Finding an Inner Monologue Representation Space in LLMs
mlpowered · x · 2026-07-07
The mlpowered team identified a small subset of representations inside language models, dubbed J-space. Functioning like the model's "inner monologue," it presents intermediate concepts, makes judgments about the current text, and can even detect prompt injection. This offers new clues for model interpretability and safety.
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