finding
active
finding:hidden-states-of-middle-entity-groups-become-less-linearly-separable-by-position-as-n-growsHidden states of middle entity groups become less linearly separable by position as n grows
Fig. 22.
Source paper
extracted_from(2025) · Yoav Gur-Arieh · Mor Geva · Atticus Geiger
Neighborhood — ranked by edge-count
Papers (1)
paper
Hypotheses (1)
hypothesis
- Positional signal is diffuse because the model encodes group positions in a poorly linearly separable way for middle groupsassociated_withsupportsTested via probes.
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.
- We hypothesize that group (b) hidden states store a representation of the statement's truthhypothesis0.786Motivating hypothesis driving the remainder of the paper's analysis after patching localization
- Patching experiments localize truth representations to these specific hidden states in LLaMA-2 models
- Explains why RevNet lacks capacity to separate states for identity-of-first-argument algorithm
- Challenges prevailing view.
- Localization result from patching experiments; identifies group (b) hidden states as the locus of truth representations
- Interpretive synthesis of DIM and cone intervention successes
- Tested in Section 4.4 calibration experiment; confirmed by findings.
- Figures 14–17.