Anthropic Releases Full Paper on LLM "J-Space"
wesg52 · x · 2026-07-07
Anthropic's interpretability team has officially released the full research paper on the language model "Judgment Space" (J-space). J-space is a small subset of model activations that reflects the model's internal state, emotions, implicit thoughts, and safety awareness, existing independently of output tokens. The research reveals a systematic distinction between a model's "reportable cognition" and "implicit internal processing," opening a new direction for AI interpretability and safety research. The paper was co-authored by Jack Lindsey, Nick Sofroniew, and the Anthropic interpretability team.
Related event: Anthropic Discovers Global Workspace Inside Claude(102 posts)→
More from Research
- Token Collapse: Study Finds Graph Soft Prompting Tokens in LLMs Fail to Encode Structure — Sapienza · 2026-10-27
- Terence Tao's SAIR Foundation revamps Mathathon with $1.5M compute per team — _sathvikr · 2026-10-27
- Mathathon 2.0 launches with Terence Tao's SAIR, $1.5M in compute for teams — _sathvikr · 2026-10-27
- The 'sweet lesson' is real: two kinds of ML problems need opposite strategies — mike64_t · 2026-10-27
- PixGR: Pixel-space generative decoding for fast feed-forward 3D reconstruction — kwangmoo_yi · 2026-10-27
- UniAction: unified action conditioning lets one world model handle mixed action types — ZeYanjie · 2026-10-27