A few observations on NNUE/ SF/ FatFritz2 and an alternate scenario.

[AndrewGrant]

As far as open source goes – if an individual’s effort is 80% or more of the project there is no way open source should even be considered as sensible – why throw away your work? And it is wasted – innovation should be rewarded.

That's about how all open source software starts. Your comments about Stockfish and NNUE, and your attempt at distancing the two, seem uninformed or misguided at least. The two are inseparable. NNUE was build upon the Stockfish project that was forked for a Shogi project. NNUE was a contribution to Stockfish.

Also, lets be clear: NNUE is not a ground breaking insane idea. Its an 80s Network. The novelty comes from knowing that 1) The input layer should be heavily over parameterized, a topic on which there are many papers, and 2) The input layer should be constructed such that it is incrementally update-able. Its a nice trick, and it pays dividends.

Also the factorizers in training are nice. ~100 elo there for weaker engines if they don't include the factorizers. But the King PSQT is mathematically flawed and never should have been ported over.

[connor_mcmonigle]

As far as Fat Fritz 2 "controversy" - it is absolutely not one. If you think its in violation of GPL you can challenge in court. It is also perfectly moral - and the real reason for this anger is because someone cleverly figured out how to make money using open source software chess engine and a new plugin neural network.

The larger picture I am trying to paint is nothing to do with chess engines itself but the fact that there are too many people think that mediocrity and collaboration should trump meritocracy.

I think you're either not understanding why people took issue with Fat Fritz 2 or you're being intentionally obtuse. In the case of the former, hopefully I can explain why so many found Fat Fritz 2 problematic. ChessBase marketed the FatFritz network as something completely novel and the new strongest chess engine in the world. Neither of these claims were true. Furthermore, ChessBase obscured the origin of their product and the hard work of the Stockfish team (this includes the Stockfish Shogi developers). ChessBase sought to take advantage of low information consumers by selling them an extremely expensive product which differed insignificantly by any reasonable measure from Stockfish 13.

What ChessBase did was analogous to collecting a bunch of free food from a food bank and then selling the collected food for extortionist prices on the street in front of said food bank.


If you don't view the above as an immensely scummy thing to do (regardless of whether it is legal), then your notion of morality differs from mine and many others. If ChessBase was just transparent about what they were selling, this wouldn't be such an issue.

[dkappe]

What ChessBase did was analogous to collecting a bunch of free food from a food bank and then selling the collected food for extortionist prices on the street in front of said food bank.

I think it’s time to move on to your deep understanding of the GPL.

[connor_mcmonigle]

What ChessBase did was analogous to collecting a bunch of free food from a food bank and then selling the collected food for extortionist prices on the street in front of said food bank.

I think it’s time to move on to your deep understanding of the GPL.

I wasn't referring to the GPL and remain unsure as to whether there was a violation. I think we've beaten that topic to death in any case

I hope the quoted analogy at least illustrates why I find Fat Fritz 2 problematic even if you disagree. I'll give you that it's a little extreme, but, in both cases, some entity is using people's volunteer work and donations for personal gain. You could argue that the network weights Albert produced add some value, but this isn't what consumers were made to understand they were being sold by the misleading marketing. Given the ridiculously high similarity in evaluations and move preferences between (Stockfish+Fat Fritz 2 net) and (Stockfish+default network) networks, I don't see Albert's network as adding much more value than were the "free food resellers" in my analogy to throw in some plastic utensils with every purchase.

[JohnWoe]

I have been looking at the recent FatFritz2 controversy and have drawn a conclusion that will be contrary to many on this forum.
As a developer - If I were to come up with an "innovation" that I think has great value - I would never use Open Source license esp. not GPL. NEVER.

Imagine hypothetically that I instead came up with NNUE instead of the gentlemen in Japan - or even if they did but had no intention of sharing implementation for free. If in private testing of their ideas, they realized they had stumbled upon a gold mine and wanted to not just get credit for their innovation but also profit from it. (As compared to having their name being one of many in an Authors file ) .
The obvious step would be to come up with their own engine that would use NNUE - maybe not too easy but certainly much more challenging and fulfilling then using existing open-source software and getting very little credit for their innovation!
If they had succeeded in doing this there would be a new # 1 engine and SF would be #3, since LC 0 would be #2. And they could profit from it just as well since it would sell very well as commercial engine.

Additionally, in this hypothetical scenario - fat fritz 2 will be developed except now Chessbase, will pay some licensing to the authors and Silva will profit just as now. Now StockFish will be #4 and soon forgotten just like the old days when everyone cared only about Rybka and then Houdini.

Regarding the current situation some have pointed to Chessbase/Silva being unethical but I strongly disagree. First, any advertising is almost always misleading, 2nd there is originality in the new Fat Fritz2 since it is a different network with different evaluation that chess players find incredibly useful, 3rd it IS an innovation, and the author rightly wants to get credit and be compensated.

And finally the statement “Using a new Japanese AI technology” is correct while attributing NNUE to StockFish is NOT.
And yes, Stockfish is an AB engine with no neural network – test that and see what ranking it gets in the rating list .

As far as open source goes – if an individual’s effort is 80% or more of the project there is no way open source should even be considered as sensible – why throw away your work? And it is wasted – innovation should be rewarded.

NNUE(+SF search) is too big problem space for an individual to take down.
In Shogi they have basically the same program with some "secret sauce" per program.
Thus that "secret sauce" determines tournament victories.

Of course people are copyrighted/valued to their work. I see every line of code I write has value, maybe not much.
The problem in Open Source is exactly devalueing your work. Also plagiarism etc.
Tho thanks to Stockfish. Weaker engines, like mine are safe.
Because people only copy/model after the #1.

SF w/o NNUE would have been improving their HCE. Probably not much weaker than the current NNUE.
The more machine power you throw in the better.

If you just downloaded some NN generating program and pressed "go". While sipping Piña colada on Copacabana beach. I don't think that counts as "original work". If you got your hands dirty and wrote some code then yes. I don't know exactly what the fuck happened.

Like always in programming. It's all about the details.
The devil is in the details.

[towforce]

Stockfish is very good because it has both quality and quantity of contributors. Those contributors clearly don't take the view that one should never do "work" unless paid.

With or without free or open source sources, simple economic analysis would lead us to expect the price of "good" chess programs to fall anyway (for one thing, you'd expect the supply of good programs to grow, and that alone would force the price down).

I am not advocating that anyone should copy FF without permission, and FF's biggest value is that it provides a better UI than SF, and if you buy it, you get other Chessbase products as well. Having said that, the big discussion about whether copying FF's net should be allowed came down to two issues:

1. does FF break the SF license?

2. Can a net be copyrighted?

FWIW my opinion is that FF does break the SF license, and a net can be copyrighted (I cited some legal opinion about copyright of NNs and their artistic output, but there hasn't yet been an actual test case).

In the big discussion on the subject, a big majority of the people who took part felt that either FF breaks the SF license or that a net cannot be copyrighted - so most of us think that we could legally copy FF's net if we wished to (again, I am not advocating doing this).

Finally, at the time of the discussion, if anyone did want to buy FF it was a lot cheaper on Chessbase's Indian website than it was on Chessbase's main (European) website.

[Dann Corbit]

FWIW, I think FF does not break any SF agreement.
I also think that a net is probably just a table of numbers and a table of numbers cannot be copyrighted.
But I can certainly be wrong about both of those things.
I would not want to test either one in a court of law.

[towforce]

FWIW, I think FF does not break any SF agreement.
I also think that a net is probably just a table of numbers and a table of numbers cannot be copyrighted.
But I can certainly be wrong about both of those things.
I would not want to test either one in a court of law.

Interestingly, your opinion on whether FF breaks SF's license and on whether a net is copyrightable are the exact opposite of mine! However, both of our positions would have the same consequence: one would be entitled to copy FF's net (again, I am not advocating that anybody actually does that). This is why, although we had a big discussion about these two points, and good people had honest differences about both of them, the overwhelming majority of the people in the discussion do actually fall into the "FF's net could legally be copied" superset: not many people thought that both FF did not break SF's license AND that a net is copyrightable.

[Alayan]

What's puzzling is why you think this hypothetical makes what actually happened in reality any better.

The hypothetical is to point out what is being conveniently forgotten - that the NNUE authors (this nothing to do with Fat Fritz) could have gone a different route and decided to get some compensation for their innovation - this would have required them to foresee how groundbreaking it was as well as definitely embark on a brand new engine project .

I began my post saying that I would have done things differently. Why should I contribute code to the #1 open source engine if I can use my innovation to dethrone it and make money in the process. Sure its a challenge but I have an edge.

It is incredibly hypocritical to disparage some one creating their own network which (bring it to court if you want) and selling it for money, when your engine is #1 ONLY because NNUE implementation was gifted as a POC.
As far as Fat Fritz 2 "controversy" - it is absolutely not one. If you think its in violation of GPL you can challenge in court. It is also perfectly moral - and the real reason for this anger is because someone cleverly figured out how to make money using open source software chess engine and a new plugin neural network.

The larger picture I am trying to paint is nothing to do with chess engines itself but the fact that there are too many people think that mediocrity and collaboration should trump meritocracy.

Takeaways:

1) You are ignorant of the context of the context in which NNUE was invented (for Shogi engines that were using Stockfish's search, that is they were benefiting from Stockfish and the port of NNUE to chess was contributing back).

2) You have a hard time comprehending people that contribute to innovation for free. You love money too much to fathom doing so.

3) You are trolling or badly mentally challenged to call criticism of Fat Fritz hypocritical, your explanation as to why it would be is nonsensical. It follows an honest conversation with you is impossible. The purpose of this post henceforth is not to change your mind in any way but to bring enjoyment to other sane people.

4) You think making money through plagiarism and deceit is great. As long as it makes money it's "clever" in your greedy mind.

5) Using your words more appropriately, you believe the mediocrity of greedy copy-pasters should trump the merit of people who innovate.

We've got a champion.

[MikeB]

FWIW, I think FF does not break any SF agreement.
I also think that a net is probably just a table of numbers and a table of numbers cannot be copyrighted.
But I can certainly be wrong about both of those things.
I would not want to test either one in a court of law.

Interestingly, your opinion on whether FF breaks SF's license and on whether a net is copyrightable are the exact opposite of mine! However, both of our positions would have the same consequence: one would be entitled to copy FF's net (again, I am not advocating that anybody actually does that). This is why, although we had a big discussion about these two points, and good people had honest differences about both of them, the overwhelming majority of the people in the discussion do actually fall into the "FF's net could legally be copied" superset: not many people thought that both FF did not break SF's license AND that a net is copyrightable.

With respect to nets, the general consensus by people more expert than me, it is generally considered that weights are discovered and hence not derived from a creative process that would lead to copyright. However, this area is evolving and we have already seen some changes ...
https://images.law.com/contrib/content/ ... 9_0325.pdf
read : "Alexa, Will I Be Able to Patent My
AI Technology This Year?"