between a phyntastic Be4 and then some bull. And what I personly want to
: know if the output makes sense to all these moves...

Ah, yes... back to the "no computer could play that..." sort of bullshit I
see. I notice that Kasparov seems satisfied now. He is not continually
raising that issue any more. Ferret found that move at ply=20. Crafty found
it at ply=20... but of course, it was still cheating, you think?

: ============

: Let me come back to *your* bad scientifical behavior. Till now I didn't
: read the reasons why you didn't search for detailed conditions of the
: Junior matches. Again, you're not the average guy who receives data and
: post it just so. But you had to ask for the conditions of their playing.
: nd where were the games?? I thought a computer had a memory to keep the
: notations??

: I admit that you didn't cheat the data you gave us. Dont mix that up.
: But you are guilty as scientist to having bought naively half cooked
: data without further details... And all indication leads us to the
: assumption that you played a little game with Hsu, who, as you told us,
: lay in the background ROTFLying. Tell him, that I'm ready to tell him
: some words about it at our next meeting... :)

If you follow, the data has *not* changed. It is exactly as I said it was,
with nothing added, nothing omitted....

: ============

: Finally if you dont excuse for your primitive outing of unknown
: ad-hominem tendencies in your self, I promiss I'll do *my* research to
: learn more about your rather late carreer at the university of
: Brimingham right in the middle of Alabama. Duh. Watch out. :)

suit yourself...

: We'll see if you're really a natural-born or more an adopted prof...

: In Germany we have a tradition of famous cooks and other honorable
: people who suddenly become Prof without all the embarrassing year-long
: studies...

no doubt...

: And then this. Do you really think that a scientist necessarily only is

: a scientist if he worked in a university and published minimum 50
: articles? And nobody from the outside whether scientist or not is
: forbidden to criticise a member of a university? LOL

A scientist is one that does "research." That's the definition. Where it
is done is unimportant. That it is done is critical.

: It's true, scientifical journals are very important. Mostly the articles

: are a preliminary stage of a later publication. But that's it, Bob.

: We have no longer the 19th century.

: Let's come directly to the internet. Today dood ideas also could be
: published by non-scientists. Ideas then could be tested by specialists
: with the known scientifical standards. No reason to look in despair when
: a postings doesn't come from a non-academic. Why throwing away possibly
: useful data?

Because *most* articles about current research are *still* published in
paper Journals, because they must be refereed and critiqued before they
are published. As a result, the quality is high. The quality of stuff
on the internet is hopeless, because *anyone* can write something and
stick it on the internet, fact, fiction, or somewhere in between. But
you don't get that, right?

: I'm convinced that a lot of good data could be provided by non-academics

: in our special field of cc. Dont you think so? (Rhetorical question, I
: know you in the meantime. But why then all the hustle?)

I believe I've said this many times. But don't start drawing your sword
until you know *how* something gets published. Doesn't matter if *you*
want it on the internet, and want it there *now*. It matters where *they*
want to publish and *when*. It is *their* story to tell, not yours to
demand.

: ================

: In awaiting a sound apology I want to repeat what you did wrong.

: I criticised you because you forgot of the duties a scientist always has
: regarding the data he gets from wherever.

: You asked me (rhetorically) if I did ever publish in a famous
: scientifical journal...

: And this is unallowed ad-hominem, at least following the accepted
: world-wide standards of science. You wanted to construct the impression
: that I couldn't be sound with my critic against your bad behavior
: because I "simply" was not playing in your league.

No it is not "unallowed." It points out that once again, you have no idea
of what you are talking about. Even if they had prepared an article before
the match started, it would be very unlikely it would have been printed yet.
Since you don't understand that important detail, you start swinging the
sword around in circles, cutting whatever gets within your small radius R.

: hey Bob my comments are not yet finished... please read below...

: >Have *you* ever published anything in a scientific journal? of course
: >not. Do you have *any* idea of the delay between submission and
: >publication of an article? of course not. here's a clue: it typically
: >takes one year from submission to publication for a good journal that is
: >not backlogged. it can take two years for a top journal that has a big
: >backlog of submissions waiting for publication. So grow up. I wouldn't
: >write this up and submit it to the Journal of DC Comics for publication
: >either.

: >: True is that they seem to travel around to get some money and public
: >: attention. Like Muhammed Ali. But he's at least on the welfare tour for
: >: the handicapped.

: >Just keep those insults coming. If nothing else, it exposes your
: >supreme ignorance to the bright light of this newsgroup...

: Will you please be so friendly to explain why these words of mine could
: be understood as insults? Thanks.

read your statement... "they seem to travel to get money and attention."
That is *not* why they do so. That is *not* why I accept invitations to
give talks either. It is never profitable, and takes a lot of time. So
your "Ali" metaphor is just more of a substance that smells bad...

: Note that you are wrong in assuming that I wanted to play in the new cc

: experts elitarian league. I do what I can, but I'm not like an impostor.
: My education is sufficient to judge about what's happening. At least on
: the base it is written here in the group.

apparently not, but that's my opinion...

: Will you please explain to the rest of the group where I showed my

: ignorance? Please dont bother to take the strongest example... :)

Generally when you type "f" in the newsreader, to start a follow-up
post. But a good example is the next line, your "American way"... BTW,
America is spelled with a capital A, in case you didn't know, just like
Germany, England, etc...

: >: But this seems to be the american way. Kasparov had to know that...

: >Ah, now we are on to "the American way". Fortunately I'm wise enough
: >to know that your way is not the "German Way." I can paint with a
: >much narrower brush than that, and only hit the people that fit the
: >description.

: No, you misunderstood my words. The event took lace in USA, no? The DB
: team plus sponsor were US-american, no?

and this would mean exactly what?

: I wrote what I claimed since end of May, that also from a sportive view

: the event and its outcome was not satisfactory if one looked a bit over
: the internal US-american spectrum. I mean we all know about the typical
: narrow-minded US-american sports philosophy... and not just in that
: topic.

: Example: the winner of the american football internal championship is
: directly World Champion too...

: Same in baseball I assume.

: This is absolutely unknown in the rest of the world outside the USA.

: Then the tradition that the winner takes it all independent of how he
: achieved to win. And that is the point for our event in May.

: The result was treated as the winning of the Wch.... But in the rest of
: the world especially among chess experts the result was more so regarded
: as a bad joke by a badly prepared Kasparov, but not at all as anything
: reliable for the qestion man or machine.

: And here I know it you also agreed, Bob!

Nope. World Championship was never at stake. Just a lot of pride by
Kasparov. Which he lost of course...

: So with my quote you "simply" can't bash me down. It was not meant as an

: anti american spam, Czub oftenly demonstrated.

: Bob, WHY do you lose your style and your control in such a primitive
: way?

good question. I didn't lose my style or control. I simply wrote
exactly what I meant..

[Rolf Tueschen]

hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:

> technical journal? (of course not) Do you have *any* idea of

[Rolf Tueschen]

hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:

>Rolf Tueschen (TUESCHEN.MEDIZ...@t-online.de) wrote:

>: Yeah, I already answered there too. Bob, you lost your control. But
>: without getting a clue of.

>: I attacked you, and it was and is justified for bad scientifical
>: behavior. Note I didn't write about you as scientist or human being.
>: Perhaps you should also concentrate on the same standards.
>: Attack my postings, but dont lose your style and "face" by joining the
>: primitive "ad-hominem" fraction. I hope you got it?

>: What you wrote here below is so primitive, you know. And it's typical
>: for the elitarian bullshit you and others supported the last days here
>: in rgcc while discussing a new rgcp.

>: But I remarked it on your side time ago. When you had a slight mutual
>: masturbative chuckle with collegue impostor Ingo...
>: Chuckle about the idiots non-experts...
>: On the other hand you were never too tired to defend IA when I put my
>: **questions** about the exact description of his impostor-like n-brain
>: promotions. Already there you lost a lot of value as scientist, I tell
>: you.

>I don't defend Althofer, because he doesn't need defending.

Thanks that you again came back to this one. I'm astonished. Because he
"simply" :) forgot to mention the exact experimental conditions. Period.
You then came up with the journals and other sources. My answer was, if
he does present his data here in rgcc for feedback he has the damn duty
to provide us with enough info about the experimental design. Period.
When you took his side following my questions you acted not as
scientist.

>What he
>does is interesting and useful, because it shows that a computer *can*
>produce good analysis, but that it doesn't always find the best move.

Ah, I see, and it needs IA to demonstrate this? I dont get it. I do the
same for about 15 years now. :)

>If you don't like his experiments, don't pay any attention to 'em.

Yeah, papa, but what shall I do if I like them...?
Cant you understand that critics are not a bad thing basically?
And critics dont always come out of hate...?

Man. we've enough brown-nosing mice, that the world could stand one more
critic.

>But
>you ought to leave your hatchet at home and not attack everything you
>don't like...

It's so sad to see you arguing in vain. Because the opposite again is
true. Someone I dont like I dont even talk to... :)
Or related to topics, you behave BTW like the SSDF teamsters on my
critical questions, I try to give my view on the stuff and I am prepared
to be also informed that my view was damned wrong. Well, that's better
in my view than always daydreaming in the same shit. No? :)

>: You seem to miss that censorship and also selection of welcomed
>: participants by their personal level of submissiveness...

>: Disgustful.

>: But let's not build a second front. The scandal of the DB team is topic
>: number one. And I feel for you and am even jealous that I was never in
>: their nearness... But I took you for a scientist. And, sorry Bob, the
>: standards are universal and not dependent of personal relationships.
>: More so, they are stable in their definitions. They cant be "simply"
>: bent.

>: And then this. You made a primitive cheat. Note, I dont call you a
>: cheater. But you made a cheat.

>: I nowhere claimed that the guys had to lift off a rocket to build up a
>: slovenly report in the American Scientist....

>"cheat" is unambiguous. I either am or I am not. So far as I know, I
>am not.

You *are* not, if you want this stabilization, but you did it. Slightly
different thing. Perhaps not for you. Bob, that's the difference, why I
took your confusion about me and mclane as almost insult. He always
writes about personal characteristics. "You are mad or ill." But I do
concentrate on what is visible to the whole world thanks to Dejas.

You cheated because you insinuated that I ever had wished them to
publish a quickie report just to satisfy my curiosity. But I said that I
wanted the data that was promissed during the show by Benjamin. 60 games
would be not such a great effort without annotations??

And the output of the DB moves is also a trivial thing if we would
translate it into readable chess notation. Period.

But not only this. A cheat is it if you see my reminders open eyed but
you always shift/twist your attention to the more deeper analysis which
"simply" took two years. (BTW I just mention here for the sake of the
argument and Dejas history, that I judge absolutely negative about this
tool of scientifical journal if it's misused for political purposes.
Here it's simply the excuse for holding back data. Because nobody but
you claimed to be eager to be informed about deeper insight of the team
about their victory... We all want "simply" the data. We have then our
own theories, Bob. No need to comment on this.)

>: Dont fool around, Bob.

In this meaning I just mentioned. Sorry.

>: What we wanted were the results and games of the pre-match skittles. Got
>: it? And I repeated it since end of May, that Benjamin, admitted, not a
>: scientist, had promissed to give them after the event. Also those of
>: Ilesscas.

>They were given to Kasparov.

What was given to him? All the games or just a little bit of the output
for the second game? Iask because I never heard of it.

>: Bob, what are you talking about scientifical invoices for the noble
>: prize commetee?? Your style is becoming more and more primitive.

>: And then the output of DB.

>: Bob, we want the data. Not how Hsu digested the weather turbulences of
>: last May. Moon ecclipses. Tooth stickers. We want the output.
>: And I'm sure you or others were ready to write a little program to be
>: able to *read* the allegedly too complicated computer language of the
>: Blue. We dont want a final analysis of all that in two years, Bob.

>: You know why?

>: Because it's not so interesting. Sorry, if I miss something, but Hsu
>: could not be the authority to talk about chess matters, no?

>you produce way too much output already. We really don't need more.

Sorry so much, but I thought I gave you an interesting challenge. You
dont like it anymore?

>: What we need is the output. And then there're enough here around who
>: could think about the interpretation of the data. First of all we all
>: want to see hgame 1, 2, and 6. Then for the gourmets the others too...

>: Bob, even myself with v e r y limited computer knowledge and the
>: same in chess, I did observe interesting stuff in the first mentioned
>: games. Moves with the need to know about the variations, depth and
>: values. Because it's so astonishing sometimes. I mean the difference
>: between a phyntastic Be4 and then some bull. And what I personly want to
>: know if the output makes sense to all these moves...

>Ah, yes... back to the "no computer could play that..." sort of bullshit I
>see. I notice that Kasparov seems satisfied now.

Ah, you wrote above that he *got* some data. So, perhaps it's because of
this?

Oh, no Bob, you again fell into the boiling pot. But I swear, *I* didn't
even *see* how you did that again. :)

>He is not continually
>raising that issue any more. Ferret found that move at ply=20. Crafty found
>it at ply=20... but of course, it was still cheating, you think?

Sorry, I apologize for having expected too much. The Be4 was my symbolic
picture to make my point. I do remember very well that you posted the
results for Crafty. But for you it must not be so difficult to
understand what I meant. You do nothing else the whole nights long when
you attend to the sessions on the net over there. You think about this
or that seeing some output... No? Bob?

>: ============

>: Let me come back to *your* bad scientifical behavior. Till now I didn't
>: read the reasons why you didn't search for detailed conditions of the
>: Junior matches. Again, you're not the average guy who receives data and
>: post it just so. But you had to ask for the conditions of their playing.
>: nd where were the games?? I thought a computer had a memory to keep the
>: notations??

>: I admit that you didn't cheat the data you gave us. Dont mix that up.
>: But you are guilty as scientist to having bought naively half cooked
>: data without further details... And all indication leads us to the
>: assumption that you played a little game with Hsu, who, as you told us,
>: lay in the background ROTFLying. Tell him, that I'm ready to tell him
>: some words about it at our next meeting... :)

>If you follow, the data has *not* changed. It is exactly as I said it was,
>with nothing added, nothing omitted....

And I already made clear that I didn't think for a second, Bob, that you
would intentiously cheat on the data. On the contrary I defended you and
rejected publical applause when I demonstrated a little oversight in
your memory about some tournamernts...

That's not the point. You're wheeling and dealing to come out of this
trap. But it's not possible. It's crystal clear that you forgot to
secure the needed data for a serious judgement about these 10-0 results.
Let's not repeat. You failed to give an explanation. And I know you good
enough that you had given it if you had found one.

You did as if the experimental conditions for the little games between
Junior and the two micros were absolutely fair. Appropiate time control
and enough reduced power of Junior/DB. But that's not the case as far
as I could see. Period.

Now you argue. but I didn't know at the time being. And I say. No, Bob,
that not an excuse, because you had to research your data good enough
yourself. The way you presented it you acted irresponsible. And note I
never attacked Hsu for this because it was you who wanted to publish it.

>: ============

>: Finally if you dont excuse for your primitive outing of unknown
>: ad-hominem tendencies in your self, I promiss I'll do *my* research to
>: learn more about your rather late carreer at the university of
>: Brimingham right in the middle of Alabama. Duh. Watch out. :)

>suit yourself...

Keep cool, it was a joke Bob. :)

>: We'll see if you're really a natural-born or more an adopted prof...
>: In Germany we have a tradition of famous cooks and other honorable
>: people who suddenly become Prof without all the embarrassing year-long
>: studies...

>no doubt...

Never heard of? Certainly it lacks then of the doctoral degree. To
mention just a tiny misunderstanding of them cooks and butchers. :)

>: And then this. Do you really think that a scientist necessarily only is
>: a scientist if he worked in a university and published minimum 50
>: articles? And nobody from the outside whether scientist or not is
>: forbidden to criticise a member of a university? LOL

>A scientist is one that does "research." That's the definition. Where it
>is done is unimportant. That it is done is critical.

Thanks. And you would surely agree that the interne/usenet is a good
place to collect interesting ideas, no? C'mon Bob. Dont destroy my
fundaments.

>: It's true, scientifical journals are very important. Mostly the articles
>: are a preliminary stage of a later publication. But that's it, Bob.

>: We have no longer the 19th century.

>: Let's come directly to the internet. Today dood ideas also could be
>: published by non-scientists. Ideas then could be tested by specialists
>: with the known scientifical standards. No reason to look in despair when
>: a postings doesn't come from a non-academic. Why throwing away possibly
>: useful data?

>Because *most* articles about current research are *still* published in
>paper Journals, because they must be refereed and critiqued before they
>are published. As a result, the quality is high. The quality of stuff
>on the internet is hopeless, because *anyone* can write something and
>stick it on the internet, fact, fiction, or somewhere in between. But
>you don't get that, right?

You seemed to have missed some of the latest scandals. Because it's my
determined opinion that the quality is absolutely NOT assured. But
beware me, I dont say, let's forget about these publications.

But here on he net we have a so called if it's allowed, second reality.
And you're for me -- I repeat myself -- the best example, that a good
academic is able to talk with a lot of different people, and of
different educational level. And that you do this is the reason we all
adore you. Therefore you should not always back your collegues if we
once and again attack him and he denies any communication. I always come
back to the big Scientist from Jena.

>: I'm convinced that a lot of good data could be provided by non-academics
>: in our special field of cc. Dont you think so? (Rhetorical question, I
>: know you in the meantime. But why then all the hustle?)

>I believe I've said this many times. But don't start drawing your sword
>until you know *how* something gets published. Doesn't matter if *you*
>want it on the internet, and want it there *now*. It matters where *they*
>want to publish and *when*. It is *their* story to tell, not yours to
>demand.

Wait, it's surely my story to wish and demand. And also to criticise as
far as I'm able. You wont really doubt that?

>: ================

>: In awaiting a sound apology I want to repeat what you did wrong.

>: I criticised you because you forgot of the duties a scientist always has
>: regarding the data he gets from wherever.

>: You asked me (rhetorically) if I did ever publish in a famous
>: scientifical journal...

>: And this is unallowed ad-hominem, at least following the accepted
>: world-wide standards of science. You wanted to construct the impression
>: that I couldn't be sound with my critic against your bad behavior
>: because I "simply" was not playing in your league.

>No it is not "unallowed." It points out that once again, you have no idea
>of what you are talking about.

I'm sure I know about. But I'm also realist and must respect if you dont
get it. Therefore I always claimed not guilty due to unintentionls... :)

> Even if they had prepared an article before
>the match started, it would be very unlikely it would have been printed yet.
>Since you don't understand that important detail, you start swinging the
>sword around in circles, cutting whatever gets within your small radius R.

You really let us forget the prose of mclane/Czub. But therefore you're
not right. I already explained. I didn't expect an scientifical article
but the data. Rough and fresh like we get it from your crafty output,
Bob.
(BTW again for the sake of truth. I knew instantely at the press
conference, when Tan (?) made this statement about the "appropiateness"
that he made a fishy statement. Well, either you have it in your guts or
you dont, Bob. I was right, because this was an excuse. Kasparov, you
said it, has his data since long... But we should have to wait till the
"never". Kasparov cant be interested in further ramblings because he
wanted a new chance. But Bob, believe me, I dont have to make allowances
for my critics.)

>: hey Bob my comments are not yet finished... please read below...

>: >Have *you* ever published anything in a scientific journal? of course
>: >not. Do you have *any* idea of the delay between submission and
>: >publication of an article? of course not. here's a clue: it typically
>: >takes one year from submission to publication for a good journal that is
>: >not backlogged. it can take two years for a top journal that has a big
>: >backlog of submissions waiting for publication. So grow up. I wouldn't
>: >write this up and submit it to the Journal of DC Comics for publication
>: >either.

>: >: True is that they seem to travel around to get some money and public
>: >: attention. Like Muhammed Ali. But he's at least on the welfare tour for
>: >: the handicapped.

>: >Just keep those insults coming. If nothing else, it exposes your
>: >supreme ignorance to the bright light of this newsgroup...

>: Will you please be so friendly to explain why these words of mine could
>: be understood as insults? Thanks.

>read your statement... "they seem to travel to get money and attention."
>That is *not* why they do so. That is *not* why I accept invitations to
>give talks either. It is never profitable, and takes a lot of time. So
>your "Ali" metaphor is just more of a substance that smells bad...

I hold it upright.

I meant that Ali does all his public work for a charity reason. Could
you give me the "raison d'etre" for Murray? Instead he could provide
this community with data, no???????????????????????

And then, you're a teacher with a strict schedule. For you it's perhaps
annoying to see the waste of time. But you're a normal human being,

But if you once smelled the tasty seductive sides of being an actor on
the scene travelling from state to state. Well, it *can* become an end
in itself.

For me, I repeat this, it's fishy, to let this group starve for
information and travelling around and letting here and there someone
pick up some snips he then posts to the group. Why not doing this:

I have this idea. You as the doyen of the group, you keep the mic close
to your lips. And you transmit some of our highlevelled or just plain
mad speculations. Comes my stuff into closer consideration I fear. :)

And then they answer to you. And you again post it here in decent style.

Bob, I'm sure that this could be a tremendous upgrading for rgcc.

No insults, no Rolfing, only questions, questions, questions.

AND answers. And also different levelled answers. Here, sorry, we cant
answer because it would be bad for a piossible rematch. And so on. But
then we know about what we talk.

Because, Bob, my elder brother, do you really think that I'm happy to
see you and all of us digging in fog?????

No, it's the scandal of their side. You can only post what you got.

But please think about this. AND FOR ALL IT COULD PROVE THAT I'M
BASICALLY INTERESTED IN INFORMATION AND NOT TROUBLE.

>: Note that you are wrong in assuming that I wanted to play in the new cc
>: experts elitarian league. I do what I can, but I'm not like an impostor.
>: My education is sufficient to judge about what's happening. At least on
>: the base it is written here in the group.

>apparently not, but that's my opinion...

It would be boring if all had the same...

>: Will you please explain to the rest of the group where I showed my
>: ignorance? Please dont bother to take the strongest example... :)

>Generally when you type "f" in the newsreader, to start a follow-up
>post.

Good to know tht you're normally a friendly guy. :)

>But a good example is the next line, your "American way"... BTW,
>America is spelled with a capital A, in case you didn't know,

No, I didn't know. And let me explain why. It's ridiculous, Bob. But in
English you write all words with minor letters. But suddenly the
adjectives have to be written with a capital letter??

>just like
>Germany, England, etc...

Ah, you now claim expert status for linguistical questions? :)

But is it allowed to lead your attention to the simple truth that I
talked about american as an adjective. And you can be sure that I write
also german sugar and not German sugar. Please notice that.

I write Germany and America. Substantive then.

Well, my English is not good enough. But I had never thought that I
could make you upset like the famous Ed? I thought you were much more
calm.

>: >: But this seems to be the american way. Kasparov had to know that...

>: >Ah, now we are on to "the American way". Fortunately I'm wise enough
>: >to know that your way is not the "German Way." I can paint with a
>: >much narrower brush than that, and only hit the people that fit the
>: >description.

>: No, you misunderstood my words. The event took lace in USA, no? The DB
>: team plus sponsor were US-american, no?

>and this would mean exactly what?

You know exactly what I meant. And I'm sure I had given fut´rther
information about how I meant this. Yes, if you want, I claim, that in
sports US-America is following the iron rule. If it suits our interests
(money, fame) then all cheating is allowed. I think especially on the
athletics. Read what Carl Lewis told the press about the situation in
US-America. It's "simply" hapocritical. The "simply" dont test against
doping. Period.

Another aspect is the low public education. I mean the show of
especially one guy on the stage in May was very supportive for my
statement. He really tried to sell chess as an boxing event. :)

All this Kasparov had to take into consideration before the show.
But I know that you agree with me about his own responsibility.
But why do you now reject my talking about this rypical american way?

>: I wrote what I claimed since end of May, that also from a sportive view
>: the event and its outcome was not satisfactory if one looked a bit over
>: the internal US-american spectrum. I mean we all know about the typical
>: narrow-minded US-american sports philosophy... and not just in that
>: topic.

>: Example: the winner of the american football internal championship is
>: directly World Champion too...

>: Same in baseball I assume.

>: This is absolutely unknown in the rest of the world outside the USA.

>: Then the tradition that the winner takes it all independent of how he
>: achieved to win. And that is the point for our event in May.

>: The result was treated as the winning of the Wch.... But in the rest of
>: the world especially among chess experts the result was more so regarded
>: as a bad joke by a badly prepared Kasparov, but not at all as anything
>: reliable for the qestion man or machine.

>: And here I know it you also agreed, Bob!

>Nope. World Championship was never at stake. Just a lot of pride by
>Kasparov. Which he lost of course...

You seem to miss the average logic. We've beaten the champ. So, we are
the best....

>: So with my quote you "simply" can't bash me down. It was not meant as an
>: anti american spam, Czub oftenly demonstrated.

>: Bob, WHY do you lose your style and your control in such a primitive
>: way?

>good question. I didn't lose my style or control. I simply wrote
>exactly what I meant..

Bob, these two things are well seperated. You wrote what you meant. But
you also lost a little bit your control, no? :))

Against me, the defender of Bob's decency against a whole lot of rgcc
troublemakers. No flowers, Bob. It's all in Dejas. That's enough for me.

[Eugene Nalimov]

It looks that there are 16 chess processors hooked to a single RS/6000 CPU
in Deep Blue Jr, and just 8 in Deep Blue. So communications overhead is
much
greater in Deep Blue Jr. That can explan less nps - chess processors just
too
fast (Bob himself explained some time ago neccessity of a balance between
number of general purpose CPU and chess processors) .

Eugene

Robert Hyatt <hy...@crafty.cis.uab.edu> wrote in article
<60dqb5$c...@juniper.cis.uab.edu>...

> Chris Carson (chris-car...@ti.com) wrote:
> : A description of Deep Blue Jr (DB Jr) can be found at:
>
> : http://www.chess.ibm.com/press/html/g.6.4.html
>
> : Deep Blue DB Jr.
> : Host computer 32-node RS/6000
> : RS/6000 SP workstation
> : Chess
> : processors 256 16
>
> : Avg. speed (in chess
> : positions per second) 200 million 10 million
>
> : Computing
> : method parallel serial
>

[Eugene Nalimov]

Is it possible that 2M nps is just a theoretical peak performance?

If I understand correctly, chess processor is idle while RS/6000 CPU
does software search, checks for extensions, probes hash table
(and is it distributed, too?), communicates with other chess processors
(and there are 7 of them), etc.

If so, than 1M effective nps seems reasonable.

Eugene

brucemo <bru...@seanet.com> wrote in article <342AD4...@seanet.com>...

> Robert Hyatt wrote:
>
> > First question, what is the "normal speed"??? We need to know that to
> > resolve the other questions. IE if "normal" is 1M nps, then 1/10th is
> > 100K, which is what I'd claim rebel does on a P6, from my tests at
> > home. If normal is 2M, then 1/10th is 200K, which is double what Rebel
> > does on a P6/200. Either case, however, is *way* over what Rebel would
> > do on a P5/90, which is why I had originally assumed the P6/200 was
> > used. I find the P5/90 numbers suspicious, because that would mean
that
> > the quote from Hsu was not as accurate as I'm used to seeing from him,
> > ie "approximately equal" and "5x-10x faster" are *not* the same...
> > If rebel ran on a P5/90, 200Knps is about 10x faster, which is not very
> > equal at all... and if we take my 1M nps for DB's single chip, it is
> > still 5x faster than the P5/90, which is still not very close...
>
> There is no way that you are going to argue that I mis-heard Campbell, or

> that Campbell is wrong about what happened and Hsu is right, by doing
this
> kind of math.
>

[Harald]

quoting a mail from chris-car...@ti.com

> Rebel on a Pentium 90 would be about 60K NPS based on data from
> Rebel homepage: http://www.xs4all.nl/~rebchess/bench.htm
>
> (20957080 Nodes / 350 sec) = 60K NPS for Rebel 8.0.

Oops, this must be an endgame-position. Mine does on P133 only about 45-
50knps

Harald Faber

[AMIR BAN]

Robert Hyatt wrote:
>

> When they publish details, we'll find out. By the way, for Rolf,
> who is always nagging about why they have published something, here
> is an important question:
>
> "Have *you* ever published something in a significant scientific
> technical journal? (of course not) Do you have *any* idea of
> the lead time from writing an article until it is published in
> such a journal? (of course not) Here is a clue: the typical
> lag from submission to publication is over one year. One year
> is considered very quick turnaround, while many journals operate
> on a 2-year delay. So back off and give them time."

In view of the huge amount of commentary, analysis and whatnot by anyone
reomotely close to this match during the match and after it, even you
Bob m ay admit that this yours is a weak statement. What kind of
preparation and lead time is needed to give some meaningful comments on
what everyone is asking ?

If you couple this with IBM's (especially Mr. Tan's) statements of
say-nothing gibberish, and IBM's very obvious policy of
keep-your-mouth-shut before, and amazingly, after the match, you will
understand that it is NOT the case that some annoyingly slow editors of
some obsocure journals are keeping the story from the world while the
the Deep-Blue team is biting its nails in frustration at the delay.

Please remember we are not waiting to hear details about the SP
architecture, or IBM's marketing hype on molecular biology. We are
waiting to hear something on computer chess and the match.

The world has noticed, and IBM are well aware, that Kasparov has raised
the suspicion of cheating, with the 2nd game in mind. Rather than
discuss and reveal everything they have on this remarkable game, they
chose to deliver a few pages of cryptic dumps marked "IBM confidential"
to Kasparov and his aides eyes only. Since Kasparov asked us to decipher
those, I have seen them (I understand something was also published on
the NY Times. I don't know if everything or only part of it). The
interesting part of it were the printouts of the 36'th and 37'th moves
of game 2 (axb5 and Be4).

Garry's cheating theory was that there was a "go-for-second-best"
button, and that this may have been used at the 36th move. This is quite
possible to implement of course, and may be quite effective. In this
regard Kasparov said something that is very significant I think: "If an
expert is allowed to override the computer just once in any game, I am
already in trouble". This still doesn't mean that this kind of cheating
took place. I didn't find any evidence for this, but I cannot rule it
out in view of what the printouts tell.

The printouts show that Qb6 was the move considered all the time. DB did
well to see Black's best counterplay after Qb6, throwing 2 or 3 pawns to
open lines and attack the white king, as shown by the PV. The evaluation
throughout was around +0.5 pawns for white and dropping slowly. Finally
axb5 was accepted without a PV with a value of 0.4-0.5.

There were several things the needed explanation in the deciphering:

- There was no explanation why the Qb6 score dropped so low with white
ahead by 2 pawns or more. I mean, it's not like you could see the
material regained or a decisive king attack mounted. The only
explanation was that the evaluation function gave the compensation, in
contrast to all PC programs who evaluate white to be more than a pawn up
there. So it seems that DB is not so much the calculating giant as the
positional wizard ?

- No PV for axb5 (but that can happen of course).

- Time management mystery: I think I could make out DB's time management
policy, but could not understand what happened in this move. The normal
time allotment was only about 3 minutes, but a move was not played by
that time, although in other moves, DB would always play a move at the
planned time. There was no obvious reason not to play the move at the
normal time, which was still Qb6. Instead DB went into a procedure it
called "panic-time", where there started a countdown to about 14
minutes. Several minutes into "panic-time" DB switched to axb5. Some
time later, not connected to any event shown in the log, with the
panic-time countdown still with 7 or 8 minutes to go, DB announced
"panic-time" and played a move. DB still had other moves to consider at
that ply level. So what happened ?

In the previous move, the 35th (Bxd6) DB thought for 15 minutes. It
seems cleat that the "panic-time" mechanism was activated there (for
some unknown reason), and the machine thought until "panic-time"
interrupted it. This is significant I think. Was DB malfunctioning at
this point of the game. Could it be that an operator, seeing the
depressing "panic-time" messages appearing for the 2nd move in a row
really panic and push the "Move Now" button ?

There is no reason to rush into any conclusions here, but the DB team
needs to give answers to the these questions:

1. What was the basis for preferring 36. axb5 over 36. Qb6 ?

2. Why was "panic-time" activated in the 36th move and why was a move
played when it finally played ?

Amir

[Robert Hyatt]

Eugene Nalimov (eug...@microsoft.com) wrote:
: Is it possible that 2M nps is just a theoretical peak performance?

: If I understand correctly, chess processor is idle while RS/6000 CPU
: does software search, checks for extensions, probes hash table
: (and is it distributed, too?), communicates with other chess processors
: (and there are 7 of them), etc.

: If so, than 1M effective nps seems reasonable.

:

I think I know the "real" number now. From what I understand, the
real DB chess procssors run at 24mhz, and take ten cycles/node, which
gives a processor thruput of 2.4M nodes per second, normally. For the
"match" we have been discussing, this was throttled way back, into the
100K range.

It's interesting that DB with 256 processors averages about 200M, when it
"could" do nearly 600M. So I am assuming (now) that this 200M nps value
just might be an effective number since they run at about 30% efficiency
with the parallel search. If that is true (where the 200M nodes per sec
comes from) then I am mighty impressed. I'm impressed if they do 200M
*before* factoring in the 30%, but if it is after, simply "wow"... :)

Bob

: Eugene

: brucemo <bru...@seanet.com> wrote in article <342AD4...@seanet.com>...
: > Robert Hyatt wrote:
: >
: > > First question, what is the "normal speed"??? We need to know that to
: > > resolve the other questions. IE if "normal" is 1M nps, then 1/10th is
: > > 100K, which is what I'd claim rebel does on a P6, from my tests at
: > > home. If normal is 2M, then 1/10th is 200K, which is double what Rebel
: > > does on a P6/200. Either case, however, is *way* over what Rebel would
: > > do on a P5/90, which is why I had originally assumed the P6/200 was
: > > used. I find the P5/90 numbers suspicious, because that would mean
: that
: > > the quote from Hsu was not as accurate as I'm used to seeing from him,
: > > ie "approximately equal" and "5x-10x faster" are *not* the same...
: > > If rebel ran on a P5/90, 200Knps is about 10x faster, which is not very
: > > equal at all... and if we take my 1M nps for DB's single chip, it is
: > > still 5x faster than the P5/90, which is still not very close...
: >
: > There is no way that you are going to argue that I mis-heard Campbell, or

: > that Campbell is wrong about what happened and Hsu is right, by doing
: this
: > kind of math.
: >

: > It is obvious who should clarify this, but I don't know if they want to

: >

[Robert Hyatt]

AMIR BAN (ms...@ibm.net) wrote:
: Robert Hyatt wrote:
: >

: > When they publish details, we'll find out. By the way, for Rolf,
: > who is always nagging about why they have published something, here
: > is an important question:
: >
: > "Have *you* ever published something in a significant scientific
: > technical journal? (of course not) Do you have *any* idea of
: > the lead time from writing an article until it is published in
: > such a journal? (of course not) Here is a clue: the typical
: > lag from submission to publication is over one year. One year
: > is considered very quick turnaround, while many journals operate
: > on a 2-year delay. So back off and give them time."

: In view of the huge amount of commentary, analysis and whatnot by anyone
: reomotely close to this match during the match and after it, even you
: Bob m ay admit that this yours is a weak statement. What kind of
: preparation and lead time is needed to give some meaningful comments on
: what everyone is asking ?

I don't agree at all. Murray's talk is but one example. Hsu has been
out of the US, giving talks on the DB hardware. Murray has given talks
around the US as well. It's not as though they are being secretive.
I think we'll see info on what they do.

: If you couple this with IBM's (especially Mr. Tan's) statements of

: say-nothing gibberish, and IBM's very obvious policy of
: keep-your-mouth-shut before, and amazingly, after the match, you will
: understand that it is NOT the case that some annoyingly slow editors of
: some obsocure journals are keeping the story from the world while the
: the Deep-Blue team is biting its nails in frustration at the delay.

Hmmm... while I don't want to start a big argument, how much have you
published in the way of technical details about your program? Gone as
far as I have? It is easy to criticize the DB guys, but don't forget
to do an occasional reality check and compare what they've written
against what most others have written. They have a pretty good record
for publishing about their "system"...

: Please remember we are not waiting to hear details about the SP

: architecture, or IBM's marketing hype on molecular biology. We are
: waiting to hear something on computer chess and the match.

You have to catch Murray or Hsu as Bruce did. There is lots of
info flowing, just not in visible locations like journals, yet. I
suspect we will see something before too much longer..

: The world has noticed, and IBM are well aware, that Kasparov has raised

: the suspicion of cheating, with the 2nd game in mind. Rather than
: discuss and reveal everything they have on this remarkable game, they
: chose to deliver a few pages of cryptic dumps marked "IBM confidential"
: to Kasparov and his aides eyes only. Since Kasparov asked us to decipher
: those, I have seen them (I understand something was also published on
: the NY Times. I don't know if everything or only part of it). The
: interesting part of it were the printouts of the 36'th and 37'th moves
: of game 2 (axb5 and Be4).

: Garry's cheating theory was that there was a "go-for-second-best"
: button, and that this may have been used at the 36th move. This is quite
: possible to implement of course, and may be quite effective. In this
: regard Kasparov said something that is very significant I think: "If an
: expert is allowed to override the computer just once in any game, I am
: already in trouble". This still doesn't mean that this kind of cheating
: took place. I didn't find any evidence for this, but I cannot rule it
: out in view of what the printouts tell.

This is a case of "the perfect crime". *IF* they wanted to cheat, it
would be impossible to prevent. However, I wonder who would want to
override DB against the best chess player in the world? I simply don't
buy that argument, and think it is an insult to them to even suggest
that it is a possibility. Please offer evidence, not "I cannot rule it
out." In the US, the constitution is founded on "innocent until proven
guilty." It is a well-conceived idea that works. We should all give it
a try here...

: The printouts show that Qb6 was the move considered all the time. DB did

: well to see Black's best counterplay after Qb6, throwing 2 or 3 pawns to
: open lines and attack the white king, as shown by the PV. The evaluation
: throughout was around +0.5 pawns for white and dropping slowly. Finally
: axb5 was accepted without a PV with a value of 0.4-0.5.

: There were several things the needed explanation in the deciphering:

: - There was no explanation why the Qb6 score dropped so low with white
: ahead by 2 pawns or more. I mean, it's not like you could see the
: material regained or a decisive king attack mounted. The only
: explanation was that the evaluation function gave the compensation, in
: contrast to all PC programs who evaluate white to be more than a pawn up
: there. So it seems that DB is not so much the calculating giant as the
: positional wizard ?

not at all. Both Crafty *and* ferret found this move, it just takes a long
time. We saw draw by repetition on Qb6 at ply=20. Remember that DB's
hardware probes the hash tables in the chess processors, but that the
chess procesors do not return any sort of path information or anything...
just a score for the depth requested. So it would be nearly impossible to
see what was going on beyond the point where the SP hands something to the
chess processors.

But finding that draw is not impossible.. when you get to Paris, ask Bruce.
He ran it to 20 plies before I did, and found that Qb6 is a draw, and that
Be4 is slightly better than a draw. I confirmed it by letting Crafty search,
although it took an incredible amount of time to do so...

[note... I hope we are at the same point in the game... I don't have the
pgn handy, but remember the qb6 vs Be4 move discussion. If this is the
wrong place or move, excuse the mistake...]

: - No PV for axb5 (but that can happen of course).

If I got concerned every time this happens to Crafty, I'd have thrown it
out long ago. IE deep search to get 0.000 on Qb6, fail high on Be4, but
no time to resolve it. When I got the Be4! at 20 (or 21, Bruce probably
remembers better than I when we were playing with this) I waited a long
time and got nothing back... But it would have played Be4 in any
case...

: - Time management mystery: I think I could make out DB's time management

: policy, but could not understand what happened in this move. The normal
: time allotment was only about 3 minutes, but a move was not played by
: that time, although in other moves, DB would always play a move at the
: planned time. There was no obvious reason not to play the move at the
: normal time, which was still Qb6. Instead DB went into a procedure it
: called "panic-time", where there started a countdown to about 14
: minutes. Several minutes into "panic-time" DB switched to axb5. Some
: time later, not connected to any event shown in the log, with the
: panic-time countdown still with 7 or 8 minutes to go, DB announced
: "panic-time" and played a move. DB still had other moves to consider at
: that ply level. So what happened ?

I've seen that happen with them many time. If you find the JICCA article
they published, there is a cryptic explanation of what triggers them to
go into a deep think... but it seems related to the node counts for each
root move, and when one starts climbing with respect to the best root move
they get suspicious. They've done this against me, in fact. I'd like to
understand it better... but Crafty will also go into extended time mode if
the first move drops too far. I print out a "--" eval notice to let me know
that the first move looks worse than expected, but early versions did not,
nor did Cray Blitz.. it would just chug thru the second, third, etc. move
trying to find one that returned a value > alpha...

: In the previous move, the 35th (Bxd6) DB thought for 15 minutes. It

: seems cleat that the "panic-time" mechanism was activated there (for
: some unknown reason), and the machine thought until "panic-time"
: interrupted it. This is significant I think. Was DB malfunctioning at
: this point of the game. Could it be that an operator, seeing the
: depressing "panic-time" messages appearing for the 2nd move in a row
: really panic and push the "Move Now" button ?

Was there any evidence to suggest this? Again, I see no reason to start
such a line of speculation without some clear evidence. It really borders
on slander...

: There is no reason to rush into any conclusions here, but the DB team

: needs to give answers to the these questions:

: 1. What was the basis for preferring 36. axb5 over 36. Qb6 ?

: 2. Why was "panic-time" activated in the 36th move and why was a move
: played when it finally played ?

I, for one, would like to know anything about what they did. But, in
reality, they certainly don't owe us an explanation. Kasparov doesn't
explain why he plays what he does, why he decides to think for 1/2 hour
in a simple position, etc. I'm certain Murray isn't a good enough chess
player to start overriding moves, however... and I'm sure he wouldn't
do it either...

[brucemo]

Robert Hyatt wrote:

> Ah, yes... back to the "no computer could play that..." sort of bullshit I
> see. I notice that Kasparov seems satisfied now. He is not continually
> raising that issue any more. Ferret found that move at ply=20. Crafty found
> it at ply=20... but of course, it was still cheating, you think?

Ferret didn't find it through ply 23.

My original post said that the move had moved up in the search order until it
was the second move searched at ply 20 or so.

This appears to have changed in the most recent version, it's now a ways back
in the search order still.

What happened with Be4, I think, is that all of the humans were going gaga,
expecting white's position to collapse. They assumed that Qb6 would win
material. I don't know if Kasparov thought that it would win material. But the
humans that I was listening to at the time all had a very defeatist attitude.

People seemed surprised by Be4, since it appeared to be one of those moves that
delays doing something good in order to do something great later.

It also seems to be a profoundly non-computer move. At the time, I could swear
I can remember hearing people suggest Be4 as a move, but there was always the
follow-up comment that a computer wasn't going to do that unless it was a crush,
and it would probably be too deep to see a crush.

But I think that what was happening is that white wasn't winning, white was
watching the eval for Qb6 fall on every ply, and eventually Be4 bubbled to the
top of the list, with a score of less than +1.00.

After 23 plies, my program still wants to play Qb6, but the score is +0.13.

bruce

[Jim Caccamise]

Robert Hyatt <hy...@crafty.cis.uab.edu> wrote in article

<60hq6b$t...@juniper.cis.uab.edu>...

> AMIR BAN (ms...@ibm.net) wrote:
> : Robert Hyatt wrote:
> : >

<<<...snipped...>>>

>
> Hmmm... while I don't want to start a big argument, how much have you
> published in the way of technical details about your program? Gone as
> far as I have? It is easy to criticize the DB guys, but don't forget
> to do an occasional reality check and compare what they've written
> against what most others have written. They have a pretty good record
> for publishing about their "system"...
>

<<<...snipped...>>>

It is understandable for commercial and/or competitive reasons why the
release of information would be controlled for a chess program/computer.
But, the official announcement of DB's developmental and competitive
retirement eliminates the normal reasons for controlling release of
information on the inner workings of DB.

Jim Caccamise

[Robert Hyatt]

Jim Caccamise (c...@magicnet.net) wrote:

: Robert Hyatt <hy...@crafty.cis.uab.edu> wrote in article
: <60hq6b$t...@juniper.cis.uab.edu>...

: > AMIR BAN (ms...@ibm.net) wrote:
: > : Robert Hyatt wrote:
: > : >

: <<<...snipped...>>>
: >
: > Hmmm... while I don't want to start a big argument, how much have you

: > published in the way of technical details about your program? Gone as
: > far as I have? It is easy to criticize the DB guys, but don't forget
: > to do an occasional reality check and compare what they've written
: > against what most others have written. They have a pretty good record
: > for publishing about their "system"...
: >

: <<<...snipped...>>>

: It is understandable for commercial and/or competitive reasons why the
: release of information would be controlled for a chess program/computer.
: But, the official announcement of DB's developmental and competitive
: retirement eliminates the normal reasons for controlling release of
: information on the inner workings of DB.

: Jim Caccamise

The demise of Deep Blue was apparently a "misquote". There should already
be a retraction available. This according to the DB team, anyway. Of
course, management can make decisions without their being involved, but it
seems doubtful it would happen like that...

[Don Getkey]

In article <342BC9...@ti.com>, Chris Carson
<chris-car...@ti.com> writes:

>Don,
>
> Thanks for posting this article (could I get the URL?)
>
> I do not know why the descrepancy. The IBM web site was
>around when we started this discussion last spring and
>represents an average speed (as stated on the web page).
>
>

This might help you, , ,

http://www.chron.com/content/search/index.html

[Jim Caccamise]

>
> The demise of Deep Blue was apparently a "misquote". There should
already
> be a retraction available. This according to the DB team, anyway. Of
> course, management can make decisions without their being involved, but
it
> seems doubtful it would happen like that...
>
>

GOOD NEWS!
I hope (as do many others) this is a misquote. It is also good news
that the DB team reacted by labeling it as such. Certainly there isn't a
consensus to move on - when there would be no need to identify this as a
misquote.

Jim Caccamise

[Amir Ban]

AMIR BAN wrote:

(Regarding DB 36'th move in the 2nd game)

I looked again at the material and I need to correct somewhat the
description of events as told in the 1st paragraph. They were even more
unusual: Actually Qb6 was on the PV right up to when the search was
interrupted by the premature "panic-time". At this time, the PV was Qb6
with an eval of +0.48. The search was interrupted with "panic time"
messages, and an announcement that the main line will be reconstructed,
which produced out of the blue axb5 with an eval of +0.63. As I wrote,
the panic time still had about 7 minutes to go, so it's unclear what
brought on this.

I guess I am still asking the same questions, but I am also curious to
know what caused a new best move to surface out of the act of stopping.
This does not happen in regular alpha-beta programs, where if you
interrupt the search, they would simply play the move from the current
PV. Perhaps in DB, when they interrupt they also take into account past
evals from lower ply depths. This sounds risky and dubious to me, but
maybe it works for them. Still, you wonder what caused DB to use what
must be described as emergency measures when it still had 7 full minutes
to think.

Amir

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

[Robert Hyatt]

Amir Ban (ami...@m-sys.com) wrote:
: AMIR BAN wrote:

: (Regarding DB 36'th move in the 2nd game)

: > The printouts show that Qb6 was the move considered all the time. DB did

: >

: I looked again at the material and I need to correct somewhat the

: description of events as told in the 1st paragraph. They were even more
: unusual: Actually Qb6 was on the PV right up to when the search was
: interrupted by the premature "panic-time". At this time, the PV was Qb6
: with an eval of +0.48. The search was interrupted with "panic time"
: messages, and an announcement that the main line will be reconstructed,
: which produced out of the blue axb5 with an eval of +0.63. As I wrote,
: the panic time still had about 7 minutes to go, so it's unclear what
: brought on this.

: I guess I am still asking the same questions, but I am also curious to
: know what caused a new best move to surface out of the act of stopping.
: This does not happen in regular alpha-beta programs, where if you
: interrupt the search, they would simply play the move from the current
: PV. Perhaps in DB, when they interrupt they also take into account past
: evals from lower ply depths. This sounds risky and dubious to me, but
: maybe it works for them. Still, you wonder what caused DB to use what
: must be described as emergency measures when it still had 7 full minutes
: to think.

: Amir

As I've said, they do something "different". I don't know exactly how
it works, but would like to. I saw this happen at different points in
the various games I've had the opportunity to look over their shoulder
on.

But the behavior doesn't sound odd to me at all, ignoring DB. I can
show you thousands of log file entries for Crafty where it plans to
play a move X after 4 minutes, but comes up with better analysis at
the last minute. This sounds to me exactly like what happened. They
simply changed their mind, and that changing their mind somehow removed
the "panic" condition.

IE, in crafty, I search for 12 plies (iterations) and get +1.3 for
every one, +/- some small amount. Then at 13, this drops to +.3. I
instantly go into a "panic time" condition and Crafty will not stop the
search "on time" because of the lower eval. But, if it finds another
move that brings the score back up to within "x" of the original 1.3,
it will pop out of panic mode and make that move.

I simply think they define "panic" differently, based on observation.
I don't think it is something they "have" to explain, as they don't have
any sort of mandate to help the rest of us (particularly commercial
competition) get better. What they do seems *very* good. I'd like to
understand it. But I'm not rattling my sword because I don't... there's
plenty of other things to work on at present...

[AMIR BAN]

Jim Caccamise wrote:
>

> >
>
> Did DB only have 7 minutes to go till time control? How much time did
> Kasparov have left?
>

No. DB had 40 minutes left when it started thinking on this move. The 7
minutes is what remained on its "panic time" plan. Let me explain: At
the start of the move, DB planned to think about 3.5 minutes if
everything looks normal, and about 14 minutes of it goes into what it
calls "panic time", which I don't know how they define, but means DB
senses some trouble. Many programs do similar things with the time
planning. DB did go into its panic mode in this move, but played a move
7 minutes before its panic allotment expired.

> I have thought that part of the DB strategy was to pressure Kasparov
> on time. I understand that DB approximately doubled it's positions per
> second capababilities from the previous match. But, this doesn't buy
> much in ply depth when you are already searching deeply. What it does
> give is the ability to play to the same depth in half the time, allowing
> less opportunity for Kasparov to think on DB's time. ( Some of DB's time
> gain could also be used for positional evaluation improvements and
> additional search extensions.)
> I think it's reasonable that the DB team would be conservative in an
> effort to avoid time pressure. Plus, their liberal search extensions
> allows conditions where excessive extensions occurs. If time control was
> approaching, it is easy to believe that some sort of "panic time" mode
> could be automatically triggered when extensive search extensions are
> occurring. This would allow an adequate search extension time buffer for
> the remaining moves in the control.

DB time strategy in this match struck many observers as not
sophisticated. They played usually faster than necessary, and their
biggest concern appeared to be not to get into time trouble even at the
cost of leaving an hour on the clock.

Amir

[Robert Hyatt]

Ilias Kastanas (ika...@alumni.caltech.edu) wrote:

: Presumably, panic-time is triggered in some situations of a
: falling evaluation. But it could be other parts of the program can
: invoke it too as an emergency measure, upon noticing a "new" danger
: or other surprise.

Actually, it is more involved. I remember from some journal article
they wrote (paraphrased, but fairly close to the idea): "if the
tree search can't be expressed as a canonical representation of the
search depth and number of pieces on the board, DB will use more time
since it is having trouble resolving a clear best move..."

I have seen this happen against Cray Blitz, where they didn't fail low,
but still took a lot more time than normal because they were seeing
something in the search that seemed "not right."

: Panic-time ran its full course on move 35. It is possible
: a watchdog mechanism limited it to half that much for move 36. Time
: allotment involves tradeoffs; e.g. keeping a reserve to be able to
: call panic-time again.

: One thing is clear: the machine has sensed some difficulty;
: and the machine is playing against the World Champion. Telling it
: "Don't think -- just move. Now!" doesn't sound like the thing to do.

: Anyway, this deeper delving into the position apparently led
: DB to 36. a:b5..., 37. Be4! -- a "human's intuition" move cutting
: the opponent's options that Kasparov marvelled about. No, I wouldn't
: call that "malfunctioning"!

: Ilias

I hate to be abrupt here (and this is *not* directed at Ilias, who seems
well-balanced), but it is *really* disappointing to see these continual
speculations about the DB team cheating. It shows poor sportsmanship on
the part of Kasparov, whom I now consider to be a 100% jackass with the
morals of an 18th century hooker. It also shows professional jealousy on
the part of a few chess-programmer types that want to continue to expound
this speculation without *any single shred* of evidence.

Sad... *really* sad. I can guarantee you that *I* don't look up to
the people that continue to speculate thusly.

Glad guys like Ilias can see the holes in the claims. I hope most
everyone does..

[Ilias Kastanas]

In article <342C5C...@ibm.net>, AMIR BAN <ami...@m-sys.com> wrote:
...

>The world has noticed, and IBM are well aware, that Kasparov has raised
>the suspicion of cheating, with the 2nd game in mind. Rather than
>discuss and reveal everything they have on this remarkable game, they
>chose to deliver a few pages of cryptic dumps marked "IBM confidential"
>to Kasparov and his aides eyes only. Since Kasparov asked us to decipher
>those, I have seen them (I understand something was also published on
>the NY Times. I don't know if everything or only part of it). The
>interesting part of it were the printouts of the 36'th and 37'th moves
>of game 2 (axb5 and Be4).

No easy task, drawing conclusions about a parallel-processes
system from just a printout. Or, for that matter, from a posting about
the printout!...

Eh... only to hear "I'm sorry, Dave, I'm afraid I cannot
do that..."

Presumably, panic-time is triggered in some situations of a
falling evaluation. But it could be other parts of the program can
invoke it too as an emergency measure, upon noticing a "new" danger
or other surprise.

Panic-time ran its full course on move 35. It is possible

[brucemo]

Amir Ban wrote:

> I looked again at the material and I need to correct somewhat the
> description of events as told in the 1st paragraph. They were even more
> unusual: Actually Qb6 was on the PV right up to when the search was
> interrupted by the premature "panic-time". At this time, the PV was Qb6
> with an eval of +0.48. The search was interrupted with "panic time"
> messages, and an announcement that the main line will be reconstructed,
> which produced out of the blue axb5 with an eval of +0.63. As I wrote,
> the panic time still had about 7 minutes to go, so it's unclear what
> brought on this.

I haven't seen this. If you can post it, please do.

I don't know what they did, but if the value of the PV was falling, or
failed low, might it have generated these messages?

bruce

[brucemo]

Robert Hyatt wrote:

> I wouldn't argue that either was right or wrong, not knowing more about who
> did what. For the hardware questions, I'd suspect Hsu is dead on. He
> designed the whole thing, and revised the design several iterations until
> he reached the DB-II configuration. I strongly believe that 1M (or a little
> less) is the right number for the current hardware, based on other
> communication with Hsu and based on the IBM web site with DB info last year.

I got more info from New York. I am not allowed to quote my source, but I can
summarize what I got in the mail.

I reported 10-0 in the first match, 38-2 overall, versus a P5/90, at time
controls of between 10 0 and 30 0, with "Tiny Blue" at 2M nps.

I also mentioned that there were "some search extensions turned off."

Apparently I should have said "some search cutoffs". I don't know where the
source of this mistake is, but I assume I heard wrong.

Also, I have been informed that the node rate on that old machine was 1.6M nps
and not 2.0M nps.

In the 10-game match, approximately 8 of those games were on a P5/90, and
approximately two were on a P6/180. Also, almost all of the games were at 30 0
time controls.

Ok, so what does this mean? What I want it to mean is that what I said before
was essentially correct. But there is a problem with this. Those search
cutoffs I mentioned had some effect upon the strength of Tiny Blue. This
effect is estimated to be large. So I have been informed that the absence of
these cutoffs degraded the search by 80-90%, although this figure is much less
in endings (according to another email I got).

So this brings us back to the issue of whether Tiny Blue is a faster searcher
than the micros it played. And it appears that this is still true, but to a
lesser extent than I reported previously.

It sounds like we are seeing the source of the confusion between what Bob said
Hsu said and what I said that Campbell said. Campbell's NPS figure did not
take into account search speed degradation due to cutoffs that had been turned
off, and the figures that Bob reported did.

There are still some disparities between what he and I have reported, but they
aren't that major. It does become kind of hard to assess "Tiny Blue".

I think it is important to not lose sight of what Bob was trying to prove by
bringing this all up in the first place, and what the data that has been
presented does seem to indicate.

Deep Thought had a very simple evaluation function, which puts it way out at
one end of the chess program ideological spectrum.

There are those of us who want to believe that our creations are better than
Deep Blue, and our reasoning would involve the quality of our own evaluation
functions, since we sure aren't going to go any faster.

Bob was attempting to show, by reporting on this match, that DB is superior,
node for node, to the micro programs. Personally I don't think that this
experiment was controlled well enough to demonstrate this -- that wasn't the
original point of the experiment, after all, all Campbell was trying to do was
find bugs. But I think it does show that Tiny Blue doesn't get run over
positionally by micros. I think I can safely assume that DB wouldn't either.

bruce

[Thomas Davie]

In article <60go3u$dp5$2...@news00.btx.dtag.de>, TUESCHEN.MEDIZ...@t-online.de (Rolf Tueschen) wrote:
>But let's not build a second front. The scandal of the DB team is topic
>number one.

Again, Deep Blue won. Kasparov lost. Get over it.

And this is unallowed ad-hominem, at least following the accepted
>world-wide standards of science. You wanted to construct the impression
>that I couldn't be sound with my critic against your bad behavior
>because I "simply" was not playing in your league.

No, that argument is an inverse ad vericundium argument, which is the
implication that one in a position of specific/specialized knowledge is more
likely to have an accurate opinion on any given subject that is of that
specific/specialized knowledge. Get your argumentation correct please.

>Example: the winner of the american football internal championship is
>directly World Champion too...

Of course, there is no other competitive professional football league in the
world which can compete at an equal level with the NFL

>
>Same in baseball I assume.

This one is a bit more debatable.

Tom

Brought to you from Manitoba, Canada
The dryest province in Confederation.

[Jim Caccamise]

Amir Ban <ami...@m-sys.com> wrote in article
<342E80...@m-sys.com>...

>
> I looked again at the material and I need to correct somewhat the
> description of events as told in the 1st paragraph. They were even more
> unusual: Actually Qb6 was on the PV right up to when the search was
> interrupted by the premature "panic-time". At this time, the PV was Qb6
> with an eval of +0.48. The search was interrupted with "panic time"
> messages, and an announcement that the main line will be reconstructed,
> which produced out of the blue axb5 with an eval of +0.63. As I wrote,
> the panic time still had about 7 minutes to go, so it's unclear what
> brought on this.
>

> I guess I am still asking the same questions, but I am also curious to
> know what caused a new best move to surface out of the act of stopping.
> This does not happen in regular alpha-beta programs, where if you
> interrupt the search, they would simply play the move from the current
> PV. Perhaps in DB, when they interrupt they also take into account past
> evals from lower ply depths. This sounds risky and dubious to me, but
> maybe it works for them. Still, you wonder what caused DB to use what
> must be described as emergency measures when it still had 7 full
minutes
> to think.
>
> Amir
>

Did DB only have 7 minutes to go till time control? How much time did
Kasparov have left?

I have thought that part of the DB strategy was to pressure Kasparov

on time. I understand that DB approximately doubled it's positions per
second capababilities from the previous match. But, this doesn't buy
much in ply depth when you are already searching deeply. What it does
give is the ability to play to the same depth in half the time, allowing
less opportunity for Kasparov to think on DB's time. ( Some of DB's time
gain could also be used for positional evaluation improvements and
additional search extensions.)
I think it's reasonable that the DB team would be conservative in an
effort to avoid time pressure. Plus, their liberal search extensions
allows conditions where excessive extensions occurs. If time control was
approaching, it is easy to believe that some sort of "panic time" mode
could be automatically triggered when extensive search extensions are
occurring. This would allow an adequate search extension time buffer for
the remaining moves in the control.

--
Jim Caccamise
( Remove X from e-mail address, anti-spam )

[Robert Hyatt]

AMIR BAN (ms...@ibm.net) wrote:
: Jim Caccamise wrote:
: >

: > >
: >
: > Did DB only have 7 minutes to go till time control? How much time did
: > Kasparov have left?
: >

: No. DB had 40 minutes left when it started thinking on this move. The 7

: minutes is what remained on its "panic time" plan. Let me explain: At
: the start of the move, DB planned to think about 3.5 minutes if
: everything looks normal, and about 14 minutes of it goes into what it
: calls "panic time", which I don't know how they define, but means DB
: senses some trouble. Many programs do similar things with the time
: planning. DB did go into its panic mode in this move, but played a move
: 7 minutes before its panic allotment expired.

I still fail to see *anything* odd here. Here is a sample from a game
against a GM (averaging about 30 secs/move) where crafty's target time is
about 30 seconds. It's "panic time" is 5x that, or 2:30, although it doesn't
disclose this in the output.

time limit 30.22
depth time score variation (12)
12 5.25 -3.162 Rh2 Bxa3 Kd4 Bb4 Ke3 d2 Rxd2 Bxd2+
Kxd2 Bb1 Ke3 Kf5 Kd4 Be4
Black(40): f7g6
predicted move made.
12-> 14.92 -3.162 Rh2 Bxa3 Kd4 Bb4 Ke3 d2 Rxd2 Bxd2+
Kxd2 Bb1 Ke3 Kf5 Kd4 Be4
13 17.07 -- Rh2
13 25.72 -4.530 Rh2 Bxa3 Kd4 Bb4 Ke3 d2 Rxd2 Bxd2+
Kxd2 Bh3 Ke3 Kf5 Kd4 Kg4 Kd5 Kxg3
13 1:58 -3.234 Rg2 Bxa3 g4 Bxg4 Ke4 Bb4 Kxd3 Kf5 Rf2
Be1 Rf1 Bxh4 Kd4
13-> 1:59 -3.234 Rg2 Bxa3 g4 Bxg4 Ke4 Bb4 Kxd3 Kf5 Rf2
Be1 Rf1 Bxh4 Kd4
time: 1:59 cpu:99% mat:-3 n:10485357 nps:88431

At ply=13, the score drops by 1.4, from -3.1 to -4.5... and you will
notice that crafty doesn't kick out after the time limit (30.22) has
expaired. It keeps searching to resolve this problem, until one of two
things occurs: (1) it completes the current iteration. Since no moves
improve things, it doesn't try another iteration. (2) it finds a move that
brings the score back to within some finite "delta" of the score from the
last iteration. In this case, either of those two could terminate the
search, because Rg2 was the last ply=1 move to examine, and it also was
enough to bring the score back to what was expected. Notice that Crafty
"mysteriously" doesn't use the entire 5x "panic time"... that it bails out
after only 4x. So I simply don't see why Deep Blue doing so is something
suddenly sinister-looking... I suppose I miss the point somehow...

: > I have thought that part of the DB strategy was to pressure Kasparov

: > on time. I understand that DB approximately doubled it's positions per
: > second capababilities from the previous match. But, this doesn't buy
: > much in ply depth when you are already searching deeply. What it does
: > give is the ability to play to the same depth in half the time, allowing
: > less opportunity for Kasparov to think on DB's time. ( Some of DB's time
: > gain could also be used for positional evaluation improvements and
: > additional search extensions.)
: > I think it's reasonable that the DB team would be conservative in an
: > effort to avoid time pressure. Plus, their liberal search extensions
: > allows conditions where excessive extensions occurs. If time control was
: > approaching, it is easy to believe that some sort of "panic time" mode
: > could be automatically triggered when extensive search extensions are
: > occurring. This would allow an adequate search extension time buffer for
: > the remaining moves in the control.

: DB time strategy in this match struck many observers as not

: sophisticated. They played usually faster than necessary, and their

: biggest concern appeared to be not to get into time trouble even at the

: cost of leaving an hour on the clock.

"not sophisticated" seems to be yet another back-handed slap at them. It
looked pretty damned clever to me, when Kasparov got short on time, it tried
to pressure him. It took extra time when it saw trouble. Moving a little
fast is not a bad plan... if it ends up affecting the tempo of the game, or
if it tends to get your opponent into time trouble...

But I would not say "not sophisticated" without knowing some details of
what they do. They might have thousands of lines of code and some real
intelligence in the time-allocation code.

: Amir

[Robert Hyatt]

Vincent Diepeveen (vdie...@cs.ruu.nl) wrote:
: In <60hq6b$t...@juniper.cis.uab.edu> hy...@crafty.cis.uab.edu (Robert Hyatt) writes:

: >AMIR BAN (ms...@ibm.net) wrote:
: >: Robert Hyatt wrote:

: Amir, can i see somehow this logs?
: Would be interesting.
: I don't have them. mailing me a bitmap of it is also no problem.
: fax: netherlands 318 550408

: >: If you couple this with IBM's (especially Mr. Tan's) statements of

: >: say-nothing gibberish, and IBM's very obvious policy of
: >: keep-your-mouth-shut before, and amazingly, after the match, you will
: >: understand that it is NOT the case that some annoyingly slow editors of
: >: some obsocure journals are keeping the story from the world while the
: >: the Deep-Blue team is biting its nails in frustration at the delay.

: >Hmmm... while I don't want to start a big argument, how much have you
: >published in the way of technical details about your program?

: How well does his program play compared to yours.
: Perhaps that is more interesting. Seems he understands things better than
: you in that respect!

They are pretty equal in the games I've seen. But that has *nothing* to do
with the comment above. *if* he had published anything in a reputable
journal, he'd have answered his own question, because it takes a *long* time
for something to appear in a journal, *after* it is submitted. A year is
considered very fast. Two years is not uncommon.

: Same comparision, sorry.

: Don't begin about Crays versus PC. I am even willing to start a match
: Diep versus Crafty on a Cray. End of this week you can email me the Cray moves
: to di...@xs4all.nl

: However that match i want to have at least 32 MB hashtables on my Pro, or
: 64 and about an hour pro move (not much more, because i must do other things
: too), versus 1 hour and 1 second (at that unix machine i give you 1 second
: extra to swap your hashtables into memory...).
: This time i also play with a tournament book which by the way looks a lot like
: a lot like your tournament book... ...so that book will this time not be
: the problem.

: > Gone as

: >far as I have? It is easy to criticize the DB guys, but don't forget
: >to do an occasional reality check and compare what they've written
: >against what most others have written. They have a pretty good record
: >for publishing about their "system"...

: This is of course true. We can easily assume that just making the system work
: (hardware chessprocessors!) is already a fulltime job.

: >: Please remember we are not waiting to hear details about the SP

: I also don't believe that the DB guys cheated. I wouldn't want to say
: this about Kasparov's performance. As we know Kasparov is always open
: for all options.

: Look how he plays now in Tilburg, Netherlands. About 700 elopoints stronger
: play than he showed against DB. He could do it he can do it, so why didn't
: he do it against DB?

: Fact is however that the DB team didn't publish a lot.
: Even more concerning is that it seems DB will not play anymore.

As I posted earlier this week. Deep Blue's "demise" was premature, according
to the DB team. It was apparently a mis-quote.

: They would of course be stupid to play with it. They might loose!

and they would probably win. But *any* outcome is possible.

: If they are however so sure that they kill other programs with it,
: then why not play next year with it?

Wouldn't surprise me to see them. They have hardly ever missed a computer
event that I recall.

: How could they loose, using Roberts redenation.

Not easily, to be sure. But they *could* lose when one game means so
much. But I'd suspect that they would have a 95% chance of winning
each game they play against computers. With enough rounds, they are
guaranteed to lose a game of course.

: Of course the PC programs will try to make dead meat of them, no matter how.
: (if it makes already a mistake in opening, then it definitely is not
: stronger than most GM's, because as we know GM's can win won positions).

Baloney. Want me to post a few GM games where the GM hung a piece, or
overlooked a mate in 1 and had to sac a queen to avoid it?

: >: The printouts show that Qb6 was the move considered all the time. DB did

: >: well to see Black's best counterplay after Qb6, throwing 2 or 3 pawns to
: >: open lines and attack the white king, as shown by the PV. The evaluation
: >: throughout was around +0.5 pawns for white and dropping slowly. Finally
: >: axb5 was accepted without a PV with a value of 0.4-0.5.
: >
: >: There were several things the needed explanation in the deciphering:
: >
: >: - There was no explanation why the Qb6 score dropped so low with white
: >: ahead by 2 pawns or more. I mean, it's not like you could see the
: >: material regained or a decisive king attack mounted. The only
: >: explanation was that the evaluation function gave the compensation, in
: >: contrast to all PC programs who evaluate white to be more than a pawn up
: >: there. So it seems that DB is not so much the calculating giant as the
: >: positional wizard ?
: >
: >not at all. Both Crafty *and* ferret found this move, it just takes a long
: >time. We saw draw by repetition on Qb6 at ply=20. Remember that DB's
: >hardware probes the hash tables in the chess processors, but that the
: >chess procesors do not return any sort of path information or anything...
: >just a score for the depth requested. So it would be nearly impossible to
: >see what was going on beyond the point where the SP hands something to the
: >chess processors.

: I agree with both parties: certain Diep versions have not much problems
: with the move. On the other hand it is very strange that there is no PV.

Check this out. By your definition, Crafty must be "very strange"...

Black(80): Kc4 [pondering]
time limit 2.44
depth time score variation (6)
9 1.28 1.509 Ke5 Bb4 Ke6 Bd2 Kd6 Be3 Bf3 Bf4+ Kc6
Kd4
9 2.47 ++ b6!!
9-> 2.67 1.835 b6
10 2.99 ++ b6!!
10-> 4.89 2.134 b6
11 6.56 2.188 b6 Be1 b7 Bg3 Ke6 Kc5 Kd7 Kb6 Kc8 Ka7
a5 Be5 Bf3

After ply 9, *no* pv. After ply 10, *no* pv. Finally, at ply=11, it comes up
with one. I see this frequently, and it is explained by the transposition table.
So that is *not* strange at all. If you look thru one of my logs, *any* log you
choose to pick at random, there will likely be a move made with *no* PV. It
is common to see the best move fail high on the last iteration, but time runs
out before the PV has been calculated. What's strange about that? Here's an
example:

Black(85): c4c5
time used: 1.06
time limit 2.13
depth time score variation (1)
6 0.11 ++ b8=Q!!
6 0.23 19.532 b8=Q Bg1 Qa7+ Kd5 Qxg1 Ke6 Qe3+ Kf5
Bc2+ Kf6
6-> 0.24 19.532 b8=Q Bg1 Qa7+ Kd5 Qxg1 Ke6 Qe3+ Kf5
Bc2+ Kf6
7 0.51 19.724 b8=Q Bg1 Qa7+ Kd5 Qxg1 Ke6 Bg4+ Kf7
Qa7+ Kg6 Kc7
7-> 0.52 19.724 b8=Q Bg1 Qa7+ Kd5 Qxg1 Ke6 Bg4+ Kf7
Qa7+ Kg6 Kc7
8 0.82 ++ b8=Q!!
time: 2.26 cpu:89% mat:13 n:214028 nps:105432
ext-> checks:38031 recaps:0 pawns:413 1rep:2594
predicted:59 nodes:214028 evals:21002
endgame tablebase-> probes done: 1182 successful: 1182

No PV for the move from depth=8 at all...

: If you have transpositiontables you ALWAYS have a PV from hash,
: and we can be sure that they use hashtables somehow. Especially
: if you use a good update system you can get even from the smallest
: hashtable a PV.

If you understand transposition tables, you *know* that you do not
always have a PV. Unless you require your hash table to be large enough
to store the entire tree. DB *can't* do that, because at 200M nodes per
second, how much memory will that take? Using my 16 bytes/entry, that
comes out to a paltry 3.2 gigabytes/second they must have. No machine
yet built has that kind of memory. Biggest machine I have seen is a 16
gigabyte Cray T90. So please don't post nonsense...

your "smallest hashtable a PV" is also nonsense... because positions out
near the tips of the PV *should* get overwritten by other more useful
positions as the search progresses... I have two hash table algorithms that
I use, one the double-table approach from Belle, one the multi-probe idea from
Cray Blitz (which Don Beal tested in his JICCA article and pretty well proved
it was better when the table is over-saturated). And neither of these could
produce good PVs when playing around with mtd(f). Nor can they produce good
PV's reliably on the current code. I've tried it many times. Maybe most
of the time, yes. But not all the time.

: You don't need to be too smart to realize this, and
: as we know the DB team is everything except stupid.

you have to be really smart to be able to do something that is, by definition,
impossible.

: >But finding that draw is not impossible.. when you get to Paris, ask Bruce.

: >He ran it to 20 plies before I did, and found that Qb6 is a draw, and that
: >Be4 is slightly better than a draw. I confirmed it by letting Crafty search,
: >although it took an incredible amount of time to do so...
: >
: >[note... I hope we are at the same point in the game... I don't have the
: >pgn handy, but remember the qb6 vs Be4 move discussion. If this is the
: >wrong place or move, excuse the mistake...]
: >
: >: - No PV for axb5 (but that can happen of course).

: At that machine No way . You always have a PV. Also with nullwindow search.

bullshit. See the two points above with logs included from Crafty. Have
you *ever* failed high, but could not complete the search after raising
beta, before you ran out of time? Either you don't play many games with
your program, or you are very inattentive while it is running, or you lose
lots of games on time. Because it *must* happen.

: For an updating system using probe=2, and using overwriting tuplse you
: can very easily prove that you have at least 2 moves of the pv, the
: move that failed high and the move after that. Of course if you
: use hashtables with small sizes (few kilobytes), it could be the case
: that the PV is not much longer. You must however be very unlucky if
: this happens.

Two moves is *not* a PV. If the average PV is 30 plies, which would be
about right for their selective search, that would be a fraction of it.
But you don't even get two moves. You get *the* move that failed high,
and that is all... because at the next ply *every* move failed low. How
could you decide which one of those to store in the table? There's *nothing*
to go on, so any move you store at ply=2 is simply random. Save the *previous*
move as best when this happens? Still completely random...

: If you don't use the Nullmove then the chance that your pv is only few
: moves becomes even smaller.

totally wrong. See above output. Would you like me to play one game with
null-move disabled and post the log? Or, even better, go back to Cray Blitz
which used non-recursive null move R=1? It happened all the time there too.
I even have logs prior to 1981 when we added null-move to Cray Blitz. And we
still had a search ending on a fail-high with no PV. And PV's that were
truncated by hash table hits...

: >If I got concerned every time this happens to Crafty, I'd have thrown it

: From a remote machine you can expect everything.
: Who checks you?

: Fact is that DB didn't see that in the same game after Ra6,Qe3 would be a draw,
: which most PC programs do see within few hours, especially with some
: bigger hashtables.

There's not been a case of a PC seeing that yet, from the original position. And
we ran several on it overnight. I ran both Crafty and Rebel 8. Bruce ran Ferret.
and we didn't see it in any reasonable time, from the original position, where choosing
the first move for white...

We proved it was a draw, by manually following various paths. And taking many hours
to try the only reasonable-looking moves. But from the first move, I'd like to see
the output from a program that finds that move. *especially* in the tournament time
control DB had to play in. I'm not interested in finding it in 24 hours. This wasn't
a correspondence game...

[Amir Ban]

brucemo wrote:

>
> Amir Ban wrote:
>
> > I looked again at the material and I need to correct somewhat the
> > description of events as told in the 1st paragraph. They were even more
> > unusual: Actually Qb6 was on the PV right up to when the search was
> > interrupted by the premature "panic-time". At this time, the PV was Qb6
> > with an eval of +0.48. The search was interrupted with "panic time"
> > messages, and an announcement that the main line will be reconstructed,
> > which produced out of the blue axb5 with an eval of +0.63. As I wrote,
> > the panic time still had about 7 minutes to go, so it's unclear what
> > brought on this.
>

> I haven't seen this. If you can post it, please do.
>
> I don't know what they did, but if the value of the PV was falling, or
> failed low, might it have generated these messages?
>
> bruce

The printouts are photocopies, and not very good ones, so they can't be
posted. I may write a full report in the JICCA if the parties involved
feel comfortable with that.

I don't quite understand your question, but I can describe what I see:
DB goes into panic-time countdown with Qb6 in the PV. So far so good.
Several minutes of silence, and then Qb6 returns with +0.48 and no PV.
Yes the eval is falling. Panic time countdown continues. So far so good
and reasonable. A minute and a half of silence follows, and then the
search is interrupted with announcement that panic time has arrived (it
still had 7 minutes or so to go). This is followed by announcement that
main line will be reconstructed (from TT, presumably), and then an eval
and PV for axb5 is given and axb5 is played.

When I saw this, I did speculate that a possible explanation was that
the sequence of events was reversed: i.e.: first axb5 was found in the
normal search, and that interrupted the search since a good enough move
was found. That would be a perfectly fine explanation. The problem is
that the printout doesn't support it. It shows the events in the order I
described, and moreover, the new main line is a reconstruction as a
result of the search being interrupted.

There are other strange things there: In the 37'th move, Be4 was
accepted with a faulty PV, the last move losing a rook outright. Frans
Morsch, who also saw this, suggested a possible explanation that DB do
not manage PV's but reconstruct them from the hash table, and he says
that sometimes introduces errors (why ?).

This all leaves me scratching my head, looking for explanations. Maybe
the machine was malfunctioning at the time. If so, it's a remarkable
sleepwalking performance. Perhaps the trouble started with move 35 but I
don't have printouts for it. Theories of intervention out of planned
dishonesty may sound farfetched, but out of operator panic much less so
...

Amir

[brucemo]

brucemo wrote:

> I got more info from New York. I am not allowed to quote my source, but I can
> summarize what I got in the mail.

Also, just got info that it *was* Rebel 8 and not Rebel 7.

bruce

[Thomas Davie]

In article <60nu24$dvi$1...@news01.btx.dtag.de>, TUESCHEN.MEDIZ...@t-online.de (Rolf Tueschen) wrote:
>Hello Tom, ok, let's continue on fields like this.

I'd prefer to talk on sane and rational grounds, but if you wish to talk
gibberish and nonsense, I suppose I could do so.

>
>But the point here quoted above was another one. Please reread the
>former posts. Taken isolated the piece of evidence you quoted your
>commentary would make sense. But the solution is simply (perhaps due to
>language, I dont know for sure) that I criticised RH on his meta-critics
>against people who put questions about the event in May...

Get your argumentation schemes correct. You were wrong. Wrong. You may wish to
peruse the following argumentation schemes; ad vericundium, ad baculum, ad
hominem, ad miseracordium and tu quoque.

>
>Dont confuse that. *I* would never, at least from my actual
>perspective..., challenge his cc related experience (that's his legue,
>for short X1), but I did challenge him in *my* legue, Tom. :)

I'm challenging you on using logically unsound argumentative schemes and being
pathologically insane. What is your league Rolf?

>Take the little paragraph of yours above and change "an accurate opinion
>*also in all other fields*", and you're exactly there, what I called
>Alekhine's trap. Referring to his nazi crap during WW II. As if his
>brilliancy in chess allowed him deeper insight in the different values
>of races... [Noooo, not again, the nazi debate, Tom, please.]

You brought it up Rolf, not I. But, if you wish, I'll resort to that argument.
You are nothing but a simpering Nazi. Now, *that* is an ad hominem argument;
I'm publicly stating that( basically it means that your arguments should be
given less credence because you are 'X', where X is some type of insult ).
Now, do I really believe that you are a Nazi? Nah. But I do believe that you
are mentally unbalanced and display certain clinical symptoms reminiscent of
hebephrenic schizophrenia.

>Another good and not so scandalous example is the often published
>insights of formerly pure natural scientists, in physics for instance
>Heisenberg. Suddenly they detect the eye/hand of God somewhere with
>their theories. Or take this really funny anecdote. Some astronaute once
>told the world that hunfortunately he didn't meet God while surrounding
>the earth with his small capsule. It seems that some officials then
>changed the space code as follows. Now each second astronauts talks
>about his msytic experience while travelling in space... :)

Yes. Ok. Sure. Whatever. But let's get back to your poor use of argumentation,
and let us not talk of the detection of God within theory.

>
>Back to Hyatt, Tom.
>
>Taken him literally, it could happen that he's the only one talking here
>because in some respects we're all way less educated/experienced.
>
>But this won't happen because we have so much to talk about, no?

Quite frankly, I'd rather talk about your behavior and reasoning. I've never
seen anything so fascinating.