FF2 verses SF, what are the important differences?

[Dann Corbit]

Code wise, the major change is to alter some constants. (See above)
One file has a dimension doubled.
The other file's changes seem purely syntactic sugar.
The rest of it is the training effort.

If you want to make a FF2 like engine that uses the current source, then merge the above changes.
However, the real difference is (I am sure) in the training effort.
There is also no guarantee that the FF2 constants will work well with the new source, since that will change evaluations.

[JohnWoe]

Syntactic sugar...

I'm afraid Stockfish isn't a programming language.

[pohl4711]

"Fat Fritz 2 is trained and developed using this NNUE technology, running inside the open source Stockfish binary, but with a completely new and different neural network that differs significantly from Stockfish's on a couple of key points:
The first is that the evaluations and positions did not come from a brute force engine such as Stockfish or Komodo, but rather came from our original neural network Fat Fritz.
The second key difference is that the neural network in Fat Fritz 2 is double the size of Stockfish 12, using 512 neurons instead of 256, offering the opportunity for double the knowledge and understanding."

[gaard]

Code wise, the major change is to alter some constants. (See above)
One file has a dimension doubled.
The other file's changes seem purely syntactic sugar.
The rest of it is the training effort.

If you want to make a FF2 like engine that uses the current source, then merge the above changes.
However, the real difference is (I am sure) in the training effort.
There is also no guarantee that the FF2 constants will work well with the new source, since that will change evaluations.

One is free and plays at a higher level than the other.
The other is 99 Euros and plays at a lower level than the other and is infringing on the others copyright.

[acepoint_de]

"Fat Fritz 2 is trained and developed using this NNUE technology, running inside the open source Stockfish binary, but with a completely new and different neural network that differs significantly from Stockfish's on a couple of key points:
The first is that the evaluations and positions did not come from a brute force engine such as Stockfish or Komodo, but rather came from our original neural network Fat Fritz.
The second key difference is that the neural network in Fat Fritz 2 is double the size of Stockfish 12, using 512 neurons instead of 256, offering the opportunity for double the knowledge and understanding."

What's the source of your quotation?

At the chessbase shop it now reads:

"Feature list:

• Fat Fritz 2* engine with a massive new neural network

• 64-bit program interface

• Includes current Fritz 17 program interface

• ChessBase Premium Account (6 months) with access to training videos, Playchess, tactics training, and much much more!

• Database with 1 million games, etc.

* Fat Fritz 2 is an original neural network that is powered by a modified version of Stockfish. Stockfish is an open-source project licensed through the GPL v3 with all due rights. The source code of Stockfish and the modifications for Fat Fritz 2 can be found on Github."

In a logical sense the first item says that FF2 is an original NN with a massive new NN .

More and more I'm convinced they didn't and still don't know what the GPL stuff is and how to deal with it

Ciao

acepoint

[hgm]

Well, provided the NNUE weights are included as a separate .bin file, there doesn't seem to be anything legally wrong with this. Or misleading. They use Stockfish in its capacity as a 'NNUE player', for their own NNUE. They clearly mention that Stockfish can be obtained for free, so it should be clear to everyone they charge you only for the NNUE and whatever else they included in the package (GUI, game database).

That they needed a modified version of Stockfish at all, due to the NNUE size doubling might just be a commercial trick. 'Potentially double the knowledge', for sure. But actually? Was the smaller net really saturated? And certainly half the speed, methinks... So it seems to over-emphasize the importance of the NNUE size, probably because they did not want to use the same as everyone else for commercial reasons, and "double the size" goes down well with the naive customer as an explanation why it "must be better" than the smaller nets that can be obtained for free.

What you think of the ethical aspects of this is mostly a matter of personal opinion. Some would say it is a ripoff, cashing on massive efforts the computer-chess community provided for free. Others would say that it is just smart business, and that they are actually doing exactly what the Stockfish developers invited them to do, by releasing Stockfish under GPL, rather than under a more restrictive licence.

[Modern Times]

What you think of the ethical aspects of this is mostly a matter of personal opinion. Some would say it is a ripoff, cashing on massive efforts the computer-chess community provided for free. Others would say that it is just smart business, and that they are actually doing exactly what the Stockfish developers invited them to do, by releasing Stockfish under GPL, rather than under a more restrictive licence.

With the notable exception of Komodo, I think the incredible success of Stockfish has killed off commercial engines. It is no longer viable to invest in the development of new totally original commercial engines because there is no money to be made. You could argue that the incredible success of Stockfish has created the environment where chessbase chose to take the approach they did. (now I will get burned at the stake). So we might well see more of this. What annoys people more probably are the exaggerated marketing claims and the initial startlingly obvious mistake of integrating the NNUE and the executable which was in most people's opinion a breach of the GPL. That has now been rectified, so there should not be any concerns over legality.

Komodo only survives because, my speculation, Larry is retired and doesn't rely on an income from Komodo to put food on the table, and likewise Mark is no spring chicken either and must have other sources of income. When these gentlemen decide to actually retire, I think that will be the last of the great commercial engines.

[hgm]

Well, you never know. There could be technological breakthroughs that make a completely different approach competitive. AlphaZero was such a breakthrough we did not see coming...

Existing engines are a kind of 'one-trick ponies': they excell at tactics, but their knowledge is dismal. AlphaZero could exploit that, by bringing its alpha-beta opponnets outside their comfort zone, and engaging them there. NNUE repairs a bit of that, but it is still a very primitive technique. There is no guarantee at all that other blind spots of the existing alpha-beta + NNUE engines won't be discovered, which will expose them as "not nearly as strong as we thought".

[Modern Times]

Yes, GPU technology is evolving at a rapid rate, we may see some sort of GPU chess implementation re-emerge as the next big step.

[Harvey Williamson]

With the notable exception of Komodo, I think the incredible success of Stockfish has killed off commercial engines.

With regards to selling just an engine you are probably right. I imagine CB only makes a fraction of its money from engine sales. Most of their income probably comes from the other products they sell. To survive now you need to package the engine with something else like a good GUI which Shredder, Hiarcs etc.. do.