Chris Olah Portfolio
Co-founder, Anthropic
Co-founder of Anthropic who works on reverse-engineering neural networks into human-understandable algorithms, after leading interpretability research at OpenAI, working at Google Brain, and co-founding the journal Distill; his blog and the "Circuits" paper series present original findings on reverse-engineering what individual neurons and features compute inside trained networks, built on a decade-long archive going back to posts like "Understanding LSTM Networks."
Location
Somewhere on planet Earth, most likely near a laptop.
Companies
Skills
Notable
- Co-founded Distill, a scientific journal focused on clear research communication (per his About page)
- "Understanding LSTM Networks" (2015): 3,700+ citations on Google Scholar for a blog post
- Co-author of "Concrete Problems in AI Safety" (2016)
- Co-author of the Circuits thread on Distill (2020), starting with "Zoom In: An Introduction to Circuits"
What makes it work
A breakdown of the choices that make this portfolio stand out.
Original research, not secondhand explanation
The Circuits series doesn't summarize other people's papers — it publishes original findings about which internal features and circuits deep networks compute, the kind of ground-up empirical work that defines actual research rather than research communication.
Diagrams built to carry the technical argument
Custom visualizations of activation space and feature interactions are not decoration — they are the proof itself, since interpretability claims are hard to state convincingly in words alone.
A research agenda visible across a decade of posts
Read start to finish, the archive traces one throughline — from "Neural Networks, Manifolds, and Topology" (2014) to circuits and transformer interpretability — showing a sustained research program, not scattered topical posts.
What AI Researchers can take from this
Specific, actionable tips to apply to your own portfolio — no generic advice.
Publish original findings, not just summaries of others' papers — a genuine research contribution is what separates a researcher's site from a technical-writing site.
When a claim is hard to state in words, build a custom diagram that carries the argument instead of describing it in prose.
Let your archive read as a single throughline over years, not a set of disconnected topics — it shows a sustained research agenda.
Projects
The homepage sorts a decade of work by what the neural network is (recurrent, convolutional, transformer) instead of by date, so Chris Olah's archive reads like an anatomy textbook in which each chapter got deeper than the one before.
Understanding LSTM Networks
A 2015 explainer that walks through the LSTM cell one gate at a time, with a hand-drawn diagram for each step.
Shows that one clear explanation of a hard mechanism can shape how a whole field learns it.
Diagram-per-gate structure
Published on colah.github.io, not in a journal
3,700+ Scholar citations
Circuits (Distill thread)
A multi-article Distill thread that starts with "Zoom In: An Introduction to Circuits" and continues through Early Vision, Curve Detectors, Curve Circuits, and Naturally Occurring Equivariance, tracing specific circuits inside InceptionV1.
Original, ground-up empirical research, the kind that defines a subfield rather than summarizing one.
Starts from single neurons and builds up to circuits
Interactive figures carry the argument
Leads straight into the later Transformer Circuits work
Feature Visualization
A 2017 Distill article on how neural networks build up their understanding of images, using optimization to render what individual units respond to.
Built reusable methods that later interpretability work, including the Circuits thread, relies on.
Interactive figures
Part of a run of Distill articles, 2017-2019
Method-focused rather than result-focused
More AI Researcher portfolios
Ready to build your portfolio?
Follow the AI Researcher roadmap — skills, projects, and timeline to get hired.