Ask Lemmy

It's similar for Chess AI (and by that I mean Chess Engines that have been developed for decades, not LLMs) except that the engines will play extremely principled and cautious Chess, right up until it gets out of the book/theory territory and it calculates 20-30 moves into the future and determines that an incredibly unintuitive and bizarre move is the "best" position, but even the top grandmasters would likely never find that specific move that the engine determined was the best move you could make.

In some cases, you can tell when someone is cheating by seeing them make top engine moves that don't do anything or really advance the board state in a meaningful way the turn you make them because it forces the opponent to make one or more less optimal moves that the player could capitalize on, but humans are terrible and seeing these patterns because they are super analytical and require precise calculation. Also, asking them to explain why they made those moves is another way to catch them in the act - and a subject of a great deal of controversy in modern chess tournaments when some players give less than convincing answers when pressed on why they made certain moves in the game.