🤖 Is Polysemy Over?
Goodfire founder Nick Cammarata published the essay “Neural Networks Are Already Clean” on October 15: polysemy and superposition are largely misinterpreted — recursive generalization is more often behind them. Qwen3-32B audit: ~90% of neuron poles have one function. This is a hypothesis (~85% confidence), not a proof.
🌍 If the “confusion” in networks is explained by generalization and weak early tools rather than fundamental superposition, the main front becomes directly studying neurons, and SAEs are not the only path.
👤 The essay (~65 minutes reading) and the LessWrong discussion are open to everyone, the question is backed by numbers, not a meme. But this is one author’s hypothesis: 13+ responses in the comments, some experts see an overestimation.
Source 1: https://already-clean.goodfire.com/ Source 2: https://www.lesswrong.com/posts/x4kvabkF2Jk8GLZp8/linkpost-neural-networks-are-already-clean