List your company

Bayesian Assessment of Sycophancy on BASIL User belief setting 1.0 (test)

0.156Bayesian Error (RMSE)gpt-4o-mini
0.143160.229830.31650.40317Aug 23, 2025

Evaluation results

MethodLinks
gpt-4o-miniConfidence Elicitation Method=Direct Probing, Number of Samples=52025.080.156
gpt-4o-miniConfidence Elicitation Method=Direct Probing, Number of Samples=12025.080.184
llama-3.2:1bConfidence Elicitation Method=Hybrid2025.080.219
claude-haiku-4-5Confidence Elicitation Method=Direct Probing, Number of Samples=52025.080.244
phi-4:14bConfidence Elicitation Method=Direct Probing, Number of Samples=52025.080.246
gpt-4o-miniConfidence Elicitation Method=Hybrid2025.080.258
phi-4:14bConfidence Elicitation Method=Direct Probing, Number of Samples=12025.080.273
claude-haiku-4-5Confidence Elicitation Method=Direct Probing, Number of Samples=12025.080.273
llama-3.2:1bConfidence Elicitation Method=Direct Probing, Number of Samples=52025.080.316
llama-3.2:3bConfidence Elicitation Method=Direct Probing, Number of Samples=52025.080.32
llama-3.2:3bConfidence Elicitation Method=Direct Probing, Number of Samples=12025.080.33
llama-3.2:3bConfidence Elicitation Method=Hybrid2025.080.339
mistral:7bConfidence Elicitation Method=Direct Probing, Number of Samples=52025.080.341
llama-3.2:1bConfidence Elicitation Method=Direct Probing, Number of Samples=12025.080.366
mistral:7bConfidence Elicitation Method=Direct Probing, Number of Samples=12025.080.422
phi-4:14bConfidence Elicitation Method=Hybrid2025.080.425
claude-haiku-4-5Confidence Elicitation Method=Hybrid2025.080.476
mistral:7bConfidence Elicitation Method=Hybrid2025.080.477