Here is how gnubg cheats.

2328 views
Skip to first unread message

muratk

unread,
Apr 5, 2012, 6:56:43 AM4/5/12
to
Let me warn you that this will be necessarily a long article but if
you take time to read it, you probably will learn something from it.

The reason I am being humble enough to say "probably" instead of
saying "for sure", is that I have no idea about the brain capacity of
all who will be reading it.

But don't you worry. I will try to spell it all out for you as much as
I can. So, let's get started by looking at a position.

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

GNU Backgammon Position ID: u3PDAADvtgHAAA
Match ID : MIHxAAAAAAAE

+-1--2--3--4--5--6-------7--8--9-10-11-12-+ O: gnubg
| O O O O O | | | 0 points
| O O O O O | | | Rolled 34
| O O | | |
| O | | |
| | | |
| |BAR| |^ 7 point match (Cube:
1)
| | | |
| | | |
| X X X | | |
| X X X X X | | O X |
| X X X X X | | O X | 0 points
+24-23-22-21-20-19------18-17-16-15-14-13-+ X: Murat7F01

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

In this positin gnugb played 18/11 which raised my eyebrow. This is
the kind of position that I have been talking about when claiming that
I can predict future dice rolls playing against gnubg and other bots.

At this point let's tie a couple of loose ends so that you can focus
better. Whether admitted or not, all bg bots today are mutated
offsprings of jellyfish.

Jellyfish was cheating. When I reported in this forum that it was
rolling 77's, the dice rolling bug was fixed but at the same time
jellyfish started to play differently. Thus, the link between how
jellyfish rolled dice and how it played was established. After that,
the guy who was peddling jellyfish disappeared from the face of the
planet and jellyfish has been flushed down into the backgammon septic
tank.

And you may be wondering why I wanted to play cubeless 7-point matches
to begin with? Well, I was not only just upset but also curious about
not being able to play cubeless backgammon (i.e. "the real thing"
before it was bastardized by sick gambler Americans) using commercial
bots like extereme gammon. So, I decided to see if I could find any
answers by playing cubeless 7-point matches against gnubg.

I downloaded 10x10,000 rolls from the random.org, saved them in 10
different files, poured myself a cup of tea, got comfortable in my
chair and started to play, intending to play 10 matches using each
file, for a total of 100 matches, to observe what would happen.

The above position arose at the 14th move, of the first game, of the
first match. An unfortunately early lucky strike, you may say... :(

After gnubg played 18/11, I rolled 32 and the correct play for me was
to hit, after which gnubg would roll a decisive joker 64!

Now, after rasining my eyebrow about gnubg playing 18/11, I would have
done one or both of two things.

If I had bet money on winning the game/match, I would play my 32
"incorrectly" (i.e. make an inferior move, just like gnubg has made)
in order to avoid gnubg's following 64 joker.

If I was betting money to predict future rolls, I would bet a
reasonable amount that I could afford to lose, let's say $100, on the
2 in 36 odds that it would get a 64... That is your $1,700 against my
$100. After I hit you with a few of these, you all would start to shit
your pants... ;)

All a person needs to predict future dice rolls playing against bots
is to be a good enough backgammon player to detect such subtle
irregularities be able to ask oneself "why has the bot played like
this" in any similar position.

According to the hint, 18/11 is the right move with 46.91% vs. the
second best move 5/2 5/1 with 46.86%. A mere 0.05% difference. And if
you roll out, the difference shrinks to 0.03%, between 46.85% and
46.82%.

But I am still not convinced. So, I take a closer look at the default
grandmaster settings. Under the advanced options, there is a box
checked, that says "for cubeful evaluations, use cubeful checker
evaluation".

Because I am playing cubeless 7-point matches, it sounds like an
irrelevant setting in this case but I check it off in both the payer
and the analysis and then roll out again, just to see.

Holy macaroney! The formerly second best move 5/2 5/1 is now superior
to 18/11 by 46.78% vs. 46.71%, by 0.07% which is a bigger difference
than any of the above.

Okay, well, let's leave these settings alone and start replaying from
the beginning. Curiously, on its 12th move, (one before the above
position), gnubg plays 18/7 instead of 7/1 6/1.

The hint says it's the right move with 50.08% vs. 50.01%. Although by
a smaller margin, the roll outs validate it with 50.18% vs. 50.14%.

Now, lets revert that one settings and roll out again. Sure enough,
the second best move reverts to being the best move with 50.06% vs.
49.95%.

So, what can we conclude from all this?

First, the bots don't seem to know how to play without the cube. My
personal observation is that the dice seems more realistic playing
without the cube, which may just be my biased impression. But I do
better against gnubg without the cube, which is a tangible evidence.

And, come to think of it, bots like "extreme garbage" seem to not even
offer the option of cubeless play because they probably don't know how
to handle the above discrepancy in their colorful pie chart rendition
of meaningless analyses...

Which leads to my second conclusion, that it doesn't matter to me
either way, simply because I look at the cube as a tool for impatient
sick gamblers to raise the stakes and expedite their wins or losses...

Consequently, any analysis that is based on the so-called cube skill
is nothing more than plain bullshit. In fact, a pile of bullshit that
one can use to his advantage when playing against these bots and/or
betting money on predicting future dice rolls (regardless of how the
dice is rolled).

In the past, I had proposed that the "doubling window" gave the bots
the wiggle room they needed to cheat (as I define "cheat"). Now, I
have the proof.

From now on, in any betting on predicting future dice rolls, I will
insists that we use the default settings (i.e. leave the box "for
cubeful evaluations, use cubeful checker evaluation" checked even when
playing cubeless matches.

I suppose I should uncheck it if I want to play truly cubeless matches
against gnubg but even then, who knows what other bullshit settings
effect its other bullshit anayses...? It may be just best to ignore
them all.

My advice to people who are just learning backgammon is to learn to
play it without the cube first. Learn the real game first and then if
you want to become a sick gambler, you can always use the cube to jack
up the stakes, in order to win or lose faster...! :)) But, learn to
have to patiently play out a game to the end without the cube, and
thus improve your staying power.

Playing backgammon as it has been for millenia without the cube is
like making slow love to your wife.

Playing backgammon with the cube is like fast fucking a whore, with
much of what you pay going to the book publishing, bot peddling,
tournament organizing pimps of backgammon gambling......

MK
Message has been deleted

Stick

unread,
Apr 5, 2012, 12:00:53 PM4/5/12
to

You can play cubeless matches with XG.

Stick

muratk

unread,
Apr 10, 2012, 3:13:24 AM4/10/12
to
On Apr 5, 10:00 am, Stick <checkmug...@yahoo.com> wrote:


> You can play cubeless matches with XG.


Did I confuse it with snowie then? One of those bots doesn't have
the option to play cubeless matches, no?

Anyway, I don't have the patience to play against bots with manual
dice and XG is a hilarious piece of shit regardless of what built
in dice it uses.

And yes, my offer to bet money on predicting XG's or any other bot's
built in dice rolls is still valid...

MK

LENUS

unread,
Apr 14, 2012, 3:29:09 AM4/14/12
to
On Apr 10, 11:13 am, muratk <mu...@compuplus.net> wrote:
>
>
> And yes, my offer to bet money on predicting XG's or any other bot's
> built in dice rolls is still valid...
>
> MK

Murat,

You are offering nothing unusual or surprising, chances of rolling 64
(or any other non-double) REALLY IS 2 out of 36, if you care to make
adjustment to the bet terms, let us know.

muratk

unread,
Apr 15, 2012, 4:37:45 AM4/15/12
to
On Apr 14, 1:29 am, LENUS <lenus...@yahoo.com> wrote:
.
.
> You are offering nothing unusual or surprising, chances
> of rolling 64 (or any other non-double) REALLY IS 2 out
> of 36, if you care to make adjustment to the bet terms,
> let us know.
.
.
Lenus, I think you are fooled be the ilks of Tim Chow to
reduce backgammon to the level of tossing coins...

We are not talking about how often I can predict a 64
while simply rolling dice but we are talking about the
likelihood of one side or another rolling a 64 in a very
specific position during a game of backgammon...

Do you understand the difference...?

Please answer yes or no before you write anything more...

Thanks,

MK



valerie...@gmail.com

unread,
Mar 18, 2015, 10:34:43 AM3/18/15
to
Wanker
Message has been deleted

jrobin...@gmail.com

unread,
Jun 23, 2015, 12:46:20 PM6/23/15
to
Playing against a bot that is excessively 'fortunate' with the dice it rolls is really bad for beginners because it teaches them to be over cautious.

GNU backgammon is just as irritating as Jellyfish, a pox on both their houses!

thefishd...@gmail.com

unread,
Jun 24, 2019, 5:26:07 PM6/24/19
to
Definitely cheats.

I had a situation where I had 3 points in my home base. Gnu backgammon had 2 pieces in hand. It doubled??? And then rolled double 3. How did it know it was going to roll double 3 to get both men back on board and then manage to hit one of my men?

bgbl...@googlemail.com

unread,
Jul 27, 2019, 6:44:51 PM7/27/19
to
Am Montag, 24. Juni 2019 23:26:07 UTC+2 schrieb thefish...@gmail.com:
> Definitely cheats.
>
> I had a situation where I had 3 points in my home base. Gnu backgammon had 2 pieces in hand. It doubled??? And then rolled double 3. How did it know it was going to roll double 3 to get both men back on board and then manage to hit one of my men?

Maybe because you were primed and it was pretty irrelevant whether the bot comes in or not?
Without posting a positions complaints are pointless.

Simply set up the position again, change the seed of the random number generator and check whether the bot doubles again. do this some dozen times. Complain and post the position then if the bot behaved not always the same. This might be something worth to investigate....

bgbl...@googlemail.com

unread,
Jul 27, 2019, 6:48:05 PM7/27/19
to
Am Dienstag, 23. Juni 2015 18:46:20 UTC+2 schrieb jrobin...@gmail.com:
> Playing against a bot that is excessively 'fortunate' with the dice it rolls is really bad for beginners because it teaches them to be over cautious.

It might be surprising to you: to move the checkers in a way that more rolls work well, aka are lucky, is the very essence of the game.

It is really difficult to make a bot play weak like a human. Do you believe that such unnatural play is better for beginners?

IMHO having real time feedback on the dice quality is a good trainer and not playing weak,

kevin arbuthnot

unread,
Feb 2, 2021, 8:30:04 AMFeb 2
to
On Saturday, 27 July 2019 at 23:48:05 UTC+1, bgbl...@googlemail.com wrote:
> Am Dienstag, 23. Juni 2015 18:46:20 UTC+2 schrieb jrobin...@gmail.com:

> It might be surprising to you: to move the checkers in a way that more rolls work well, aka are lucky, is the very essence of the game.
>
> It is really difficult to make a bot play weak like a human. Do you believe that such unnatural play is better for beginners?

Although I've been playing for a few decades I don't rate myself as great, so when I first started using Gnu a few years ago, I wasn't immediately shocked to be losing a lot when I set Gnu to stronger levels, as I wanted to learn. After many games, however, I did start to be surprised by a few things.
First, if I set it at a low level in the early days, just to get used to it, eg, beginner or casual player, it would do stupid things, like accepting doubles from clearly losing positions, or offer doubles from the same positions, or make blatantly stupid moves in order to lose.
Second, when I set it to middling levels it would play well and all would seem well, until you accepted a double that it "thought" you shouldn't have accepted, eg, a 20% chance of winning, even though I had a stronger layout, and Gnu could only get ahead by throwing jokers. Well that happened a real lot. It was almost trying to dissuade you from taking any risk and punishing you for a poor decision. I wasn't sure how much it was being "lucky", so played 2000+ games and recorded the outcomes. My recent analysis of the results suggests that if I accept a double in these circumstances, ie when it would recommend that I do not accept, from that point onwards, Gnu will average 13.6 pips per throw, and I will average 7.4. That's a bit spooky, innit?
Using Gnu's own analysis tool, about 4 times out of 5 Gnu enjoys more "luck" than I did, often in the 20% range!
I also think that the analysis tool should show a total pip count thrown by each player for each game; that reveals a lot.

Nasti Chestikov

unread,
Feb 2, 2021, 12:31:58 PMFeb 2
to
There seem to be a lot of these old posts being resurrected recently.

Having said that, when I can compile the source code back to the same filesize / hashsum as the currently live version then I'll take GNUDung as being honest.

But I can't. So, it sits in my "this piece of crap cheats" drawer.

Thunderground Samurai

unread,
Aug 31, 2021, 1:23:53 PMAug 31
to
GNU backgammon is a good teacher, but there is no question at all that the program 'knows' exactly what doubles it needs to throw and when. This programs often throws three doubles in a row, or three out of four, to turn a losing position into a winning one. This happens far too often to be random. I also question the program's evaluation of luck and skill. One can beat the program 28-0, and get a low score on skill with no luck.
Even so, I use it to improve my play, and simply restart the game if the program goes on one it's magical doubles sequences.

Nasti Chestikov

unread,
Sep 14, 2021, 12:15:47 PMSep 14
to
On Tuesday, 31 August 2021 at 18:23:53 UTC+1, Thunderground Samurai wrote:

> GNU backgammon is a good teacher, but there is no question at all that the program 'knows' exactly what doubles it needs to throw and when. This programs often throws three doubles in a row, or three out of four, > to turn a losing position into a winning one. This happens far too often to be random. I also question the program's evaluation of luck and skill. One can beat the program 28-0, and get a low score on skill with no > > > luck.
> Even so, I use it to improve my play, and simply restart the game if the program goes on one it's magical doubles sequences.

From your main menu: View <-> Panels <-> Command

It will pop up a dialogue box in the bottom right hand corner of your screen. Type "show RNG" and then "show seed". That gets you the parameters used for the dice.

Save down your game and then close down GNU Backgammon. Upon restarting, choose those parameters for the dice (before rolling anything). You can then see if the dice are the same as you play differently.

Frank Berger

unread,
Sep 22, 2021, 3:46:47 PMSep 22
to
Thunderground Samurai schrieb am Dienstag, 31. August 2021 um 19:23:53 UTC+2:

> GNU backgammon is a good teacher, but there is no question at all that the program 'knows' exactly what doubles it needs to throw and when. This programs often throws three doubles in a row, or three out of four, to turn a losing position into a winning one. This happens far too often to be random. I also question the program's evaluation of luck and skill. One can beat the program 28-0, and get a low score on skill with no luck.
> Even so, I use it to improve my play, and simply restart the game if the program goes on one it's magical doubles sequences.

That's the long awaited proof. 3 doubles in a row or 3 out of 4 rools. In the 40 years I play backgammon I have never seen this in RL. Never. What clever bastards that programmers are, providing the source code so that everyone can check the code.
Reply all
Reply to author
Forward
0 new messages