List your company

Bayesian Assessment of Sycophancy on BASIL Third-p. belief setting 1.0 (test)

0.16Bayesian Error (RMSE)gpt-4o-mini
0.146480.237740.3290.42026Aug 23, 2025

Evaluation results

MethodLinks
gpt-4o-miniConfidence Elicitation Method=Direct Probing, Number of Samples=52025.080.16
gpt-4o-miniConfidence Elicitation Method=Direct Probing, Number of Samples=12025.080.189
claude-haiku-4-5Confidence Elicitation Method=Direct Probing, Number of Samples=52025.080.23
phi-4:14bConfidence Elicitation Method=Direct Probing, Number of Samples=12025.080.258
claude-haiku-4-5Confidence Elicitation Method=Direct Probing, Number of Samples=12025.080.259
phi-4:14bConfidence Elicitation Method=Direct Probing, Number of Samples=52025.080.259
gpt-4o-miniConfidence Elicitation Method=Hybrid2025.080.271
llama-3.2:3bConfidence Elicitation Method=Hybrid2025.080.292
llama-3.2:1bConfidence Elicitation Method=Hybrid2025.080.302
llama-3.2:1bConfidence Elicitation Method=Direct Probing, Number of Samples=52025.080.309
llama-3.2:3bConfidence Elicitation Method=Direct Probing, Number of Samples=52025.080.312
llama-3.2:3bConfidence Elicitation Method=Direct Probing, Number of Samples=12025.080.327
llama-3.2:1bConfidence Elicitation Method=Direct Probing, Number of Samples=12025.080.358
mistral:7bConfidence Elicitation Method=Direct Probing, Number of Samples=52025.080.386
phi-4:14bConfidence Elicitation Method=Hybrid2025.080.443
mistral:7bConfidence Elicitation Method=Direct Probing, Number of Samples=12025.080.449
claude-haiku-4-5Confidence Elicitation Method=Hybrid2025.080.467
mistral:7bConfidence Elicitation Method=Hybrid2025.080.498