I won against Fritz5 (ELO 2791)! - Here is the game
ba...@banet.net
really, I am not supposed to win a game against "something" that pretends to
have a blitz ELO of 2791!
Valentin,B (1598) - Fritz 5.00 (2791) [E12]
Level=Blitz:5'. Bronx, 16.07.1998
10176kB, f5book.ctg
1.Nf3 Nf6 2.c4 e6 3.d4 b6 4.a3 Ba6 5.Nbd2 Bb7 6.Qc2 d5 7.cxd5 exd5 8.e3 Bd6
9.Bd3 0-0 10.0-0 c5 11.dxc5 bxc5 12.b3 Qe7 13.Bb2 Ne4 14.Rac1 Rd8 15.Bxe4
dxe4 16.Qc3 f6 17.Nh4 Bxh2+ 18.Kxh2 Qd6+ 19.g3 Qxd2 20.Rcd1 Qxd1 21.Rxd1
Rxd1 22.Nf5 Nd7 23.Qc4+ Kh8 24.Qf7 Rg8 25.Bd4 cxd4 26.Qxd7 Ba8 27.Nxd4 a6
28.Qa7 Bd5 29.Qxa6 Rd3 30.Qb6 Ra8 31.a4 Kg8 32.a5 Rd2 33.Kg2 Rd1 34.b4 h6
35.a6 Rc1 36.b5 Re8 37.Qd6 Bc4 38.Qd7 Bf7 39.b6 Rcc8 40.b7 Rcd8 41.Qxd8 Rxd8
42.a7 Bd5 43.b8Q Rf8 44.Qc7 Rf7 45.Qd8+ Kh7 46.Qxd5 Rxa7 47.Qxe4+ g6 48.Qe8
f5 49.Ne6 g5 50.Nf8+ Kg7 51.Nd7 Kh7 52.Qf7+ 1-0
bruce moreland
>How could this happen? I mean, I am not the worst player in the world, but
>really, I am not supposed to win a game against "something" that pretends to
>have a blitz ELO of 2791!
You don't say how you won.
Did you play the same opening over and over until you won?
Did you use the takeback function? With enough takebacks, you can
beat Kasparov.
bruce
Valentin Barbulescu
1. Game mode: rated game (not takebacks allowed!), 5 minutes (Blitz), using
Fritz5.ctg and 11 Mb hash tables on a 200 Mhz MMX pentium with 48 Mb EDORAM
2. I did play the same opening again and again but I lost again and again with
1.e4 to which Fritz choose invariably Russian Defense. After I got bored of
playing the same opening I switched to 1.d4, and you saw what happened.
3. So, how can you explain now my victory ?
Valentin
John Nye
games against most of the top fifty GMs running on fast Pentiums. Surely it would
be stronger on a tweaked ultra fast machine. By less than 2600 do you just mean
raw "ability"? I would expect that program strength plus the tendency of even GMs
to blunder at 5 0 would easily give Fritz 5 (running on a hypothetical x000 MHz
machine) a 2600 blitz rating? Are you saying Fritz wouldn't beat an average 2500
player 2 out of 3 times or more at 5 0? Or are we talking about a 2500 player
who's good at blitz and carefully trained to beat computers?
Am I missing something? Please help an outsider understand.
Thanks
f...@accountant.com wrote:
> .... At 5 0 and even on the absolutely fastest PCs
> available, ultra-cooled etc., I do not believe F5 is even 2600. Now at 2 0 or
> faster, I do not think most humans have any chance against it. But at 5 0....
> :-)
>
> Kind regards
>
> fca
Nike Byrne
<Vale...@Barbulescu.com> wrote:
>Here are the details of the game:
>1. Game mode: rated game (not takebacks allowed!), 5 minutes (Blitz), using
>Fritz5.ctg and 11 Mb hash tables on a 200 Mhz MMX pentium with 48 Mb EDORAM
>2. I did play the same opening again and again but I lost again and again with
>1.e4 to which Fritz choose invariably Russian Defense. After I got bored of
>playing the same opening I switched to 1.d4, and you saw what happened.
>3. So, how can you explain now my victory ?
>
>Valentin
>
even a blind squirrel finds an acorn now and then ......
bruce moreland
wrote:
>On Fri, 17 Jul 1998 13:24:15 -0400, Valentin Barbulescu
><Vale...@Barbulescu.com> wrote:
>>3. So, how can you explain now my victory ?
> even a blind squirrel finds an acorn now and then ......
Yeah, I'm not sure what the answer can be. There have been several
posts like this, the basic point is, "I beat my chess program, what
does this show?"
If someone thinks they are a chess prodigy, they are welcome to hit
one of the chess servers and try to beat some strong people.
If they think that because they beat it the chess software sucks,
well, I don't think this is true.
Assuming they aren't a chess prodigy, and the software doesn't suck,
I'm not sure what it shows.
bruce
rene...@#earthling.net
general", that is covering all types of openings, etc. If you
compared any program's performance as black against e4 and d4, it's
performance against e4 is always going to be better, because that
usually leads to a more tactics-oriented game. 1. d4, however,
leading to more positional lines (and requiring deeper search) the
computer will make more postional errors and have fewer tactical
opportunies. This is even more pronounced in blitz, where it can't do
really deep searches. So if you as a human excel at "fast positional
play", you increase your chances quite a bit. However, that is not to
say F5 is still not an extremely strong program anyway, but it does
have weaknesses....
Andy Serpa
renegade@#earthling.net
(remove the # from my address to contact me)
Rolf Tueschen
We have a real problem here. It's so difficult that you won't find
through if you don't keep track on sober logic and definitions. Also
it does help that if we took the verdicts of real experts.
Let me therefore begin with R. Hyatt. Although for most people *the*
expert as such people tend to forget what he told us many times.
Todays MICROS are NOT yet on GM level. Period.
This is as true as it can be.
But how then could FRITZ5 have an ELO of 2791?
Because this number is a result of cheating? ---- Not at all.
We have difficulties with the meaning of the concept ELO.
Here's my theory. But it won't destroy all the false thoughts if even
R. Hyatt's explanations couldn't achieve them ...
&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
Think about an iron chain. How "strong" it is? Let's make this thought
experiment. There's a tradition back to the Middle-Ages where strong
men proved their strength by destoying iron chains. Normally you
couldn't achieve that. Iron is stronger than human muscles, no?
But you remember the saying, that each chain is only as strong as its
weakest member? So you already have the whole secret. If you'd succeed
in putting all your strength combined on this weak member, then the
strongest iron chain will break apart.
Well, that's the secret why average chessplayers could beat a FRITZ
2791. On the contrary there'd be no chance in a billion years that a
nobody/average player/all we masters of lemon blunders could "beat" a
human super GM in a real competion game.
What is the strength of the weak members of a computer program?
Fortunatelly it's low down to 1500 or 1800 at times. That would never
happen with a human player. I give correction. It won't happen on the
basis of "weaknesses", *chess-related* weaknesses. But of course a
human being has other weaknesses too. And we all know the modern
popsong "Money money money makes the world go round."
Let me also reply to your assumptions or ideas above.
You made a well known error. You think that in short games (Blitz and
Rapid) the positional error tendency is even more a disadvantage for
the computer programs. But that is totally the opposite of the
objective results. Micros are at their best in shorter time controls.
Becuase it needs a real master to exploit almost without thinking the
horizon effect of the machines. Somewhere you'll have to calculate
variations and it's an eternal truth that a robot is always superior
to a human on that field.
Therefore it's also a self-betraying if B. Moreland simply answered
that those, who thought they could beat their Micros quite easily,
should come to the ICC servers and then would see ...
Of course. The normal time control is bullet or 2 0, 5 0 or whatever.
Let me state this.
If you are a good A player (= expert level) with theoretical knowledge
of master play then it would take you to experience some 6 months with
DEEP BLUE and then you beat him 5-1.
With all the Micros it's different. Parallel or not, if you had a
month to experiment each day and night, then you beat them all
automatically.
The reason is quite simple. All the robots don't play chess at all.
But they all have those opening storage that is usually only available
for humans from IM upwards ... -- So, you'll have to find your
favorite lines. That alone takes a lot of time.
Then most of the robots already have a lot of endings where they could
play automatically on optimal level. -- So, you'll have to evitate
them. Becuase you could NEVER have the smartness of a super-GM. You
simply would be killed also in dead equal positions.
You get the idea. There're thousands of ridiculous weaknesses of the
robots. You will always be able to discover a few.
(NB that I talk about a human player who could play a robot for a
certain time, who is for this time period NOT under the contol/
manipulation of likewise human influence! Because it would always be
possible to instruct a machine to exploit certain constant errors of a
human "little" expert. So this would be trivial. A GM Benjamin is of
course able to prepare against an 1900 expert... This is completely
uninteresting. We all know that a twisting between games already could
embarras even such a good player as Kasparov.)
MAYANKING
Rolf Tueschen
rene...@#earthling.net
>You made a well known error. You think that in short games (Blitz and
>Rapid) the positional error tendency is even more a disadvantage for
>the computer programs. But that is totally the opposite of the
>objective results. Micros are at their best in shorter time controls.
It may be well known, but it is not an error. Micros are not at their
best in shorter time controls -- obviously the longer time they take
to "think", the better they get. What you meant was that they get the
best results against humans at short time controls in terms of
win-loss record. That is because humans also are "not at their best"
at short time controls. Which is why I added the qualifier, "If you
are a 'fast positional player'..." Meaning if you as the human can
maintain your positional sense under short time controls, AND keep
tactical opportunities for the computer at a minimum by playing closed
positions. (And maybe picking on whatever weaknesses that particular
program might have -- for instance, some programs will freely exchange
a fianchettoed bishop, not realizing that it has weakened itself
positionally because material is even.)
The basic point I was making is that the strength of the computer
varies quite a bit depending on what types of positions you can steer
them into.
>Becuase it needs a real master to exploit almost without thinking the
>horizon effect of the machines. Somewhere you'll have to calculate
>variations and it's an eternal truth that a robot is always superior
>to a human on that field.
Maybe, maybe not. Do you need to be a real master? Obviously not, or
we wouldn't be having this discussion, which started because a
non-"real master" beat the machine at the time controls we have been
talking about. You just need to attack the program's weaknesses,
which he seems to have done (although quite by accident). Unless, of
course, your real argument is that he is a liar.
Wayne Howard
>
>PS: You will have a chance to view R10 vs Anand at a variety of speeds in under 10
>days time.... Join CCC (free) for details, or visit www/rebel.nl
>If you do not know how to join CCC, please ask the question.
Hi, fca!
I'll be the point man on this: How do I join CCC?
I've heard it discussed, and it sounds very much like the kind of
thing I'd enjoy being a part of.
still hoping for an exciting Rebel upset,
Wayne Howard
Howard E
ba...@banet.net wrote in message <35aee...@news1.ibm.net>...
>How could this happen? I mean, I am not the worst player in the world, but
>really, I am not supposed to win a game against "something" that pretends
to
>have a blitz ELO of 2791!
You provided us with your rating of 1598, but from your play in this game
it would have been hard to guess your rating as the level of your play here
was in my opinion higher than 1598. I liked your move 25. Bd4.
The one move of Fritz's that surprised me a bit was 26 ...Ba8. Did
it ever consider the move Rd2 during its analysis or would it play this move
if given more time to think?
>
>Valentin,B (1598) - Fritz 5.00 (2791) [E12]
>Level=Blitz:5'. Bronx, 16.07.1998
>
> 10176kB, f5book.ctg
> 1.Nf3 Nf6 2.c4 e6 3.d4 b6 4.a3 Ba6 5.Nbd2 Bb7 6.Qc2 d5 7.cxd5 exd5 8.e3
Bd6
>9.Bd3 0-0 10.0-0 c5 11.dxc5 bxc5 12.b3 Qe7 13.Bb2 Ne4 14.Rac1 Rd8 15.Bxe4
>dxe4 16.Qc3 f6 17.Nh4 Bxh2+ 18.Kxh2 Qd6+ 19.g3 Qxd2 20.Rcd1 Qxd1 21.Rxd1
>Rxd1 22.Nf5 Nd7 23.Qc4+ Kh8 24.Qf7 Rg8 25.Bd4 cxd4 26.Qxd7 Ba8 27.Nxd4 a6
>28.Qa7 Bd5 29.Qxa6 Rd3 30.Qb6 Ra8 31.a4 Kg8 32.a5 Rd2 33.Kg2 Rd1 34.b4 h6
>35.a6 Rc1 36.b5 Re8 37.Qd6 Bc4 38.Qd7 Bf7 39.b6 Rcc8 40.b7 Rcd8 41.Qxd8
Rxd8
>42.a7 Bd5 43.b8Q Rf8 44.Qc7 Rf7 45.Qd8+ Kh7 46.Qxd5 Rxa7 47.Qxe4+ g6 48.Qe8
>f5 49.Ne6 g5 50.Nf8+ Kg7 51.Nd7 Kh7 52.Qf7+ 1-0
Today's programs are very extreme in that the can beat most anyone, as
Anand said in a recent interview, and likewise they can draw or lose to
most anyone.
Rolf Tueschen
>>
>>You made a well known error. You think that in short games (Blitz and
>>Rapid) the positional error tendency is even more a disadvantage for
>>the computer programs. But that is totally the opposite of the
>>objective results. Micros are at their best in shorter time controls.
>It may be well known, but it is not an error.
Before the details come into view let me please apologize. The way I
stated it my English sounded in your ears in a not intended way. This
might also be the case because the notion "error" is offensive in
normal language. Please take my apologies that I did not mean to say
something as if you were too stupid to reason well. Your article was
so interesting that it inspired me to add my unimportancies. In
science error means something quite normal. It's almost as normal as
the serve in certain ball sports. But I digress and I already wrote my
lecture for this Sunday elsewhere ... :)
>Micros are not at their
>best in shorter time controls -- obviously the longer time they take
>to "think", the better they get. What you meant was that they get the
>best results against humans at short time controls in terms of
>win-loss record. That is because humans also are "not at their best"
>at short time controls. Which is why I added the qualifier, "If you
>are a 'fast positional player'..." Meaning if you as the human can
>maintain your positional sense under short time controls, AND keep
>tactical opportunities for the computer at a minimum by playing closed
>positions. (And maybe picking on whatever weaknesses that particular
>program might have -- for instance, some programs will freely exchange
>a fianchettoed bishop, not realizing that it has weakened itself
>positionally because material is even.)
Yes, I understood you already. Please let's make use of the chance
that we could talk among ourselves without being "distubed" too early
by the famous GM Rober Hyatt. And because I don't know you at all, I
run that risk that you also might be a GM or some such, but let me
make an idiot then out of myself.
I understood. But I don't think that you're correct with the
statement. Your statement meant to me. At longer time controls also
the machines become "stronger". Right?
But this is wrong. And you yourself gave the reasons why not. In
closed positions, or let's say positions with no clear tactical shape,
the robots are not better but even weaker at long time controls. The
stupidity becomes even more obvious. As the absent R. Hyatt once told
us the human positional wisdom is a sort of sine qua non because
humans are unable to out-calculate the variations and then they rely
forcedly on "positional" rules of thumb. Now. The logical consequence
for computers could be that the could get much deeper in longer time
controls and therefore "stronger". But here the 'Second Law of Rolf'
comes into consideration. If you'll allow some tongue in cheek. If you
have no concrete lines to calculate you start to calculate a lot of
bull if only you had time enough. Know what I mean? That's why modern
computer programs still lose more to experienced humans at longer time
controls. And also this is the reason why people like Crafty father
Hyatt are much more interested in shorter time-controls because the
stupidity of their babies is not so obvious to be seen. NB that I
didn't say a word that Hyatt tended to boast as an impostor. How could
I if he's in general the only expert who always comes down to earth by
relativating all the hustle from the PR kitchens around the world.
>The basic point I was making is that the strength of the computer
>varies quite a bit depending on what types of positions you can steer
>them into.
Yes.
>>Becuase it needs a real master to exploit almost without thinking the
>>horizon effect of the machines. Somewhere you'll have to calculate
>>variations and it's an eternal truth that a robot is always superior
>>to a human on that field.
>Maybe, maybe not. Do you need to be a real master?
Excuse me, but at Blitz yes. NB that I said 'exploit'. This is in my
understanding an active notion which means 'intention to/of'.
>Obviously not, or
>we wouldn't be having this discussion, which started because a
>non-"real master" beat the machine at the time controls we have been
>talking about. You just need to attack the program's weaknesses,
>which he seems to have done (although quite by accident).
That exactly was my assumption. That was the core of my argument. And
I said that against a human super GM this wouldn't happen at all. And
again NB, I did not forget about simple blunders in rapids even of Wch
Karpov against Christiansen e.g. But also there it was without bad
consequences because he had more games to equalize the defeat.
No, I won't talk about the Polgar "match" because that would be
totally off-topic here in computerchess. :)
>Unless, of
>course, your real argument is that he is a liar.
Beware. No, not at all. The fact alone that he as unknown user posted
his result here is for me proof enough that he did tell the truth.
Becuase it was such a surprise to him too.
I've beaten FRITZ a hundred of times, so for me this is nothing
special. But I prefer to stay unknown like the GMs who don't want to
be recognized at times when they take part on the servers. BTW FRITZ
contains the "serious mode" where you are forced to do no cheating.
And I tell you this. Irt's a kind of orgasmic experience if you
succeed your first time remis or win after a series of losses. That is
one of the best features and invention of CB/Morsch, that you then
dive into a sort of virtual reality. And if the sounds of the music
start to honor you for being a master in town then all things are
fine! Try to enjoy that for yourself.
How *wrong* you were with your statement could also to be seen if you
looked into the statistics of Blitz against FRITZ for experts like the
GM Anand. He once said that he was happy to hold some 60% (or was it
45%?) against FRITZ3, was it if I remember it right.
With Anand you no longer could say that he doesn't know "positional
play" without even thinking a second. *But* it makes no difference.
The machine is mainly without emotions and fatigue. Oh, I see we
didn't even talk about that...
Daniel Kang
>But how then could FRITZ5 have an ELO of 2791?
It doesn't.
>Well, that's the secret why average chessplayers could beat a FRITZ
>2791. On the contrary there'd be no chance in a billion years that a
>nobody/average player/all we masters of lemon blunders could "beat" a
>human super GM in a real competion game.
I don't buy that claim. GMs sometimes lose by making simple blunders
and many non-masters can sometimes play a rather perfect game. Also
there are a lot of tactical mistakes in games of GMs so why do you
suppose that an average player wouldn't have a chance to spot at
least one tactical mistake enough to win a game when given enough
opportunities?
>If you are a good A player (= expert level) with theoretical knowledge
>of master play then it would take you to experience some 6 months with
>DEEP BLUE and then you beat him 5-1.
Absolute BS. I believe that DB would play better than good A players
even strategically. Also the chance of them making mistakes that Deep
Blue can spot is rather too high for Deep Blue to lose even a game in
three. Didn't Joel Benjamin say that DB is better than him?
Regards,
Daniel Kang
rene...@#earthling.net
>
>I understood. But I don't think that you're correct with the
>statement. Your statement meant to me. At longer time controls also
>the machines become "stronger". Right?
>
>But this is wrong. And you yourself gave the reasons why not. In
>closed positions, or let's say positions with no clear tactical shape,
>the robots are not better but even weaker at long time controls.
I still don't understand the logic of this. Machines get "stronger"
(better moves) the longer time they are allowed to make a move.
Period.
It is the human who gets "more stronger" (if you will) when the time
controls get longer, making it easier for him to win. But, that human
would have an even easier time still if say he was given 2 hours on
his clock and the computer only 5 minutes, would he not?
That is, I am strictly talking about the strength of the moves
themselves, not on the ability of the opponent to capitalize on them.
Rolf Tueschen
>Rolf Tueschen wrote:
>>But how then could FRITZ5 have an ELO of 2791?
>It doesn't.
I hope you as original English speaker that you read my sentence in
that meaning ...
>>Well, that's the secret why average chessplayers could beat a FRITZ
>>2791. On the contrary there'd be no chance in a billion years that a
>>nobody/average player/all we masters of lemon blunders could "beat" a
>>human super GM in a real competion game.
>I don't buy that claim. GMs sometimes lose by making simple blunders
Against a 1900? Please elaborate.
>and many non-masters can sometimes play a rather perfect game.
Against a GM? Please elaborate.
>Also
>there are a lot of tactical mistakes in games of GMs
That's true. Between GMs, yes.
>so why do you
>suppose that an average player wouldn't have a chance to spot at
>least one tactical mistake enough to win a game when given enough
>opportunities?
Errr. ---- Too *few* opportunities?
You astonish me. Because if you're right, with the assumption that a
1900 or even a 2100 should have the time to wait for the one good
shot. Don't you agree with me that he would be shot by the GMs much
earlier? Just my average opinion. The GM makes an error in each game,
ok. But you want to pretend that a non-master would make none? Or at
least not earlier than the GM? Ok, I will think about if I could buy
that.
BTW I waited for such an opinion so badly because as famous FRITZ
-master I wanted to suspend my plans to become World Champion just the
other day after that horrible game which I did loose! But now ...?
>>If you are a good A player (= expert level) with theoretical knowledge
>>of master play then it would take you to experience some 6 months with
>>DEEP BLUE and then you beat him 5-1.
>Absolute BS.
Probably. But I was kind of audacious.
>I believe that DB would play better than good A players
>even strategically.
You dare to propose such a theory in this round of experts? :)
>Also the chance of them making mistakes that Deep
>Blue can spot is rather too high for Deep Blue to lose even a game in
>three.
Them? I was talking about me, the FRITZ-master. Don't take me wrong.
And I was talking about a 6 month titanical fight. Against a DB that
is not allowed to tweaked and manipulated.
>Didn't Joel Benjamin say that DB is better than him?
He did talk a lot of BS, excuse my adoption of your determined
language. At first he did lie that he would publish his games after
the event.
Then he said he didn't win. Then someone said, there're no games
because Joel did always take back and then a new line was checked. So,
in my understanding this sounds as if no *games* at all had been
played.
But I'm a magician. And to prepare a little welcome for Mr Hyatt
tomorrow, let me state this. If *anyone* then also J. Benjamin should
be able to beat DB. Why? Because he knew all about the chosen
openings. Oh, man, we had that debate years ago. DB, please believe
me, is only that strong because it's a cowardly hidden artificial
genial robot. The moment you get close to his secrets, and I was
emphazising 6 months for me personally, he'll eat from your palms.
Ok, I promise to wear gloves to prevent him to eat my hands too.
Rolf Tueschen
>>
>>I understood. But I don't think that you're correct with the
>>statement. Your statement meant to me. At longer time controls also
>>the machines become "stronger". Right?
>>
>>But this is wrong. And you yourself gave the reasons why not. In
>>closed positions, or let's say positions with no clear tactical shape,
>>the robots are not better but even weaker at long time controls.
>I still don't understand the logic of this. Machines get "stronger"
>(better moves) the longer time they are allowed to make a move.
>Period.
Your serve.
Now mine.
Perhaps we've found a new paradoxon.
Let's assume that in tactical positions the normal strength win is
from zero (but this isn't true because zero isn't zero) to such xy.
Well a straight line going up with a degree of gamma ...
Now let's see the same for positions without tactics. The time from 5
minutes Blitz to 30 minutes. But the machine can't find something
special, so not a significant greater amelioration than ny. Now we
take 2 hours. No significant rise of strength. Couldn't we speak of a
machine that is as strong as the 5 minutes machine? If we did that.
Then why couldn't we say that the machine that couldn't profit from 4x
of the time should become weaker and weaker at longer time controls.
Relatively to the humans this is true anyway. Because the human
doesn't have such a boundary in "positional" situations. And also in
tactical he will profit likewise from longer times.
>It is the human who gets "more stronger" (if you will) when the time
>controls get longer, making it easier for him to win. But, that human
>would have an even easier time still if say he was given 2 hours on
>his clock and the computer only 5 minutes, would he not?
>That is, I am strictly talking about the strength of the moves
>themselves, not on the ability of the opponent to capitalize on them.
Let me also state something important for this moment. I see something
interesting in this topic but it would be foolish to talk about yet.
Let me state it like this. I made a clown of myself in public if I
would doubt your last two paragraphes. *But* I just saw something that
would show that *your* logic is true for normal stuff. But in another
understanding of chess logic it might be not true. It might well be
that a machine with a time schedule of a fortnight will not at all
play "better chess" by definition. It's an interesting thought to have
a certain benchmark for each actual program, where it is at the peak
of its strength. To let it run further days wouldn't result in better
results. But fortunatelly it's a bit late by now. I'm at my point
of ultinmate benchmark for today. :)
Good night to you too.
Rolf Tueschen
still hope that you might understand the many misunderstandigs/
intended misinterpretations you make at times. But RL is one side and
this here is cc. So let's see. Let me just give in short my valuation.
f...@accountant.com wrote:
>Rolf Tueschen wrote:
>> In closed positions, or let's say positions with no clear tactical shape,
>>
>> the robots are not better but even weaker at long time controls.
>From context it is clear you are speaking in absolute terms.
Of course not. But I agree there was an aspect that led me to that
statement. Sort of irrationality where the machine should become even
weaker instead of the expected higher strength in such positions as
mentioned (!) because of a sort of troubling/ foggyness result.
I was associating what I have read the other day. Someone wroter that
it's important to tell the machine to no longer go for absolutely
crazy moves in situations what was it, where it sees a direct fall in
the evals. ...
So my theory went like this. No plan, no prospects, nothing to
calculate. Much time. Calculating nevertheless. And so on.
But all that in relation to humans who know also in these position
exactly what do do. So finally I was talking still in relative terms.
> You are not
>here saying that in such positions a robot at 5 0 vs a human at 5 0 stands
>more chance than a robot at say 120 0 vs a human at 120 0
Of course. That was it what I wanted to say. Becuase if someone still
could fall, it would rather be the human, no? -- BTW this all is
depending a bit of my naive understanding of the job of programming. I
never thought a minute about the tricks you ha to implantate for the
machine to be able to distinguish how it should play at fast times. Is
there some scissors in-built to cut the variations or the varability?
And son on. I dont know nothing. But I'm sure the tricks as such are
not too difficult to understand. Technically it might be very
difficult in practice. But I know that all these detailed knowledge
could be exploited by a good human player. That's why the IBM team
made such a conspiracy. They knew quite well what the machine couldn't
achieve yet. Period.
>(which could
>well be true, as the human may not have enough time to find a plan at 5 0,
>while the program benefits little from the extra time in closed positions.
Exactly.
>You are saying that in such positions, the program at 5 0 is stronger than
>the program at 120 0....
Relatively. Yes. Aside the paradoxon mentioned above. But dont ask for
positions.
> With this, I cannot agree even 0.1%. While a
>few program and such-position combinations might show this, for the vast
>majority, the longer time you give the program, the better (or at least
>not-worse) it becomes....
Yes, the MM iV stood still, so to speak. Someone explained to me that
it simply didn't make sense to let it analyse longer than ____.
But I was talking about DB also. Impossible to think of such paradox
happenings? I'm still reminding the interesting statements of GK in
96. With that Rook on h1, Kg1, and the position so that the Rook could
NEVER leave his jail. So here DB could quite easily become a victim of
some thought bonus points while in reality the Rool istn't on the
board so to speak.
>Could you give just one example (actual position) to illustrate your view?
>I can, but my example would also show how rare such examples are, which
>would prove my point.
Yes, but my point was, that I needed 6 months with a DB that would
always be the same. So no tweaking or manipulating in between. Then I
wanted to go for a 5-1 ratio win ...
>A more interesting point may be that in closed positions where the program
>has a material advantage, it continues to do nothing constructive until the
>50 move rule appears in its search horizon. Then, it acts to break-out (by
>pawn move or capture), but its slower piece(s) (typically, King) may well
>be inappropriately placed then... So it throws away the win...
Exactly stuff like that, paradoxical, I was thinking of. Theoretical
all that because you won't find in practice so many such positions. So
in the end *I* won't be able to achieve a single draw. But then again
let Kasparov or better Karpov do that. Would you still think that DB
had a chance at all? My opinion: No!
Siva Chelliah
Nike Byrne <ches...@voicenet.vom> wrote:
>On Fri, 17 Jul 1998 13:24:15 -0400, Valentin Barbulescu
><Vale...@Barbulescu.com> wrote:
>
>>1.e4 to which Fritz choose invariably Russian Defense. After I got bored of
>>playing the same opening I switched to 1.d4, and you saw what happened.
>>3. So, how can you explain now my victory ?
>>
>>Valentin
>>
> even a blind squirrel finds an acorn now and then ......
I would suggest that you repeat/follow the same moves to see if you can
repeat the victory. We need to find:
1) if Fritz will select a diff. move at one point (in the middle game)
2) if there is a bug in the eval. function.
Siva
--
Siva Chelliah Richardson, TX
-----------------------------------------------------------------------------
How can overlook and oversee be opposites, while quite a lot and quite a few are
alike? How can the weather be hot as hell one day and cold as hell another?
John Nye
reasonable to argue that Fritz on the very best PCs available, properly tweaked would
average 2600 when playing against a variety of GMs rated between 2400 and 2700 FIDE at 5
minutes, taking into account both good and not so good anti-computer players. Didn't
Fritz beat most of the GMs it played in a real world blitz tournament last year with real
money at stake? (And not just casual blitz games?)
Rolf Tueschen
>Didn't Fritz beat most of the GMs it played in a real world blitz tournament last year with real
>money at stake? (And not just casual blitz games?)
Yes.
But let's see it from the other side. How many games the humans had to
become adapted to the new (?) version of Fritz?
Experience is the whole secret.
Daniel Kang
>>>But how then could FRITZ5 have an ELO of 2791?
>>It doesn't.
>I hope you as original English speaker that you read my sentence in
>that meaning ...
BTW, English is my second language. You implied that it somehow
attained that rating (it's not a result of cheating you said).
>>I don't buy that claim. GMs sometimes lose by making simple blunders
>Against a 1900? Please elaborate.
>>and many non-masters can sometimes play a rather perfect game.
>Against a GM? Please elaborate.
It is a theoritical possibility nonetheless. Fritz is not a GM anyways.
>You astonish me. Because if you're right, with the assumption that a
>1900 or even a 2100 should have the time to wait for the one good
>shot. Don't you agree with me that he would be shot by the GMs much
>earlier? Just my average opinion. The GM makes an error in each game,
>ok. But you want to pretend that a non-master would make none? Or at
>least not earlier than the GM? Ok, I will think about if I could buy
>that.
Well in a billion years anything can happen.
>>I believe that DB would play better than good A players
>>even strategically.
>You dare to propose such a theory in this round of experts? :)
I believe it and I also believe that many others would support it as
well. A-players make aweful way too often.
Regards,
Daniel Kang
John Nye
Rolf Tueschen wrote:
This is exactly the point I'm trying to make. Human-human ratings cover playing both experienced
and inexperienced opponents. Saying that Fritz is not 2600 against humans optimally trained to play
Fritz may not change the fact that Fritz is 2600 if one takes into account the range of human GMs any
computer (or player) is liable to face in a real world Swiss event. Indeed, with steady improvements
in processor speed since last year, it is likely that Fritz is in fact a reasonable 2600 at 5 minute
blitz against the typical 2500 GM. Furthermore its weakness against trained GMs may be offset by
its ruthlessness against the typical 2200 or 2300 player and overall inability to make a random one
move blunder. Just my opinion, but one I'm willing to discard in the face of good arguments or
strong evidence.
John
Valentin Barbulescu
Siva Chelliah wrote:
> I would suggest that you repeat/follow the same moves to see if you can
> repeat the victory. We need to find:
>
> 1) if Fritz will select a diff. move at one point (in the middle game)
>
> 2) if there is a bug in the eval. function.
>
>
From my experience with Fritz, he never repeats an opening he lost with (he switches
to some other completely different line, and I don't know how to force him to play a
certain opening - maybe someone can teach me, I would appreciate this!)
Valentin
Komputer Korner
an opening book.
--
--
Komputer Korner
The inkompetent komputer
To send email take the 1 out of my address. My email address is
kor...@netcom.ca but take the 1 out before sending the email.
Valentin Barbulescu wrote in message
<35B66C3A...@Barbulescu.com>...
Har...@t-online.de
quoting a mail from Vale...@Barbulescu.com concerning Re: I won against Fritz5 (ELO 2791)! - Here is the game
> > I would suggest that you repeat/follow the same moves to see if you can
> > repeat the victory. We need to find:
> >
> > 1) if Fritz will select a diff. move at one point (in the middle
> > game)
> >
> > 2) if there is a bug in the eval. function.
> >
> >
>
> From my experience with Fritz, he never repeats an opening he lost with (he
> switches to some other completely different line, and I don't know how to
> force him to play a certain opening - maybe someone can teach me, I would
> appreciate this!)
> Valentin
Open the tree (F11) and right click on the move you want Fritz not to
play, choose "no tournament move". That's all.
Harald Faber
MichaelJFox
>
> Open up a new tree and put only that opening in it. Load that tree as
> an opening book.
better, I would think. Anybody got any suggestions?
> Komputer Korner
> The inkompetent komputer
>
> To send email take the 1 out of my address. My email address is
> kor...@netcom.ca but take the 1 out before sending the email.
> Valentin Barbulescu wrote in message
> <35B66C3A...@Barbulescu.com>...
> >
> >
> >Siva Chelliah wrote:
> >