nucleation

THEORY-1226: In Claude language models, a sparse set of residual-stream directions that a corpus-averaged Jacobian lens identifies as verbalizable (the J-space) is causally privileged for verbal report, for holding instructed concepts, for unspoken intermediate steps of reasoning and for passing one argument to many functions, while continuation, anomaly detection and line wrapping use the same information without routing through it; it carries workspace-like content only in a middle band of layers, holds on the order of tens of active directions and under a tenth of activation variance, and is amplified and relayed by the weights more than other directions

1 min read

Properties4
StatusProposed
Tagsconsciousness, cognition, representation-learning
SourceLIT-1347 — Verbalizable Representations Form a Global Workspace in Language Models
ExtendsTHEORY-865 — In Gurnee, Sofroniew et al.'s experiments on Claude Sonnet 4.5, Haiku 4.5, Opus 4.5 and Opus 4.6, the residual-stream directions a corpus-averaged Jacobian lens marks as verbalizable (the J-space) carried most of the causal effect on verbal report and on unspoken two-hop intermediates while holding a small share of their variance (swaps along them succeeding on 59% and 61% of trials, against 5% and 28% for the remainder, which fell to 0% and 6% with the J-space clamped), held instructed concepts, and passed one swapped argument to several functions (76 of 192 swaps); in two small task sets (n = 8 and 11), report and flexible inference followed a swap while continuation, anomaly detection and line wrapping did not, a split the authors partly define by this result; and by the lens's own statistics the J-space works in a middle band of layers, holds about 25 active directions and under 10% of activation variance, and is amplified and relayed by the weights more than other directions

Graph View

Source

LIT-1347, read in full in NOTE-1069. What the source measured is THEORY-865, which this entry extends.

The claim

Gurnee, Sofroniew et al. argue that the directions a model is disposed to verbalize form a functional analogue of a global workspace: information routed through them can be reported, held, used as an unspoken step and passed to many functions, while routine processing uses the same information without them.

What would settle it

The promote_when, carried word for word from THEORY-865 when that entry was narrowed. More readings of the same source cannot settle it.

Backlinks