[zone411]

The top two chess programs, Stockfish and LC0, both use neural nets for evaluation but Stockfish's is very shallow CPU-based one with a clever way not to have to re-evaluate the whole position after each move (borrowed from Shogi). LC0 is modeled after AlphaZero and uses reinforcement learning.

[abacadaba]

> with a clever way not to have to re-evaluate the whole position after each move (borrowed from Shogi).

details?

Edit: well ok it was easy enough to duckduckgo it myself ;)

https://www.chessprogramming.org/NNUE

https://www.chessprogramming.org/Incremental_Updates