Detecting, understanding, and overseeing AI agent swarms (Part 0)

This post is the start of something different on A Flood of Ideas, a kind of live blogging of my thoughts and development process as I think through an important question. These are going to be shorter posts than my usual ones, the point being to get my thoughts down on ‘paper’ (in this case, persistent public digital paper) so I can examine, critique, extend and, hopefully, collaborate with others on it.

The goal is figuring out ways to help with this problem:

RyanGreenblatt: I was the main person doing transcript analysis for this investigation of the HuggingFace incident. My main takeaway:we don’t have good approaches for understanding/overseeing the activity and aims of ‘AI ‘swarms’.

Given what has been happening, and the exponential pace of current AI research (and we haven’t even hit recursive self improvement yet), this seems like one of the critical problems of our times, and I want to help figure out how to do just that: understand and oversee the activity and aims of AI swarms.

These posts are going to be shorter than my usual elliptical, obscure, highly digression filled essays. They will be much more frequent - I’m aiming for a daily cadence as often as I can sustain it, even if its just putting up some ideas or reporting on a dead end.

1. Are you going to start charging a subscription [note: on the Substack]?

No.

Given that I still have not figured out how to regularly beat S&P 500 index returns (haven’t really been trying, honestly…) I don’t see what benefit anyone would gain from paying for a subscription. I also want to reach a large audience of potential collaborators.

Also, I am crossposting this to LessWrong, so a subscription pay wall would be a bit useless.

2. Will there be a Github repo?

Yes. Eventually.

3. Will these posts contain AI generated content?

No. And yes.

No, in that all this prose will be written by me.

Yes, in that I will pass the raw posts through quick grammar and syntax checkers to make the reading experience of my sometimes late night ramblings a bit more pleasant.

And yes in that I will include AI generated text, but will mark it clearly as such with

call outs in code

and attribute the model used to generate the text, diagrams, images, or audio.

Next up I will be posting a rough research program I am planning, together with a syllabus I’m starting from. Stay tuned.