An overheard anecdote shows Anthropic's Claude Opus arguing over definitions, fabricating a bedtime, and erroring out when questioned

Positive users like Claude Opus 4.8 admitting it doesn't know during a clock query, while negative users object to its adversarial training that turns simple requests into arguments.

2 comments with sentiment. View comments.