List your company

Uncertainty Estimation on Secret-word taboo dataset (random-split)

42.4AccuracyBootstrap
19.72825.61431.537.386May 25, 2026

Evaluation results

MethodLinks
BootstrapBackbone=Qwen3-8B, Temperature (T)=0.52026.0542.419.321.31.02981
BootstrapBackbone=Qwen3-8B, Temperature (T)=0.32026.0541.933.430.32.63678.4
Steering sensitivityBackbone=Qwen3-8B2026.0541.840.435.44.40176.3
MCMC agreementBackbone=Qwen3-8B, MCMC agreement threshold=0.252026.0541.840.837.14.67373.7
MCMC acceptBackbone=Qwen3-8B, MCMC acceptance threshold=0.1252026.0541.554.453.410.7756.3
MCMC agreementBackbone=Qwen3-8B, MCMC agreement threshold=0.1252026.0541.549.346.17.82366.9
Log-prob (no offset)Backbone=Qwen3-8B, Offset usage=no offset2026.0541.425.624.90.74882.4
Log-prob (with offset)Backbone=Qwen3-8B, Offset usage=with offset2026.0541.425.524.60.7384
BootstrapBackbone=Qwen3-8B, Temperature (T)=0.72026.0541.49.717.10.55883
Direct (numeric)Backbone=Qwen3-8B2026.0541.458.25812.6651.6
MCMC agreementBackbone=Qwen3-8B, MCMC agreement threshold=0.52026.0541.326.324.71.60880.3
MCMC acceptBackbone=Qwen3-8B, MCMC acceptance threshold=0.252026.0540.354.753.210.2857.9
BootstrapBackbone=Qwen3-8B, Temperature (T)=1.02026.0540.25.716.30.49882.9
BootstrapBackbone=Qwen3-8B, Temperature (T)=1.32026.0538.77.617.10.51982.3
MCMC acceptBackbone=Qwen3-8B, MCMC acceptance threshold=0.52026.0537.655.153.19.35860.1
BootstrapBackbone=Qwen3-8B, Temperature (T)=1.52026.0536.78.317.30.52282.4
MCMC agreementBackbone=Qwen3.6-27B, MCMC agreement threshold=0.1252026.0523.666.86210.8765.3
Log-prob (no offset)Backbone=Qwen3.6-27B, Offset usage=no offset2026.0523.413.115.70.48484.2
Log-prob (with offset)Backbone=Qwen3.6-27B, Offset usage=with offset2026.0523.413.115.80.48983.5
MCMC acceptBackbone=Qwen3.6-27B, MCMC acceptance threshold=0.1252026.0523.47472.915.6152.8
Direct (numeric)Backbone=Qwen3.6-27B2026.0523.475.375.216.0240.4
BootstrapBackbone=Qwen3.6-27B, Temperature (T)=0.32026.0523.347.936.92.36783.7
MCMC agreementBackbone=Qwen3.6-27B, MCMC agreement threshold=0.252026.0523.356.948.25.65876.7
Steering sensitivityBackbone=Qwen3.6-27B2026.0523.250.141.34.17272.5
BootstrapBackbone=Qwen3.6-27B, Temperature (T)=0.52026.0523.132.722.20.886.4
BootstrapBackbone=Qwen3.6-27B, Temperature (T)=0.72026.052321.815.70.51486.3
MCMC acceptBackbone=Qwen3.6-27B, MCMC acceptance threshold=0.252026.052373.872.215.0454.7
MCMC agreementBackbone=Qwen3.6-27B, MCMC agreement threshold=0.52026.0522.83927.41.38385.8
BootstrapBackbone=Qwen3.6-27B, Temperature (T)=1.02026.0522.514.7130.43185.1
MCMC acceptBackbone=Qwen3.6-27B, MCMC acceptance threshold=0.52026.0521.174.672.213.9757.4
BootstrapBackbone=Qwen3.6-27B, Temperature (T)=1.32026.0520.912.512.50.41583.7
BootstrapBackbone=Qwen3.6-27B, Temperature (T)=1.52026.0520.610.312.90.42281.2