Video wird geladen...
Video konnte nicht geladen werden
I asked my autonomous AI to solve AI alignment. Fascinating to watch. Agents: - Task execution: text-davinci-003 - Task creation: text-davinci-003 - Task reprioritization: text-davinci-003
64,053 Aufrufe • vor 3 Jahren •via X (Twitter)
9 Kommentare

Damn you gotta start adding music

Oh tight. Wait, should I start a Tik tok? Take requests and just post videos of the output.

solved

Ugh. I wish we hadn't defined the term "alignment" to cover both "do what I actually want" (human-to-computer), and "we need to agree now on what we want" (humans to each other). Is there a term for just the 1st challenge -- aligning what you ask for with what you truly wanted?

Funny you say that, because it’s suggesting I take a mortgage on my house. Needs more compute or something - it’s for the greater good.

excellent. yohei, pls use your big brain for this, we have no idea how to solve alignment

How do you stay within with 4k token limit as it keeps extrapolating ?

It’s a balance of figuring out how much past context you want to pull in. If you pull zero, it stays within limit easily but no prev context. If you pull 5, you get decent context but it can get large. So then you can consider adding a summary step.

Why text-davinci-003 instead of 3.5-turbo?


