I decided to challenge the Claude LLM to a game of Connect Four. A painful half hour later, Claude was easily defeated. The hard part was Claude’s inability to operate a coherent model of the game. Placing my moves in the wrong columns, placing extra pieces in a turn, many attempts to violate gravity, and many assertions that I had 3 (or even 4) in a row where I did not. I took to giving it the correct board time and again since it couldn’t manage to keep it together despite many attempts. 😂
