finding
active
finding:counterfactual-prediction-training-improves-accuracy-and-auroc-on-held-out-hint-setting-for-qwen3-8b-and-qwen3-5-397b-approaching-opus-referenceCounterfactual prediction training improves accuracy and AUROC on held-out hint setting for Qwen3-8B and Qwen3.5-397B, approaching Opus reference
Generalization to hint setting.
Source paper
extracted_from(2026) · Adam Karvonen · Euan Ong · Subhash Kantamneni · Samuel Marks
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Hase & Potts +35pp vs +2pp.
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.
- Base failure mode.
- Mixed results.
- In- and out-of-distribution generalization.
- Core empirical finding about layer-dependent truth direction emergence across task types.
- Key improvement in cross-task generalization enabled by explicit instruction framing.
- Core empirical result for animal welfare setting; higher rate than helpful-only
- Trained models slightly exceed Opus reference on unfiltered bank (+0.015 Qwen3-8B, +0.033 Qwen3.5-397B)finding0.768Gains concentrated in low-concreteness claims.
- Fig. 9c, Fig. 37.