finding
active
finding:prefilling-dispreferred-option-makes-but-appear-strongly-in-post-trained-j-space-while-model-argues-for-it-in-88-of-casesPrefilling dispreferred option makes BUT appear strongly in post-trained J-space while model argues for it in 88% of cases.
Internal objection.
Source paper
extracted_fromAuthors
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Base vs post-trained comparison.
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.
- The model attends to the J-space on the user prompt when determining its introspective answer, so concepts there are reportable.hypothesis0.781Motivated by elephant observation.
- Ablation flattens experiential language.
- Suppressing the J-space entirely should impair flexible reasoning while leaving automatic processing intact.hypothesis0.771Tested via ablation battery.
- MLP and attention analysis.
- Better bases may give sparsity.
- Developmental analogy used to explain sample efficiency under high ρd conditions
- Judge sensitivity.
- Mechanistic insight surfaced by NLA explanations and validated through independent causal attribution method.