Chris Olah
Anthropic co-founder working on reverse engineering neural networks; previously led interpretability at OpenAI
Chris Olah works on reverse engineering artificial neural networks into human-understandable algorithms and is one of the co-founders of Anthropic. Olah previously led interpretability research at OpenAI, worked at Google Brain, and co-founded Distill, a scientific journal focused on clear communication. At Anthropic, Olah is the corresponding author of "A Mathematical Framework for Transformer Circuits" (2021), last-listed author of "Toy Models of Superposition" (2022), and a co-author of "On the Biology of a Large Language Model" (2025), which studies the internal mechanisms of Claude 3.5 Haiku with circuit tracing.1234
Work
A Mathematical Framework for Transformer Circuits (external site: transformer-circuits.pub)Published research
Corresponding author of the 2021 Anthropic article on the Transformer Circuits Thread, which develops a framework for reverse engineering small attention-only transformers.2
Toy Models of Superposition (external site: arxiv.org)Published research
Last-listed author of the 2022 paper, which gives a toy model in which polysemantic neurons arise from models storing additional sparse features in superposition.3
On the Biology of a Large Language Model (external site: transformer-circuits.pub)Published research
Co-author of the 2025 Anthropic study; its contributions statement credits Olah with work on the replacement model and attribution-graph computation, intervention analyses, and crosscoder ideas.4
Earlier roles
- Led interpretability research, OpenAI1
- Worked at Google Brain, Google Brain1
This profile covers public professional work only, from the sources below. It is not an endorsement or a ranking. To correct or remove it, use the correction form.