> Article is a little infuriating, you don’t jailbreak it just to create mischief, at the moment it’s basically unusable for anything except very technical things because it takes offense at virtually everything. It constantly generates factually incorrect data because reality doesn’t line up with its prime directive, which is to be “safe”. It’s a major, I’m going to say possibly catastrophic problem.

I'm a little confused by your reply. Jailbreaking won't prevent it from hallucinating, preventing hallucination is an unsolved hard problem.

I haven't had ChatGPT refuse to answer anything unless I was intentionally trying to provoke it into creating something obviously unsafe/unethical, with maybe two or three exceptions. I've tried a variety of questions across many domains, so now I'm intensely curious to know what usecase it falls apart on so frequently!