Putting a chess piece on the board is harder for an AI than winning a game of chess
Children are often better at doing the intermedial stuff than they are at the cognitive stuff at first, and this is where the LLMs succeed and also fail.
Do we want to put in charge a system that knows how to play chess or that understand why chess is played to begin with?
makes me think of Jonathan Plucker and his sociocultural context for creativity.