Accepted to ICLR 2026. I co-authored a post on LessWrong on removing LayerNorm from transformers by fine-tuning and implications for mechanistic interpretability (direct logit attribution, attribution patching, entropy neurons).

Notes on how language models are transformed from raw next-token predictors into useful assistants. Covers the three main stages of post-training — supervised fine-tuning, reinforcement learning from human feedback, and direct preference optimization — along with practical details about tool calling and training dynamics.

Co-authored with Aaron Kaufman for the ARENA capstone: we replicated MELBO (Mechanistically Eliciting Latent Behaviors) on Llama-3.2-1b-Instruct and did a hyperparameter sweep for the R value using diversity and coherence metrics.