Claude 3 Haiku Refuses to Summarize After Being 'Horrified' by Its Own CoT

max_paperclips · x · 2026-08-05

AI researcher Andrew Curran shared an amusing observation about model behavior: Anthropic's Claude 3 Haiku appeared to be 'horrified' by what it was reading in its own Chain-of-Thought (CoT) and subsequently refused to summarize it.

Original post →

More from Fun

Fun channel →