Comment by commieneko
18 hours ago
Interesting. I pasted the article URL into Claude Opus 4.8, along with some questions about uses for cells that couldn't reproduce and Claude thought about it for a while, and then got murdered by the guardrails. I was invited to edit the question and try again; in a different chat. Or use a dumber model.
I suppose I can see why. But at the time I was just curious about the idea of "mule" cells.
What are the guardrails here?
I've read than even a lot of high school biology questions can set off safety guardrails on Claude.
A lot of high school biology underpins the most immediate interesting aspects of chemistry and biology, and also the most volatile and dangerous ones.
I'm curious what you asked? I had Opus 4.8 ingest the URL and give me some ideas about what's possible and it eventually got to AI and self repairing/improving factories along with listing risks etc.
1 reply →
They're paranoid about people using AI to synthesise anthrax or something. You also can't ask it how to build a nuclear bomb.
Which shows why guardrails on AI are just dumb. What harm could come from answering your question? None.
Red blood cells can't reproduce and are "mule" cells - bags of hemoglobin. They are not really alive, so maybe not exactly what you were thinking of.