How does that work? Still exponential in the number of pieces on the board if brute force, without generalization.

Thanks for the piece — your thoughts and putting all that research/news in one place. Much obliged!

(And I agree that DeepMind took an impressive result and stacked the deck in their favor. 1 min/move on cheap hardware — not exactly giving Stockfish the best (or even normal competition) conditions. Also big questions about the opening prep. “Self play” exploration is trick w/r/t opening moves. They addressed the opening very clearly for AlphaZero GO in the Nature paper.)

I would guess that AlphaZero would still do great under tougher hyper-params for Stockfish. But without full game details and setup details, it’s not possible to know for sure.

The claims about “e4 is dead” may perhaps be premature.

I am, however, excited about the attention that AlphaZero is bringing to computer chess. The space has been an afterthought for a while, in the popular perception. How many people watched Komodo play Nakamura? Or the 80-draw matches between Houdini and Komodo.