MIT Researchers Build Mathematical Theory for Anthropic's Jacobian Lens

MIT researchers published a mathematical theory of Anthropic's Jacobian lens (J-lens), explaining why the interpretability tool captures slow, coherent, working-memory-like features, building on its earlier discovery of global-workspace-like evidence in Claude.

2026-10-27 ~ 2026-10-27 · 3 related posts

• New theory paper explains what Anthropic's Jacobian lens actually detects in LLMs — wgilpin0 · 2026-10-27
• A mathematical theory of Anthropic's J-lens: why it finds working-memory-like features — burny_tech · 2026-10-27

1 near-duplicate retellings: burny_tech