finding
active
finding:openbookqa-interacts-far-less-with-bottom-spectrum-than-memorized-data-memorized-1-6x-higherOpenBookQA interacts far less with bottom spectrum than memorized data (memorized ~1.6x higher)
Activation ratios predict robustness.
Source paper
extracted_from(2025) · Merullo, Jack · Vatsavaya, Srihita · Bushnaq, Lucius · Lewis, Owen
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Discussion.
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.
- Spectrum of brittleness.
- Evidence against test-time compute hypothesis.
- Shows VS not only maintains but can slightly improve factual accuracy compared to baseline methods
- Key insight about predictive learning's potential.
- With raw data, wider spacing has little effect on accuracy except at high S and near forking points.finding0.721Raw data behavior.
- Load-bearing motivation for multimodal approach; frames the cognitive advantage of joint modalities.
- Key cross-modal alignment result
- Evidence that Multimodal-CoT can operate without human-annotated reasoning chains by using large models to generate pseudo-rationales.