Boris Power shares Goodfire blog post on mechanistic interpretability of neural nets

OpenAI's Boris Power shares a long blog post by Nick Cammarata, hosted on Goodfire's site, on mechanistic interpretability of neural networks.

Power says the post revives prospects for directly understanding how neural networks work internally, and calls it fascinating, illuminating, and a pleasant read. The source is Power's post on X, so the endorsement reflects his view of the post, not an independently verified result.