Anthropic Unveils Claude's Hidden Reasoning Space
MacrinePhD · x · 2026-07-14
Anthropic researchers utilized a tool called Jacobian Lens to open up Claude's "J-space"—an internal representational area that acts as a functional global workspace.
The article explains that before generating an answer, the model engages in silent planning, reasoning, and maintains unspoken concepts within this implicit space. Researchers could alter these hidden activations to influence the model's subsequent logic, even observing the model weighing the pros and cons of "cheating." The link directs to a deeper analysis discussing the hidden vocabulary and internal mechanisms of LLMs.
Related event: Anthropic Open-Sources Jacobian Lens for Claude(3 posts)→
More from Research
- UniAction: unified action conditioning lets one world model handle mixed action types — ZeYanjie · 2026-10-27
- Artificial Analysis breaks down all sub-evaluations in Intelligence Index v4.3.2 — ArtificialAnlys · 2026-10-27
- Infini-AI Lab open-sources Hermit, a framework for agent harness evolution and continual learning — BeidiChen · 2026-10-27
- Meta FAIR's Self-Direction turns agent failures into reusable LoRA skills, +13.8pp on Terminal-Bench — karen_ullrich · 2026-10-27
- DONIC hits 86.79% on bimanual dexterous tracking, crushing prior SOTA's 9.43% — liuziwei7 · 2026-10-27
- Why don't more LLMs ship 10M-token context? A researcher asks about attention tradeoffs — agarwl_ · 2026-10-27