claim
active
claim:structure-is-reflected-consistently-in-both-behavior-and-internal-representations-and-decodable-with-linear-probesStructure is reflected consistently in both behavior and internal representations and decodable with linear probes
Finding (2).
Source paper
extracted_from(2026) · Eric Bigelow · Raphaël Sarfati · Daniel Wurgaft · Owen Lewis +4
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (3)
finding
- Beliefs linearly decodable.
- Distance matrices of My and Mz correlate r=.71 (p<.001) for Arbitrary domain, lower similaritysupportsWeaker shared structure.
- Shared structure.
Claims (1)
claim
- Central thesis.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Interpretive synthesis of DIM and cone intervention successes
- The causal hypothesis motivating the use of causality (intervention) as the lens connecting representation and behavior geometry.
- Open theoretical problem CIMC acknowledges: precisely characterizing the representational format of perception
- Load-bearing theoretical claim providing the conceptual foundation for DAS.
- Acknowledged gap in CIMC's theory: the exact computational description of the perceptual regime is not yet specified
- The paper's deepest interpretive claim, asserting that representation structure and behavioral structure are not coincidentally aligned but deeply connected.
- Establishes that the observed linear structure is not merely a representation of text probability
- Manifold structure is present in representations and causally influences downstream behaviorclaim0.779Interpretation of steering results.