3 ms·
Think of AI tools as unbelievably good auto-complete or intellisense. A universal anything-to-anything translator. It's incredibly powerful because it's somethi
by gizmo 3y ago
Think of AI tools as unbelievably good auto-complete or intellisense. A universal anything-to-anything translator. It's incredibly powerful because it's something humans are not very good at.
But it's a mistake to think AIs are "just" bullshitters. If you ask a GPT to auto-complete some bullshit for you, it will happily do so. But it will do many tasks that provide value to society just as easily.
Nobody knows how society will be transformed by AI. But if AI generated BS is a problem then I wouldn't be surprised to see AI powered skepticism assistance as well. We all have terrible blind spots, and maybe in an AI world it will be harder to delude yourself than in the present day? No predicting how this will turn out.
- jprete 3y agoI think LLMs and RLHF are inherently unsuited to producing aligned, useful, and fact-oriented AI. Something else is needed.
- gizmo 3y agoGPT4 (with decent prompting) provides very accurate and unbiased summaries of complex and contentious political questions. It's by design not a fact-oriented system but it is capable of producing very good factual answers.
- blibble 3y ago> It's by design not a fact-oriented system but it is capable of producing very good factual answers. this is only by chance though, and you have no way of distinguishing it