Comment by azan_
5 days ago
There's zero reason to believe that. If AI becomes better at mental tasks than humans (that's assumption, not something I claim will 100% happen), why would human oversee AI instead of AI supervising another AI?
5 days ago
There's zero reason to believe that. If AI becomes better at mental tasks than humans (that's assumption, not something I claim will 100% happen), why would human oversee AI instead of AI supervising another AI?
Bc humans can be responsible for approving / denying decisions in the end. Bchumans have the context a brand new prompt might miss. Bc the most intelligent and brightest and smartest AI ever built will do the wrong thing if given the wrong prompt. It can't presume to know things it was never told or has no way to know
Was the last time you've used LLM two years ago? Current SOTA makes good judgement calls and works well even with shitty prompts. And current SOTA is the worst these models will be.
I use unreleased models at work every day. When they work they work great. But almost every day I run into a case where they make a mistake. Also, the humans prompting them may be themselves wrong or understand the problem wrongly or not include important context. That's not a solvable problem either really.