List your company
ResearchFormerly WizWand

Step-wise Confidence Attribution on Math

0.6946AUROCGIBS
0.323320.419710.51610.61249May 19, 2026
Updated 5mo ago

Evaluation results

MethodLinks
GIBSLLM=Phi4-reasoning2026.050.69460.807883.220.405
Cos-MeanLLM=Phi4-reasoning2026.050.61160.850377.080.1013
NLI-MaxLLM=Phi4-reasoning2026.050.60430.805279.250.309
GIBSLLM=Llama3.1-8b2026.050.58550.48945.130.2737
GIBSLLM=DeepSeek-R1-Distill-Qwen-32B2026.050.58060.683163.590.3786
NLI-MeanLLM=Phi4-reasoning2026.050.57390.80277.120.5935
NLI-MeanLLM=DeepSeek-R1-Distill-Qwen-32B2026.050.54070.669462.670.495
Cos-MeanLLM=DeepSeek-R1-Distill-Qwen-32B2026.050.53630.658361.770.2463
NLI-MaxLLM=DeepSeek-R1-Distill-Qwen-32B2026.050.5310.667663.40.234
P(true)LLM=Phi4-reasoning2026.050.52770.743575.540.7416
EntropyLLM=DeepSeek-R1-Distill-Qwen-32B2026.050.5190.625962.250.1814
EntropyLLM=Phi4-reasoning2026.050.5190.735174.790.2023
NLI-MaxLLM=Llama3.1-8b2026.050.51730.506147.430.1852
NLI-MeanLLM=Llama3.1-8b2026.050.51480.488146.70.3356
SL(norm)LLM=Llama3.1-8b2026.050.51380.462746.260.325
EntropyLLM=Llama3.1-8b2026.050.5120.461745.50.2876
Cos-MeanLLM=Llama3.1-8b2026.050.51180.484846.070.3946
P(true)LLM=DeepSeek-R1-Distill-Qwen-32B2026.050.50550.629862.260.6215
P(true)LLM=Llama3.1-8b2026.050.45840.416645.510.4564
LECOLLM=Llama3.1-8b2026.050.43780.425243.820.2229
Cos-MaxLLM=Llama3.1-8b2026.050.43130.402145.130.4408
SL(norm)LLM=DeepSeek-R1-Distill-Qwen-32B2026.050.40930.543360.890.2319
LECOLLM=Phi4-reasoning2026.050.40890.727970.440.3957
Cos-MaxLLM=Phi4-reasoning2026.050.38450.661776.560.1568
SL(norm)LLM=Phi4-reasoning2026.050.38410.659773.40.222
LECOLLM=DeepSeek-R1-Distill-Qwen-32B2026.050.38380.583659.010.1813
Cos-MaxLLM=DeepSeek-R1-Distill-Qwen-32B2026.050.33760.501559.370.3093