research
most of my time goes toward mechanistic interpretability. i'm particularly interested in superposition and polysemanticity right now. trying to understand how features are represented across neurons when there are more features than dimensions. the work coming out of Anthropic and others on sparse autoencoders is exciting.
on the economics side, i'm thinking about how AI changes the production function. not in a vague "AI will change everything" way. more like: what happens to labor share of income when marginal costs of cognitive work approach zero? what happens to comparative advantage between nations?
writing
working on a longer piece about why prediction markets haven't taken off despite being demonstrably better than alternatives. also drafting something about the intersection of AI governance and economic policy. most countries are treating these as separate problems. they're not.
reading
re-reading parts of "the alignment problem" by brian christian. also going through a lot of the recent mech interp literature from Anthropic's interpretability team. on the econ side, daron acemoglu's work on AI and labor markets.
other
tracking prediction markets more seriously. putting money where my mouth is on a few questions. also trying to write shorter sentences. it's harder than it sounds.