30 Weird Chess Algorithms: Elo World
suckerpinchYouTube ↗
Further reading 2
Did the game agent see too little or optimize the wrong thing? →A limited input can hide distinctions needed for a decision. A misleading objective can reward a bad move even when the state is visible.What cannot be recovered from the current observation? →Ask which distinctions the input representation discards. Earlier observations, feedback or prior knowledge may change what can be inferred.
Caption notes
Compare chess agents that see occupied squares but cannot directly distinguish pieces. Games and Elo ratings test heuristics, search and learned board reconstruction.
Original passages · 6
it could learn a lot about the position just by trying out movesthis has to have some mistakes since the representation is ambiguousthis player performs significantly better than random due to its knowledge of openings but also significantly worse than the blinded playersmaking these engines better isn't gonna make them beat lousy any more often than alwaysthe computer gets to see where the pieces arethe neural network predicts a board that's invalid