Hacker News Reader

Oh god that's genuinely way more amusing than I thought llm systems were capable of.

The more I use LLMs the more I have actually gravitated to looking at the humor of LLMs as a imperfect proxy measure of "intelligence".

Obviously this is problematic, but Claude 3.5 (and now 3.7) have been genuinely funny and consistently funny.

This is actually by far the best example of humor by an LLM I’ve ever seen