My speculation is that the “safety” of major LLM providers like Facebook won’t be substantial. They are probably designed so that “popping the lock” isn’t difficult. A fig leaf and shrug to placate outsiders, while allowing their actual audience to carry on.
My speculation is that the “safety” of major LLM providers like Facebook won’t be substantial. They are probably designed so that “popping the lock” isn’t difficult. A fig leaf and shrug to placate outsiders, while allowing their actual audience to carry on.
I’m not sure that it’s plausible to add meaningful safety to open source LLMs.
Incidentally, that does worry me some in the long term.