Notes and experiments
A place for my research notes, puzzle solutions, and ideas I am still working through.
October 2, 2026 · Decision theory · Cooperative AI
My log of a two-day ML4GOOD project, October 2–3. The project gave me the external incentive to dive into decision theory, something I expect to benefit me in the long run but would not normally make time to study.
August 16, 2026 · Technical AI safety · Research statement
My motivation, research direction, and current engagement with technical AI alignment.
July 21, 2026 · Technical AI safety · BlueDot 1st Puzzle
A controlled experiment in making a small MLP encode country through a prescribed sphere-shell or helix-tube manifold.
Read on LessWrong
July 20, 2026 · Technical AI safety · BlueDot 1st Puzzle
My investigation of a country feature that resists linear readout and instead forms a basin-like nonlinear representation in a small MLP.