Further, the prompt also says to "choose the next move" instead of the best move.

It would be fairly hilarious if the reinforcement training has made the LLM unwilling to make the human feel bad through losing a game.