finding
active
finding:on-t-add-only-larger-models-trained-at-higher-frequency-exhibit-grokkingOn T_ADD only larger models trained at higher frequency exhibit grokking
§4.2.
Source paper
extracted_from(2026) · Jing Huang · Daniel Wurgaft · Rachit Bansal · Laura Ruis +5
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Interpretation of OLMo T_ADD results.
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.
- Fig. 5.
- Models trained directly with asynchronous updates would exhibit even greater robustness than synchronously trained modelshypothesis0.762Hypothesis that motivated the asynchronous robustness comparison experiment
- Prior finding from related work that aligns with ESR being strongest in the largest model tested
- Core framing.
- Fig. 7.
- Faithfulness decreases for easier problems and bigger models.
- AOs never saw fine-tuned model activations in training.
- Comparative prediction motivating future work contrasting different approaches to LLM self-knowledge