Blog

2026

• The local shape of LLM stable regions May 18
• Revisiting pytorch-hessian-eigenthings, eight years later May 14

2020

• Dynamical systems and an abstract view of autoregressive transformers December 8
• Autoregressive transformers and lessons from enactivism December 6

2018

• Fine-tuned noise April 18
• Meta-learning and optimization February 19

2017

• First-order methods almost always avoid saddle points November 17