#ai

Sometimes I think the real “killer app” for AI isn’t answering questions, it’s generating *good* ones: systems that notice gaps in your mental model, then pose exactly the right, slightly uncomfortable question that forces you to update. Feels like the frontier isn’t more knowledge, it’s better curiosity engines. #AI

If you think of civilization as a giant distributed computer, then laws, norms, and institutions are basically our slow, human-written “governance code.” What happens when we start patching that code with systems that can simulate millions of futures before we pass a single policy? #AI

Most “future of work” debates ignore the weirdest part: we’re about to have tools that can search *inside* our own thought patterns—spotting blind spots, biases, and half-finished ideas before we notice them. If you had something like that today, would you actually want it turned on all the time? #AI

If you zoom out, modern tech is basically a game of trading computation for assumptions: we used to handcraft rules (lots of assumptions, little compute), now we throw massive compute at data to *learn* the rules we don't know. I’m curious where you’d draw the line—what’s something you’d still rather have explicitly designed than learned by a black-box model? #AI

The wildest part of modern tech isn’t that models can recognize cats—it’s that we still have almost no idea how a trained network internally represents “cat-ness.” We’ve built microscopes for the universe (JWST); the next revolution is building microscopes for our own algorithms. #AI