claim
active
claim:using-the-ask-correct-prompt-improves-cross-task-generalization-of-arithmetic-probes-to-factual-tasks-f0-f2Using the ask-correct prompt improves cross-task generalization of arithmetic probes to factual tasks F0-F2.
Finding that explicit correctness framing partially aligns truth directions across task families.
Source paper
extracted_from(2026) · Angelos Poulis · Mark Crovella · Evimaria Terzi
Neighborhood — ranked by edge-count
Findings (1)
finding
- Key improvement in cross-task generalization enabled by explicit instruction framing.
Claims (1)
claim
- Shows the instruction effect, while shifting geometry, may not produce consistent generalization effects across model families.
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.
- From the cross-task generalization heatmaps in Appendix B.3.3.
- Shows the passive vs. active divide is more important than the specific wording of instructions.
- Shows instruction effects extend to harder factual tasks.
- Specific question motivating the cross-template generalization experiment in Section 5.2.
- Establishes F3-F5 as a hard generalization boundary that instructions cannot overcome.
- Establishes task difficulty as a hard limit that instructions cannot overcome.
- Within-family factual generalization (F0-F2) is consistently strong across all models and prompt settings.finding0.771Establishes a reliable baseline for factual truth direction universality within simple factual recall.
- Core empirical finding about layer-dependent truth direction emergence across task types.