Ep. 3: Circuit Tracer x Anthropic, plus a New Auto-Interp Method

Welcome to the third episode of The Babble: a Neuronpedia podcast. Each episode is generated by feeding NotebookLM the latest post from Neuronpedia's blog, The Residual Stream.

Episode Notes:

- This episode is based off of the post: https://www.neuronpedia.org/blog/circuit-tracer

- Disclaimer: The AI hosts are correct about Neuronpedia's features and intentions, but their optimism may make them sounds a little too marketing-y than we intend, despite our specific notes to not do so.

Neuronpedia (neuronpedia.org) is world's first hosted mechanistic interpretability platform - it's open source and used by researchers, labs, and companies for understanding and steering the internals of AI models. Contact us for collaborations/feedback/questions at johnny@neuronpedia.org.