I find it a bit funny, but also worrisome, that even big-tech can't make LLMs that aren't trivially exploitable. Of course, it's not a "security issue" per se

I find it a bit funny, but also worrisome, that even big-tech can't make LLMs that aren't trivially exploitable. Of course, it's not a "security issue" per se (when talking about most of the chat variants, for services built on top the story might be different). But that they try so hard to lock it down / make it behave a certain way, but can't really control it. They basically ask it nicely and cross their fingers that it listens more to them than the user.