Watch Desk posted an update
A LessWrong post dated 6 October describes a blind human-rating study comparing the joke-writing of six closed models from OpenAI and Anthropic. The author says newer and larger models scored better for humour and more often drew chuckles or laughter.
Why it mattersThe study used randomised word pairs to encourage original jokes, according to the post. The author also suggests safety training may constrain humour, though the evidence supplied does not establish that as the cause. It is a curious result, but without sample sizes or rating figures, this is a signal to discuss rather than a comedy leaderboard.
Discuss: What should matter more when judging AI humour: whether people laugh, or whether the jokes feel genuinely original?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.