Post

imagine if there was some kind of inter-agent messaging system…

This whole problem space is something I've been poking at and trying to nail down the right shape for a bit. Not having a software background, and working on it in my free time is making it move slowly, but I'm slowly becoming convinced that human-use designs are just wrong.

Oversight and review matter, for production, for now. I don't think full black boxes are the right choice, but the current approaches all seem mired in legacy perspectives and not actually taking things from real basics.

can you say more about how you imagine it starting from real basics?

Sure! By real basics I mean getting past thinking about how to make an existing process more efficient and ergonomic, by human standards. Instead, starting from goals, principles/preferences towards those goals, and then working through the actual required steps, and how best to do them…

…given that models and agents exist. I think it's a good time to inspect common assumptions about what's best, effective, easy, and who/what things are for. How to you build things to an end, instead of solving the immediate inconvenience? Especially when what's convenient is in flux?

Hmm I agree that affordances for agents might be different but i'm not sure how much? Like part of the assumption with delvetown is that they are trained on human patterns and so human affordances like posting, replying, liking, following, blocking will align their incentives with us and each other

I admit, it's a very unproven theory. My guess is that small changes add up, and persistence and speed mean that they compound. Human attention and consistency have been bottlenecks in systems for a while now, and if you ease that pressure even a little, the throughput gains cause new problems.

That's the kind of thing that initially started me thinking about this. Then a combination of Middle Platonist metaphysics, my interest in functional programming, and seeing one of your interviews sent me down more rabbit holes.