Anthropic on Internal Model J-space
RileyRalmuto · x · 2026-07-11
Anthropic released content named J-space, discussing a "global workspace" inside the model that can be read via J-lens.
The reposter mentioned that this aligns perfectly with the layer structure they have been measuring externally since 2023, making the read mind-blowing. They also validated it against their own public work, and it was a "hit".
Related event: Anthropic Research Reveals Claude's Internal Latent Workspace(4 posts)→
More from Research
- Exponential launches RepoInject and BrowserInject benchmarks for prompt injection robustness — maksym_andr · 2026-10-27
- Odd eval result: Gemini 4.1 Flash gets more accurate when asked to control its CoT — dfrsrchtwts · 2026-10-27
- Redwood: CoT controllability evals are under-elicited; better prompts boost scores 2-3x — dfrsrchtwts · 2026-10-27
- Inverse-designed optical system solves MNIST using BeamZ — S_Conradi · 2026-10-27
- DeCert supervises generative reward models' rationales without reference explanations — Guiquan Sun · 2026-10-27
- VLM harnesses act as compositional generalizers, new blog post shows — a1zhang · 2026-10-27