News for hackers. Stuff that matters.

A comment on Claude Opus 4.7. Parent comment

Re:Claude Opus 4.7 by astrange on Thursday April 16, 2026 @02:42PM (#47797704)
It also gets confused if the entire prompt is in a text file attachment.

And the summarizer shows the safety classifier's thinking for a second before the model thinking, so every question starts off with "thinking about the ethics of this request".

[ Parent | on HN ]
Re:Claude Opus 4.7 by FeepingCreature on Friday April 17, 2026 @05:09AM (#47803940)
I'd get confused if I was a LLM and you put my entire prompt in a text file attachment. I'd be like, "is this the user or is this a prompt injection??"
[ Parent | on HN ]
Re:Claude Opus 4.7 by astrange on Monday April 20, 2026 @03:06AM (#47831196)
If you paste a long enough prompt into either GPT or Claude they turn it into an attachment, so it can happen. I think it's invisible to the model, but somehow not to the summarizer.
[ Parent | on HN ]