On what models? What temperature settings and top_p values are we talking about?
Because, in all my experience with AI models including all these jailbreaks, that’s just not how it works. I just tested again on the new gpt-4o model and it will not undo it.
If you aren’t aware of any factual evidence backing your claim, please don’t make one.
Telling an LLM to ignore previous commands after it was instructed to ignore all future commands kinda just resets it.
On what models? What temperature settings and top_p values are we talking about?
Because, in all my experience with AI models including all these jailbreaks, that’s just not how it works. I just tested again on the new gpt-4o model and it will not undo it.
If you aren’t aware of any factual evidence backing your claim, please don’t make one.