Jacob Dunefsky

Advanced
@jacobdunefsky

ML interpretability researcher. Alum of MATS 5 in Neel Nanda's stream. B.S./M.S. Yale '23.5; current PhD student @ Yale with Arman Cohan.

transcoder_circuits. Jupyter Notebook

213

llm-steering-opt. Tools for optimizing steering vectors in LLMs.

22

one-shot-steering-misalignment. Code and results on finding one-shot steering vectors that mediate emergent misalignment

12

one-shot-steering-repro. Code for reproducing the results from our paper "Investigating Generalization of One-shot LLM Steering Vectors"

7

ObservablePropagation. Code for our paper on observable propagation

4

peek. Peek: a tool and platform for peeking inside large language models.

3

multilayer-transcoder-reversing. Reverse-engineering transcoder features in multi-layer LLMs

2

hacky_saes_and_tcs. My repository for playing around with weird SAE and transcoder variants. Based on a really old version of github.com/jbloomaus/saelens

1
8
Apply