Chris Olah

Chris Olah co-founded Anthropic and leads its interpretability research, which tries to work out, neuron by neuron and circuit by circuit, what a trained model is actually computing. His own site describes the work as reverse engineering neural networks "into human understandable algorithms."

He has no university degree. Raised in Toronto, he left the University of Toronto after about a year, and in 2012 received a 100,000 dollar Thiel Fellowship, which pays people under 20 to skip or leave college. He told the 80,000 Hours podcast that the adults around him "totally came around" once someone had given him the money. His blog posts explaining neural networks with diagrams became standard teaching material.

He started at Google Brain as an intern in 2015 and stayed three years as a research scientist, co-writing the DeepDream work and feature-visualization research that showed what individual neurons respond to. In 2017 he co-founded Distill, an online journal for clearly explained machine-learning research, and edited it until it went on hiatus in 2021. In 2018 he moved to OpenAI to lead interpretability, where his team published the Circuits project and found "multimodal neurons" in the CLIP model that fired for a concept such as Spider-Man whether it appeared as a photo, a drawing or a word (published in March 2021). He left with Dario Amodei's group at the end of 2020.

At Anthropic his team applied dictionary learning to language models, extracting millions of interpretable features from Claude 3 Sonnet in May 2024; amplifying one of them produced "Golden Gate Claude," a model that steered every conversation toward the bridge. In May 2026 he spoke at the Vatican presentation of Pope Leo XIV's encyclical on AI, arguing that outsiders must oversee companies like his own.