Jan Betley comments on Generating the Funniest Joke with RL (according to GPT-4.1)

[Jan Betley]

Cool experiment! Maybe GPT-4.5 would do better?