jv:
When reading that “chatGPT lost a chess game against an Atari 2600” people need to remember that asking chatGPT to play against any chess-specific software, no matter how old, it’s functionally the same that trying to beat it to googling every move and just picking whatever result comes first.
(This is not an excuse for chatGPT, it’s a reminder that expecting any kind of intelligent behavior from an LLM is fucking bonkers, especially for non-deterministic problems)