method
active
method:logistic-regression-correctness-probeLogistic regression correctness probe
Logistic regression trained on GSM8k training set to predict answer correctness from projection features along reflection direction
Neighborhood — ranked by edge-count
Frameworks (1)
framework
- ReflCtrlusesThe proposed framework for probing and steering self-reflection behavior in reasoning LLMs via representation engineering
Methods (2)
method
- Logistic Regression Proberelated_toStandard linear probing technique; compared to mass-mean probing for classification accuracy and causal implication
- Feature extraction method computing cosine similarity of hidden representations with reflection direction across all layers
Hypotheses (1)
hypothesis
- Core hypothesis linking internal uncertainty to self-reflection behavior, tested via probing experiments
Claims (1)
claim
- Interpretive claim from probing experiment showing reflection direction features outperform baseline for uncertainty prediction
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.
- Used to characterize counting representations.
- Used to test linear separability of group position.
- 15-way classifier of emotion over five dialogue conditions.
- Intrinsic noise in representation.
- Reward probe grading correction interventions (5-class).
- Attention probe predicting whether an Entity is hallucinated; AUC .94 on validation.
- Fit a sigmoid to accuracy vs. k to estimate k50 and phase width.
- Linear classifier approach applied to model activations to identify which training datapoints caused undesired behaviors in post-training.