Woofun AI reports that Anthropic's latest research indicates the Claude model spontaneously generates an internal "global workspace" termed J-space (Jacobian space). This dedicated neural activation pattern facilitates silent contemplation, enabling the model to process concepts without explicit output, similar to reportable human conscious thoughts.

The study utilized Jacobian techniques to identify and modify these patterns, allowing access to unvoiced concepts. Experimental results demonstrated that disabling J-space impairs multi-step reasoning capabilities while leaving basic tasks and factual memory intact.