List your company

Error detection on Held-out problems (n=200) (test)

0.953AUROCHidden State Probe
0.662840.738170.81350.88883May 10, 2026

Evaluation results

MethodLinks
Hidden State ProbeCost=1 fwd pass, Model=Qwen2.5-3B2026.050.953
Self-Consistency (N=5)Cost=5× gen, Model=Qwen2.5-3B2026.050.823
P(True)Cost=1 query, Model=Qwen2.5-3B2026.050.721
CCSCost=1 fwd pass, Model=Qwen2.5-3B2026.050.718
Sequence Log-probCost=free, Model=Qwen2.5-3B2026.050.676
Verbalized ConfidenceCost=1 query, Model=Qwen2.5-3B2026.050.674