Tag
An article discussing a new interpretability technique called the Jacobian Lens and the discovery of J-space, a region in LLMs where verbalizable representations form a global workspace, marking a significant advance in understanding LLM reasoning.
Anthropic releases research on a global workspace in language models, prompted by observations that Claude has unspoken thoughts in reasoning traces.
This paper presents evidence that language models maintain a privileged set of internal representations ('verbalizable representations') that form a global workspace, analogous to human conscious access, and introduces new interpretability techniques for identifying and intervening on these representations.
Anthropic's new research identifies a 'J-space' of internal activations in language models that acts as a global workspace for deliberate reasoning, distinct from automatic fluent output. The findings reveal that models can hold and report internal thoughts not expressed in their final output.
The author presents a method to make LLMs verbalize calibrated confidence by using a linear probe on mid-layer states and a small trained bridge to confidence logits, requiring only 200 labeled examples and no weight modification. This is linked to Anthropic's global workspace paper explaining the know-say gap.
Anthropic published research into internal model reasoning, releasing the J-Space lens code and a demonstration on Qwen 3.6 27B.
This tweet discusses Anthropic's new research on a global workspace in language models, noting that Karpathy is not in the author list and emphasizing that the work was completed during the Sonnet 4.5 era, criticizing those who simply see it as hype.
This paper introduces the Jacobian Lens (J-lens) and J-space to show that LLMs like Claude Sonnet 4.5 maintain verbalizable internal representations that function like a global workspace, causally used for flexible reasoning—demonstrated through intervention experiments.
Anthropic's new research reveals a global workspace in language models, showing a striking parallel between the conscious and subconscious divide in human brains and the internal reasoning of Claude.
Anthropic's new paper presents evidence that modern language models like Claude have developed a 'global workspace' (J-space) of internal neural patterns that are reportable, controllable, and used for flexible reasoning, distinct from automatic processing.
Anthropic's new research discovers a 'global workspace' in language models analogous to conscious processing in the human brain, finding a divide similar to conscious and unconscious thought.