mediocreatbest

103 readers

1 users here now

My collection of links, articles, projects, and more that I find interesting.

founded 2 years ago

MODERATORS

Taming AI Bots: Prevent LLMs from entering "bad" states using continuous guidance from the LLM ("is this good? bad?") to avoid bad states. (arxiv.org)

submitted 2 years ago by mediocreatbest to c/mediocreatbest

3 comments fedilink hide all child comments

you are viewing a single comment's thread
view the rest of the comments

[–] chaorace 1 points 2 years ago* (last edited 2 years ago) (2 children)

Awesome. Love to see that we have our own local LLM researcher on the instance. Pretty cool paper, too!

It's impressive to see how much your group was able to conclude given the limitations imposed by working on a closed model. Really makes you wonder what might be possible if you could have a supervisor AI agent actively inspect the internal state of GPT as it runs, eh?

[–] mediocreatbest 1 points 2 years ago

Oh! I should have clarified that I'm not the author of that paper, I just read it and wanted to share.

That idea of a supervisor AI agent is essentially the idea, but instead of supervising the weights or internal activations, it's actually supervising the tokens of output instead. Very cool stuff!

I have some more notes that I forgot to copy over here, so I'll share those tomorrow 😅

load more comments (1 replies)