I think it is the case that when the deep learning approach Stockfish used became superhuman it very quickly became dramatically superhuman within a few years/months despite years of earlier work and slow growth.
Are you thinking of Alpha Chess Zero? Stockfish didn’t have anything to do with deep learning until they started using NNUE evaluation (which currently uses Leela Chess Zero training data).
Are you thinking of Alpha Chess Zero? Stockfish didn’t have anything to do with deep learning until they started using NNUE evaluation (which currently uses Leela Chess Zero training data).