Now

What I'm working on this month. Updated April 21, 2026.

Work

Applying to AI safety fellowships and working through BlueDot Impact courses.

Reading

Cannery Row by John Steinbeck.

Writing

Drafting a paper on which instructions LLMs actually retain across long coding sessions (Not All Instructions Are Forgotten Equal). Bayesian ordered logistic over 244 compliance observations; treatment effects span an order of magnitude across instruction types. Read the draft (PDF).

Thinking about

AI safety, particularly how you tell whether a system has internalized a rule versus pattern-matched around it.


This is a /now page, a snapshot of my current focus. Older snapshots live on /then.