finding
active
finding:task-features-token-order-fourier-modes-are-more-present-in-larger-models-and-higher-frequency-runs-and-correlate-with-test-accuracyTask features (token order, Fourier modes) are more present in larger models and higher-frequency runs, and correlate with test accuracy
Fig. 7.
Source paper
extracted_from(2026) · Jing Huang · Daniel Wurgaft · Rachit Bansal · Laura Ruis +5
Neighborhood — ranked by edge-count
Papers (1)
paper
Hypotheses (1)
hypothesis
- Hypothesis box in §3.2.
Findings (1)
finding
- App. B.4.
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.
- Width buys features in order of utility, so only larger models learn infrequent or complex tasksclaim0.772Interpretation of Theorem 3.
- Fig. 13.
- With raw data, wider token strides give more accurate approximations at same token budget.finding0.770Cost tradeoff.
- With smoothed data, spacing and S trace the same Pareto frontier of accuracy vs token cost.finding0.769Cost tradeoff.
- Fig. 22.
- Poisoning results.
- When task gradient norms differ greatly, large-norm tasks have not converged while small-norm tasks have nearly convergedhypothesis0.764Motivates setting αk = max norm to enable further learning on under-converged tasks
- key claim about the benchmark's unique diagnostic value