finding
active
finding:with-sae-trained-on-96-6-pt-3-sft-0-4-rl-data-investigator-never-winsWith SAE trained on 96.6% PT, 3% SFT, 0.4% RL data investigator never wins
RM features at rank 181, outside top 100.
Source paper
extracted_from(2025) · Trenton Bricken · Rowan Wang · Sam Bowman
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Win rate correlates with SAE feature quality.
Hypotheses (2)
hypothesis
- Explains poor features with three-dataset SAE.
- Cf. 1/35 chat+grep win.
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.
- Prediction following diversity-undertraining logic.
- Fig 2 right.
- SAE latents depend on training data.
- Quantifies SAE reconstruction damage.
- Tool ablation.
- App. B.
- Realistic affordances result.
- No features correspond to Oxford study.