Now

San Francisco. I'm heading to Berkeley on June 15 for the Generator residency through August 28.

What I'm working on. I'm designing evaluations for the epistemic capabilities of frontier AI models: sycophancy, selective omission, calibration. A Coefficient Giving grant is funding my transition into AI safety research.

What I'm building. A relationship operating system that lives inside my Obsidian vault, nudges me when contacts go stale, and surfaces serendipity I'd otherwise miss. Mostly scripts and Claude Code skills.

What I'm publishing. I sent the first edition of a quarterly note to friends last week. If you'd like to be on it, send me a line.

What I'm reading. Scott Alexander's Unsong (rationalist biblical fanfiction, somehow it works) and Victor Haghani's The Missing Billionaires (on bet sizing).

What I'm learning. Polish, getting back to B2. I'm a heritage speaker who lost most of it in elementary school. I journal in Polish a few nights a week.

What I'm curious about. Whether AI systems can be evaluated on whether they update their beliefs in response to evidence, the way we'd want a good forecaster to.

This weekend. I'm at Recursive 2026, an AI alignment and evals conference in SF and Berkeley.

Last updated: May 15, 2026