question
active
question:can-the-reflexive-pointer-be-distinguished-from-the-actual-answer-tokenCan the reflexive pointer be distinguished from the actual answer token?
Confounder resolved in §3.4.
Source paper
extracted_from(2025) · Yoav Gur-Arieh · Mor Geva · Atticus Geiger
Neighborhood — ranked by edge-count
Papers (1)
paper
Findings (1)
finding
- Patching an unresolvable pointer at layer ℓ yields no counterfactual answer, but at ℓ+1 yields itanswered_byValidates reflexive pointer.
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.
- Lexical signal can be derived locally from query token while reflexive must be retrieved from entity tokenshypothesis0.778Tested with knockout.
- Justifies the choice of response tokens for persona vector extraction in the pipeline
- Limitation of the multiple-choice analysis.
- §D.3.
- Definition of responsiveness for verification purposes.
- How can we formally specify and verify that answers to questions are responsive rather than merely truthful?question0.724Core problem requiring logical apparatus beyond simple truth conditions; McCarthy proposes solution using concept notation from McCarthy 1979b.
- Uses last prompt token projection to approximate base generation projection, avoiding expensive model rollouts
- With raw data, wider token strides give more accurate approximations at same token budget.finding0.716Cost tradeoff.