Louis Jaburi

I graduated as a PhD student at Imperial College under the supervision of Ana Caraiani in 2024.

I worked on arithmetic geometry, representation theory and number theory. In particular, this includes the p-adic geometry of Shimura varieties, and Emerton-Gee stacks.

I am now working on various flavours and shapes of AI safety with focus on interpretability, theory of deep learning, training dynamics at EleutherAI.

News and Updates:

05/2025: Just put out two parts on the philosophical foundation of mechanistic interpretability: I. Is mechinterp good? and II. What is good mechinterp?
02/2025: Check out our new blog post exploring complexity, (hierarchial) generalization, and its relationship with various empirical metrics.
10/2024: New paper on rigorous mechanistic interpretability on toy models.