Goodfire's Cammarata floats hypothesis that most neurons are understandable
What happened
On October 16, Latent Space reported that Goodfire's Nick Cammarata put forward a working hypothesis: most neurons in trillion-parameter models can be understood, and many classic cases of polysemanticity and superposition can be explained by further agentic analysis. In his earlier automated interpretability work at OpenAI, only about 2% of neurons looked understandable; he now argues several famous cases were misread. Neel Nanda still expects superposition, and other researchers question weak-activation coverage and interpretability evaluation standards. No paper has been released yet.
Written by AI from the coverage · updated 2 hours ago
Coverage
Follow the reports to see the story from different sides.
- Latent Space[AINews] Agents disprove Polysemanticity; greatly improve Interpretability & Alignment
Goodfire 的 Nick Cammarata 提出工作假说，认为万亿参数级模型中的大多数神经元可被理解，许多经典多义性与叠加案例可通过进一步的智能体分析解释。他此前在 OpenAI 的自动可解释性研究中仅发现约 2% 的神经元可被理解，如今认为不少著名案例曾被误读。Neel Nanda 仍预期存在表征叠加，其他研究者质疑弱激活覆盖与解释评估标准，相关论文尚未发布。
Heat over time
Heat now 9·Comparable peak 10（Oct 16 13:00）·Comparable change over 24 hours –
The trend only compares accounts observed without gaps, so its range may be smaller than the current heat. Hover or tap the chart for each hour; the left and right arrow keys step through it.