wesg

Karma: 491

OR PhD student at MIT working on interpretability.

Find out more here: https://wesg.me/

Refusal in LLMs is mediated by a single direction

Andy Arditi, Oscar Obeso, Aaquib111, wesg and Neel Nanda

Apr 27, 2024, 11:13 AM

246 points

95 comments10 min readLW link

SAE reconstruction errors are (empirically) pathological

wesgMar 29, 2024, 4:37 PM

106 points

16 comments8 min readLW link

Finding Neurons in a Haystack: Case Studies with Sparse Probing

wesg and Neel Nanda

May 3, 2023, 1:30 PM

33 points

6 comments2 min readLW link 1 review

(arxiv.org)