alt Hacker NewsBut on a much more serious note, the violating/breaking of the guardrails when making bombs is terrible. I'd have called it unforgivable, but LLMs are a tough beast to tame in the best of situations... and I'm not really sure if chatgpt ever deserved to be forgived.
It's also ironic that Fable hits guardrails for nothing, and a literal terrorist group is making bombs and merrily skipping over guardrails.
I feel like I'm missing something though. You can open LM Studio right now and download any model with "heretic" or "uncensored" in the name and it will happily do anything you want with no restrictions whatsoever. What's the point of trying to jailbreak ChatGPT? Is it that much better if all you want is just some instructions to make bombs or whatever? (admittedly - I have no idea if these instructions are actually worth anything, but the models will not object to any question)