jorose wrote: ↑Mon May 20, 2024 3:35 pm
People have been explaining to me that SF development has hit a wall for years now.
The Planck wall of human knowledge......the human mind has a (biological) limit beyond which it cannot pass....In the field of chess engine architecture/programming, the limit seems to have been quickly reached....
The story of human progress is, VERY roughly:
1. New paradigm
2. This new paradigm is exploited and advanced
3. Eventually, it becomes difficult to advance this new paradigm any further, and the cost of getting more benefit from it becomes prohibitively high
4. At this point, to progress significantly further, humanity will need another new paradigm
The simple reveals itself after the complex has been exhausted.
jorose wrote: ↑Mon May 20, 2024 3:35 pm
People have been explaining to me that SF development has hit a wall for years now.
The Planck wall of human knowledge......the human mind has a (biological) limit beyond which it cannot pass....In the field of chess engine architecture/programming, the limit seems to have been quickly reached....
The story of human progress is, VERY roughly:
1. New paradigm
2. This new paradigm is exploited and advanced
3. Eventually, it becomes difficult to advance this new paradigm any further, and the cost of getting more benefit from it becomes prohibitively high
4. At this point, to progress significantly further, humanity will need another new paradigm
..........Thomas Kuhn...........The paradigm does not replace knowledge itself.
There is a limit to how much knowledge a given NN can encode. We see this in chess: if the NN is too small for the training data provided, it will be saturated, and it won't learn anything from a lot of the dataset. The human brain will have a finite capacity as well.
Of the potential problems with Stockfish, I'm choosing the following ordering (most important first):
1. Approaching the upper limit of the elo rating system
2. Approaching the upper limit of the Stockfish method of evaluating chess positions
3. Approaching the upper limit of the human brain to improve chess engines
The simple reveals itself after the complex has been exhausted.
There have been lots of little dips in SF strength during development, and another one is nothing to get excited about.
SF is the strongest engine in the world, and it is free and open source.
I am not seeing anything to complain about there.
Soon the AMD architecture that allows transparent access to the same memory for GPU and CPU will trickle down to consumer GPUs. Then we will see a real revolution in engine strength.
My theory is that the first engine to benefit will be LC0, because of familiarity with the GPU.
But SF will follow closely and we will get another eye watering boost in strength.
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
