This post was written as part of MATS 9.1 under the mentorship of Richard Ngo, and was written during Iliad Fellowship, to all of whom my thanks. LLM Usage: prose drafted by Claude from my outline, talk materials, and notes. I edited thereafter. There is some residual Claude cringe in...
This post was produced as part of the Iliad Fellowship under the mentorship of Dmitry Vaintrob. Tl;dr: Power-law ("heavy-tailed") distributions have universality theorems similar to those which make Gaussians common. We observe many things in ML are power-law distributed, most robustly and interestingly, the spectra of weight matrices. I explain...
This post was produced as part of MATS 9.1 under the mentorship of Richard Ngo. It is not part of my main research project, but the ideas have been an important conceptual anchor to me. Epistemically, treat this as watercooler talk. Please feel free to share additional or contradictory work...
This post was written as part of MATS 9.1 under the mentorship of Richard Ngo. This post is the first of several I will be writing on using natural selection to understand artificial intelligence and agency. This post will show how noisy selection on genome structure can make evolution effectively...
I got an enormous amount of personal and intellectual value from my 1.5 years in the TU Delft Debate Club. I did British Parliamentary debate in the Netherlands in 2022-2023, about twice a week. I went to ~5 external tournaments. It made me wiser, more personable, more open-minded, and, I...
(Note: Edited 26 Feb. 2026 to include the whole text, rather than a blurb) This post was written during MATS 9.0 under the mentorship of Richard Ngo. The argument in brief: Neural scaling laws — the power-law improvement of loss with model size, data, and compute — can be derived...
(Note: Edited 26 Feb. 2026 to include the whole text, rather than a blurb) In this post I will try to outline an idea which I have imbibed by studying physics, but which I think is generally applicable as a tool for thinking. This is the idea: because we can...