I was also stunned by how poorly ChatGPT performed at chess.

Well, it’s trained on an enormous blob of human-generated data Including human mistakes, which are probably more common and therefore take up a bigger percentage of its dataset than correct answers

It also lacks the power of deductive reasoning and cannot recognize patterns or model a complex heirarchy of likely outcomes. It's literally just a search engine that can process queries in a conversational tone and synthesize search results into mediocre college student level prose.

I’m especially amused by the people who think that it “knows” what facts are (something humans themselves have arguments about) You *know* a bunch of its responses there are simply whatever text strings were in proximity to text strings like “it’s a fact that” and “the fact of the matter is”