method
active
method:best-of-n-sampling-with-probe-scoringBest-of-N sampling with probe scoring
Test-time compute scaling guided by reward probes with majority-vote over action.
Neighborhood — ranked by edge-count
Papers (1)
paper
Frameworks (1)
framework
- Core pipeline turning probe-read features into scalable reward functions for open-ended tasks.
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.
- At n=256 best-of-n, probe scoring outperforms Gemma LLM-judge scoring by nearly 15 percentage pointsfinding0.789Test-time scaling comparison.
- Dot product between hidden state and concept vector averaged across 5-layer window around best layer; measures model's internal emotive state
- Attention probe predicting whether an Entity is hallucinated; AUC .94 on validation.
- Hypothesis generation.
- Interpretability tools that decode information from internal model activations; here, linear probes are used for data attribution.
- A technique to filter model outputs; Redwood Research's project mentioned.
- Probing approach avoiding supervision to sidestep complexity-accuracy tradeoff
- Key interpretive claim from Case Study II distinguishing probe accuracy from causal relevance