finding
active
finding:turn-averaged-features-flag-an-incorrect-answer-digit-exclusion-puzzle-while-per-token-features-find-none-of-the-error-recognitionTurn-averaged features flag an incorrect answer (digit-exclusion puzzle) while per-token features find none of the error recognition
Example A.
Source paper
extracted_from(2026) · Kevin Der · Harish Kamath · Thompson, Ben
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (2)
claim
- Central claim.
- Example A interpretation.
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.
- Prediction from coverage results.
- §5.5.
- LongBench v2 example E.
- Example B.
- Example C.
- Pairwise coverage.
- Turn-averaged features have highest embedding similarity 0.663 at k=3; per-token max lowest at 0.617finding0.802Consistent across k=1,2,3,5.
- Averaging acts as a low-pass filter that smooths away per-token detail and retains broad conceptsclaim0.799Intuition.