Thread · 4 tweets · 5 May 2023

Yes, and yes :) This is at the "I think this is philosophically correct/interesting" one-page of MD with links stage. Those are some of the questions. Which personal network would you trust with what? Can models just communicate in English? Etc.

What interests me is building an architecture that supports a form of social cognition that somehow mirrors or overlays ours. We tend to think of ourselves as cognitive individualists but in practice without an epistemic network we're pretty stupid.

This leads us to create synthetic intelligence as often individualistic, as a translation of how we misunderstand our own intelligence. From that you get this idea of a worldwide all-knowing oracle. That's dumb and it concentrates power.

I'm curious about what an alternative would look like; I'm VERY VERY far from anything practical, or even just a bad demo :)