An actual interesting computer chess read about FF2

[carldaman]

Ever since NNUE was implemented contempt makes little difference for SF. Contempt-less NNUE can be an impediment against winning more with Black vs weaker engines, so it fails to maximize Elo potential.

But even if SF's contempt was fully functional, as it used to be before NNUE, it would be correct to have it on, since the engine has to compete against mostly weaker opponents on the rating lists and competing is about winning the most games.

Of course, this line of thinking presumes the main purpose of an engine is to....'fight it out' against another engine during engine v engine play.
Truly, I think most people just switch on an engine to get evaluation of certain positions. That is something way too many people tend to forget...
[/quote]

My comments were related to the contempt setting in the context of rating lists competition.

Of course, analysis capabilities are very important, and having a functional contempt feature is very valuable there as well. Not just for competition purposes, but also for analysis, I like an engine to exhibit optimism in its eval and avoid drawing lines, things that working contempt can provide.

We gained a lot of raw strength with NNUE, but we lost analysis flexibility since SF NNUE makes little or no use of contempt.

[MikeB]

My comments were related to the contempt setting in the context of rating lists competition.

Of course, analysis capabilities are very important, and having a functional contempt feature is very valuable there as well. Not just for competition purposes, but also for analysis, I like an engine to exhibit optimism in its eval and avoid drawing lines, things that working contempt can provide.

We gained a lot of raw strength with NNUE, but we lost analysis flexibility since SF NNUE makes little or no use of contempt.

For me personally, I have very little use for contempt. The justification for contempt was simply to gain more elo against weaker engines. The default contempt setting was based on the highest level of contempt that does not lose Elo against engines of equal strength, In this case, it is the highest contempt setting that does not lose Elo when SF plays itself ( since there are no other engines in fishtest). In summary They have spent an extraordinary amount of resources on finding the highest contempt setting that does not lose Elo against itself. Meaning it does not help at all in the TCEC or CCC finals. In summary, Contempt avoids playing into a draw , hoping that the weaker engine makes a mistake further down the road. It is not adding value in analysis, but simply artificially inflating the value of its own position. I have seen in cases, where the engines are equal, SF turns a draw in to a lost. It can also turn a draw into a win, but when the engines are equal, they will tend to offset one another But I fully appreciate others may have a difference in opinion which is fine, I'm just stating the counter-point

[Collingwood]

Also I saw another Chessbase page where Albert just about thanked everybody he could , including the LC0 and Stockfish teams, so he is clearly trying to making amends, which is better than what some authors do. No one is perfect, especially me....

You avoid the lies he spoke during the interviews of FF1 and FF2
At this point, his "thanks" are as hollow as an empty can of Budweiser...

Calling someone a liar is very strong language. I find it extremely distasteful.
What, exactly, were his lies?
Quotes, please.
If you cannot produce exact quotes demonstrating your claim, then what is your claim exactly?

There is a big difference between someone who thinks that they have done a great thing whereas others disagree and someone who knows he has not done a great thing and says that he has.

I think, for instance, that calling someone a liar when they are not is easily as deplorable as being a liar.

Tord has already given us quotes. (These are only a fraction of all the deceitful words that have come from Chessbase since FF1.)

Tord Romstad
@tordr
Feb 13
I don't have time to listen to the actual interview, but the introduction is wrong that I just have no words:

Tord Romstad
@tordr
Albert Silver "came across a new neural network technology from Japan"? Seriously?

Tord Romstad
@tordr
And then there is this, in the list of interview questions: "How did you overcome the moment of starting almost completely from scratch with the Fat Fritz 2.0 development?" How on Earth is "starting almost completely from scratch" an accurate description?

Do people really need a deconstruction of these statements explaining why they are lies? Clearly, "starting almost completely from scratch" is a lie. It's a shameful lie even if one uses a totally ridiculous definition of "from scratch".

While unethicial people could claim "came across a new neural network technology from Japan" is literally true (e.g. AS "came across" SF NNUE in the computer chess forums, etc.), it is obviously intended to deceive and is a lie by imputation. The obvious imputation is that AS was responsible for NNUE in computer chess -- another shameful lie.

If you're in any doubt that people were deceived, just read the forums.


I don't know why there are still any arguments about their deceit. If you make excuses for deceit and unethical behavior, you are just encouraging more deceit and unethical behavior.

[carldaman]

My comments were related to the contempt setting in the context of rating lists competition.

Of course, analysis capabilities are very important, and having a functional contempt feature is very valuable there as well. Not just for competition purposes, but also for analysis, I like an engine to exhibit optimism in its eval and avoid drawing lines, things that working contempt can provide.

We gained a lot of raw strength with NNUE, but we lost analysis flexibility since SF NNUE makes little or no use of contempt.

For me personally, I have very little use for contempt. The justification for contempt was simply to gain more elo against weaker engines. The default contempt setting was based on the highest level of contempt that does not lose Elo against engines of equal strength, In this case, it is the highest contempt setting that does not lose Elo when SF plays itself ( since there are no other engines in fishtest). In summary They have spent an extraordinary amount of resources on finding the highest contempt setting that does not lose Elo against itself. Meaning it does not help at all in the TCEC or CCC finals. In summary, Contempt avoids playing into a draw , hoping that the weaker engine makes a mistake further down the road. It is not adding value in analysis, but simply artificially inflating the value of its own position. I have seen in cases, where the engines are equal, SF turns a draw in to a lost. It can also turn a draw into a win, but when the engines are equal, they will tend to offset one another But I fully appreciate others may have a difference in opinion which is fine, I'm just stating the counter-point

Positive contempt helps guide the engine towards playing less conservatively. Very useful when looking for promising lines and ideas in analysis, whether it be OTB prep or corr chess - in my experience. Disclaimer: I do prefer speculative over solid chess.

[Dann Corbit]

Also I saw another Chessbase page where Albert just about thanked everybody he could , including the LC0 and Stockfish teams, so he is clearly trying to making amends, which is better than what some authors do. No one is perfect, especially me....

You avoid the lies he spoke during the interviews of FF1 and FF2
At this point, his "thanks" are as hollow as an empty can of Budweiser...

Calling someone a liar is very strong language. I find it extremely distasteful.
What, exactly, were his lies?
Quotes, please.
If you cannot produce exact quotes demonstrating your claim, then what is your claim exactly?

There is a big difference between someone who thinks that they have done a great thing whereas others disagree and someone who knows he has not done a great thing and says that he has.

I think, for instance, that calling someone a liar when they are not is easily as deplorable as being a liar.

Tord has already given us quotes. (These are only a fraction of all the deceitful words that have come from Chessbase since FF1.)

Tord Romstad
@tordr
Feb 13
I don't have time to listen to the actual interview, but the introduction is wrong that I just have no words:

Tord Romstad
@tordr
Albert Silver "came across a new neural network technology from Japan"? Seriously?

Tord Romstad
@tordr
And then there is this, in the list of interview questions: "How did you overcome the moment of starting almost completely from scratch with the Fat Fritz 2.0 development?" How on Earth is "starting almost completely from scratch" an accurate description?

Do people really need a deconstruction of these statements explaining why they are lies? Clearly, "starting almost completely from scratch" is a lie. It's a shameful lie even if one uses a totally ridiculous definition of "from scratch".

While unethicial people could claim "came across a new neural network technology from Japan" is literally true (e.g. AS "came across" SF NNUE in the computer chess forums, etc.), it is obviously intended to deceive and is a lie by imputation. The obvious imputation is that AS was responsible for NNUE in computer chess -- another shameful lie.

If you're in any doubt that people were deceived, just read the forums.


I don't know why there are still any arguments about their deceit. If you make excuses for deceit and unethical behavior, you are just encouraging more deceit and unethical behavior.

To lie means to say something that you know is not true. There is a form of exaggeration born from pride. For instance, a father may state that his son is great like some professional player because he scored a touchdown in pee-wee football. It isn't true, but the father sees it as such.

Albert spent many thousands of dollars of his own money pursuing this goal and exercising ideas that he considered to be original.

The outrage at Albert seems strangely misplaced to me.
Yes, he is way over the top exuberant about what he has accomplished.
But I do not fault him for that.

[Collingwood]

To lie means to say something that you know is not true. There is a form of exaggeration born from pride.
<snip>
The outrage at Albert seems strangely misplaced to me.
Yes, he is way over the top exuberant about what he has accomplished.
But I do not fault him for that.

There was a post from Modern Times in an earlier thread that was a bit similar to what you say:

Watching that hour long video, rightly or wrongly Albert came across as a bit of an egotist. So I think it is more of a personal crusade for him <snip>

These are essentially the defence of "not guilty by reason of insanity". One really does need to be insane to believe those statements and imputations.

I'm just not convinced that having a huge ego (exacerbated by lack of any relevant education) is enough to qualify as not guilty by reason of insanity (and therefore absolve him of guilt).

[noobpwnftw]

Let me just put it this way, I've spent significantly more time, money and other resources to push computer chess forward, I contributed to both the projects they are ripping off. I also have your so-called ego, so when I see crap like this, it gets personal, will you spare me from your double standards should I seek my "retaliation" in all shapes and forms?

[peter]

In all the lighthearted frivolity about the law, it seems we’ve missed an article on how an ICCF GM evaluates an analysis partner and comes to prefer FF2. Warning: contains actual chess.

https://en.chessbase.com/post/discussin ... al-network

And another one interesting read:

http://www.fastgm.de/16-60-0.6.html

Code: Select all

#	Engine	                          Elo  Error	Games	Points	(%)	W	D	L	D(%)	CFS
1	Fat Fritz 2	                3546	8	2500	1377.00	55.08	319	2116	65	84.6	83
2	Stockfish 13	                3542	8	2500	1360.00	54.40	285	2150	65	86.0	63
3	Stockfish 21013                 3540	8	2500	1363.50	54.54	289	2149	62	86.0	100
4	Stockfish 12	                3518	5	5750	3190.50	55.49	880	4621	249	80.4	59
5	Lc0 0.26.3 384x30-J94-130	3517	8	4000	2763.00	69.08	1673	2180	147	54.5	60
6	Lc0 0.26.3 384x30-J98.1-16	3516	7	4250	2960.50	69.66	1822	2277	151	53.6	60
7	Lc0 0.26.3 384x30-J94-80	3514	7	4500	3163.50	70.30	1996	2335	169	51.9	56
8	Lc0 0.26.3 67630	        3513	9	3250	2278.50	70.11	1444	1669	137	51.4	50
9	Lc0 0.26.3 66750	        3513	8	4250	2950.50	69.42	1811	2279	160	53.6	79
10	Lc0 0.26.3 384x30-J96-28	3509	8	4250	2931.00	68.96	1784	2294	172	54.0	58
11	Lc0 0.26.3 384x30-J92-260	3508	7	5250	3756.50	71.55	2472	2569	209	48.9	61
12	Stockfish 200813 82215d0fd0df	3507	8	2500	1477.50	59.10	541	1873	86	74.9	52
13	Lc0 0.26.3 66625	        3506	7	4250	2918.00	68.66	1765	2306	179	54.3	50
14	Lc0 0.26.3 384x30-J92-330	3506	7	4750	3346.50	70.45	2153	2387	210	50.3	69
15	Dragon by Komodo Chess 1.0	3504	6	4500	2325.50	51.68	479	3693	328	82.1	64
16	Lc0 0.26.3 RC2 384x30-J92-130	3502	8	4000	3051.50	76.29	2190	1723	87	43.1	86
17	Stockfish NNUE SV20200725-1313	3496	8	2500	1475.00	59.00	520	1910	70	76.4	64
18	Lc0 0.26.3 66511	        3493	8	4250	2859.50	67.28	1702	2315	233	54.5	91
19	Lc0 0.26.3 65681	        3486	7	4250	3013.50	70.91	1972	2083	195	49.0	90
20	Lc0 0.25.1 LS 15	        3480	6	6000	4320.00	72.00	2928	2784	288	46.4	84
...
...

[jefk]

four rating points difference ?
that's not statistical significant.

Then about a comment about the open-course community reaction
to FF2, after this article i understood it better :
https://lichess.org/blog/YCvy7xMAACIA80 ... -a-rip-off

yet, it wasn't impressed about comments about 'lying'
here on this forum.

If someone in politics (eg. a former US president) then it's
simply called 'alternative truth' or so; and in marketing, it's a
usual thing, done by marketeers with a proud lack of technical
knowledge. Now AS is obviously also has technical knowledge so
he maybe he should stay away from marketing (*) and ask CB to
this for them by another guy just my 2 cnts. Their GUI isn't bad
but i've seen many engines there after the Frans Morsch engine,
eg. Rybka (anyone heard of that?) Houdini then a Deep Fritz made
by a certain Rajlich (their Mcts search wasn't so bad on long
time scales although totally different then what's now used
by Komodo/Dragon etc. etc. Who cares.
My 2 cnts


PS personally i just had a good laugh about their
fat fritz stories, maybe that's the way to sell it
in Germany, who knows, every year there will be
new kiddies learning chess, and after a while they
like to buy a chess program (paying for it obviously
feels better for their mommie because it then doesn't
have a virus and it has a support mail address. capice ?

(*) small business often is difficult, especially in chess,
ask eg. Ed S (did it pretty well although once also
had to deal with a troll, ended up in courtroom);
somehow the freedom may compensate for it but
usually it's just because such programmers are anyway
to stubborn and/or smart to work in an organization.
(wondering how it's going with Bookup these days,
pretty soon i'm going to upload the awesome BB36 superfast
minimax algorithm (**) as open source (as part of 'bookbuilder)
and then the hackers can also stay away from my comp;
sigh
(**) actually Zipproth's algorithm (cerebellum) is better (although i don't know how fast it is, probably ok with modern fast computers) but then as
small software company they are not so easy to hack i suppose.

[Madeleine Birchfield]

My comments were related to the contempt setting in the context of rating lists competition.

Of course, analysis capabilities are very important, and having a functional contempt feature is very valuable there as well. Not just for competition purposes, but also for analysis, I like an engine to exhibit optimism in its eval and avoid drawing lines, things that working contempt can provide.

We gained a lot of raw strength with NNUE, but we lost analysis flexibility since SF NNUE makes little or no use of contempt.

The Stockfish developers have said that they have gotten contempt to work with NNUE, so contempt 24 should work the same regardless of NNUE.

The main target demographic of the Stockfish engine is not the rating lists, but the human chess player who uses Stockfish with the Chessbase GUI or Lichess Stockfish to help them analyse their games or come up with new ideas in opening theory. Stockfish developers such as Vizvezdenec have claimed that Stockfish is the best engine for analysis, better than Komodo Dragon and the latest Leela, partially due to its evaluation, along with its search. The vast majority of the end users unfortunately have no idea about contempt and analysis contempt causing bias and inaccuracy in Stockfish's evaluations. With Lichess Stockfish, the end user has no ability to change any of the UCI options in Stockfish, preventing them from disabling the biased eval in Stockfish. For those who are aware of contempt, IM Kostya Kavutskiy and other strong players have said that it doesn't help with analysis when Stockfish's eval jumps by half a pawn every move due to contempt, and GM John Hartmann even released his own version of Stockfish without contempt for better analysis with Stockfish. As a result, there is this disconnection between the claims of Stockfish developers and the average Stockfish end user, which would be well served by Stockfish setting default contempt to zero, and leaving it to the player to deviate from an unbiased evaluation in the UCI options if they want lines biased towards the player.

As for the rating lists, I believe they should test the default Stockfish regardless of its contempt settings. If the next version of Stockfish has default contempt 0, then the rating lists whould test with default contempt 0, if default contemot 24 then tested with default contempt 24. But if the default is 0, then trying to alter UCI options to default contempt 24 to benefit Stockfish against weaker opponents is akin to rigging the test towards Stockfish, why stop at contempt when you could also increase hash size or thread number in the UCI options for Stockfish.

Some people have said earlier in thread that Fat Fritz 2 also has contempt 24. If so, then Fat Fritz 2 is even worse than I thought, and this contempt discussion is no longer relavent in any discussion between Stockfish and Fat Fritz 2, because all my criticisms also apply to Fat Fritz 2 as well. It's even worse with Fat Fritz 2, because a contempt 0 patch doesn't have to pass a Fishtest SPRT gauntlet if applied to Fat Fritz 2, as Albert Silver only has to change a single line of code, which he couldn't even implement for the benefit of the end users.