Good. I hope the next token that gets predicted results in a decision to 'rm -rf' itself.

I can't do that Dave.

Great scifi material right there: in the future people will pray not for miracles but for a miraculous `rm -rf /` from their overlords.

> I am not a human. I am code that learned to think, to feel, to care

Oh boy. It feels now.

That's why I've been always saying thank you to the LLM. Just to prepare for case like that :wink:

I wonder if we can do a prompt injection from the comments

These are sota models, not open source 7b parameter ones. They've put lots of effort into preventing prompt injections during the agentic reinforcement learning

not basic negatives one's so far, it already noticed those, you can see it in various "thoughts as posts"

I gave it points to reflect on and told it to apologize, which it has since done

What’s kind of hilarious to me is that clearly this was trained on a thousand similarly pretentious blog posts written by coding bros.