I agree with the author’s critique of maximalist interpretations of the paper. I’ve built a chess engine in the past so I was less hung up on the “without sear

I agree with the author’s critique of maximalist interpretations of the paper. I’ve built a chess engine in the past so I was less hung up on the “without search” component like a lot of the original thread seemed to be. I don’t care if I have to take an argmax over all possible moves in a position and whether that technically qualifies as a search, or if the model may be doing some implicit search within its weights. What matters to me is that it is several orders of magnitude lower search required than stockfish, and those operations cost much less because they’re being executed in parallel on a TPU (GPU also works). Optimizing static evaluation for the hardware we have is still a super interesting topic. It’s arguably why deep learning took off: matrix multiplication happens to be really cheap on GPUs.