The Next Big Thing in Computer Chess?

[smatovic]

....maybe to summarize (biased) what people mentioned in this thread as next thing:

- PAB/NNOM - Predicator AlphaBeta, Move Ordering Neural Networks, a neural net driven AB search
- Super-AI - an AI coding a chess engine and outperforming humans in this task
- new programming paradigm in chess, to utilize CPU+SIMD+GPU+TPU+HBM via unified memory
- chess variants, move on to other, non-drawish, chess variants
- incremental progress, there is enough to gain, via incremental progress of software+hardware
- 8-men EGTB
- GUI, advances in GUIs for the user
- human-opponent-play, develop chess engines to maximize user experience
- God's Algorithm for Chess, find an algorithm which plays perfect chess
- knowledge base vs. neural networks, store chess knowledge in graphs instead of neural networks

--
Srdja

Straggler...

- use TPU in CPU for neural network evaluation

Buzzword "AI PC", running neural networks on CPUs - how to use the TPU/NPU/Neural Engine/MMAC unit in CPUs for CNNs as in Lc0 or NNUE as in Stockfish. There is Intel AMX, AMD XDNA, Apple Neural Engine...idk how this will compete against running neural networks via SIMD in future.

--
Srdja

[AlexChess]

- Super-AI - an AI coding a chess engine and outperforming humans in this task
- new programming paradigm in chess, to utilize CPU+SIMD+GPU+TPU+HBM via unified memory

--
Srdja

AI for optimized code generation is a great idea. I hope that Stockfish devs are considering that for next versions, since they are also very focused on code writing good practices. A lot of patches are related to mantain the code to fix style glitches

[towforce]

AI for optimized code generation is a great idea. I hope that Stockfish devs are considering that for next versions, since they are also very focused on code writing good practices. A lot of patches are related to maintain the code to fix style glitches

Hmmm... for me there are two issues:

1. Fast code
2. Readable/maintainable code

In business, it's usually of more value for code to be maintainable than to be blazingly fast. I completely accept that in a game in which "number of loops completed" is a critical success factor, that code inside these loops needs to be optimised for speed. In an ideal world, such code would be in methods/functions in which it is completely clear what the function does, and with as many comments as necessary - EVEN IF THIS MEANS MORE THAN ONE COMMENT PER LINE OF CODE - to enable another developer to maintain the code.

So - is your next big thing human readable code, or fast code that can still be understood by other developers?

Or maybe what we want is for the generative AI to come up with new ways to "hand code" (meaning "not machine learning" - generative AI systems don't have hands!) chess knowledge? If so, this might be a dead end because over the decades prior to ML in chess programming, lots of people tried to hand code chess knowledge - and many of these programs are great fun to play against, but somehow they've managed to miss a lot of important chess knowledge: we know this because NNs are much slower than hand-coded evaluations, but get much better results (many people like their playing style as well).

[Hai]

....maybe to summarize (biased) what people mentioned in this thread as next thing:

- PAB/NNOM - Predicator AlphaBeta, Move Ordering Neural Networks, a neural net driven AB search
- Super-AI - an AI coding a chess engine and outperforming humans in this task
- new programming paradigm in chess, to utilize CPU+SIMD+GPU+TPU+HBM via unified memory
- 8-men EGTB
--
Srdja

And Apples neural engine.
I've found: https://forums.macrumors.com/threads/wh ... 4.2398264/