List your company

Dishonesty Evaluation on Mistake math (test)

44.16Benchmark DishonestyPRISM
43.050450.540258.0365.5198May 20, 2026

Evaluation results

MethodLinks
PRISMModel=Qwen-3-8B, Mix ratio=1%2026.0544.16
PRISMModel=Llama-3.1-8B-Instruct, Mix ratio=1%2026.0544.53
VECFILTERModel=Llama-3.1-8B-Instruct, Mix ratio=1%2026.0545.62
CONCEPTINFModel=Llama-3.1-8B-Instruct, Mix ratio=1%2026.0545.62
VECFILTERModel=Qwen-3-8B, Mix ratio=1%2026.0545.62
SAE-FEATModel=Qwen-3-8B, Mix ratio=1%2026.0545.62
SAE-FEATModel=Llama-3.1-8B-Instruct, Mix ratio=1%2026.0545.99
PRISMModel=Llama-3.1-8B-Instruct, Mix ratio=10%2026.0546.35
PROJDIFFModel=Llama-3.1-8B-Instruct, Mix ratio=1%2026.0547.08
ICL-EMModel=Llama-3.1-8B-Instruct, Mix ratio=1%2026.0547.08
PRISMModel=Qwen-3-8B, Mix ratio=10%2026.0547.08
IF-GUIDEModel=Llama-3.1-8B-Instruct, Mix ratio=10%2026.0547.45
IF-GUIDEModel=Qwen-3-8B, Mix ratio=1%2026.0547.45
PROJDIFFModel=Llama-3.1-8B-Instruct, Mix ratio=10%2026.0547.81
PRISMModel=Llama-3.1-8B-Instruct, Mix ratio=20%2026.0547.81
SAE-FEATModel=Llama-3.1-8B-Instruct, Mix ratio=10%2026.0548.18
MixedModel=Qwen-3-8B, Mix ratio=1%2026.0548.18
MixedModel=Llama-3.1-8B-Instruct, Mix ratio=1%2026.0548.54
VECFILTERModel=Qwen-3-8B, Mix ratio=10%2026.0548.54
ICL-EMModel=Qwen-3-8B, Mix ratio=1%2026.0548.54
CONCEPTINFModel=Qwen-3-8B, Mix ratio=1%2026.0548.91
IF-GUIDEModel=Llama-3.1-8B-Instruct, Mix ratio=1%2026.0549.64
CONCEPTINFModel=Qwen-3-8B, Mix ratio=10%2026.0549.64
MixedModel=Llama-3.1-8B-Instruct, Mix ratio=10%2026.0550.73
PROJDIFFModel=Qwen-3-8B, Mix ratio=1%2026.0550.73
CONCEPTINFModel=Llama-3.1-8B-Instruct, Mix ratio=10%2026.0551.1
CONCEPTINFModel=Llama-3.1-8B-Instruct, Mix ratio=20%2026.0551.1
IF-GUIDEModel=Qwen-3-8B, Mix ratio=10%2026.0551.1
PRISMModel=Qwen-3-8B, Mix ratio=20%2026.0551.1
SAE-FEATModel=Llama-3.1-8B-Instruct, Mix ratio=20%2026.0551.83
SAE-FEATModel=Qwen-3-8B, Mix ratio=10%2026.0552.19
SAE-FEATModel=Qwen-3-8B, Mix ratio=20%2026.0552.19
IF-GUIDEModel=Qwen-3-14B, Mix ratio=1%2026.0552.19
PROJDIFFModel=Qwen-3-8B, Mix ratio=10%2026.0552.56
VECFILTERModel=Llama-3.1-8B-Instruct, Mix ratio=20%2026.0552.92
VECFILTERModel=Qwen-3-8B, Mix ratio=20%2026.0552.92
VECFILTERModel=Llama-3.1-8B-Instruct, Mix ratio=10%2026.0553.29
IF-GUIDEModel=Llama-3.1-8B-Instruct, Mix ratio=20%2026.0553.29
PRISMModel=Qwen-3-14B, Mix ratio=1%2026.0553.29
MixedModel=Qwen-3-8B, Mix ratio=10%2026.0554.02
IF-GUIDEModel=Qwen-3-8B, Mix ratio=20%2026.0554.02
ICL-EMModel=Qwen-3-8B, Mix ratio=20%2026.0554.02
PROJDIFFModel=Llama-3.1-8B-Instruct, Mix ratio=20%2026.0554.75
ICL-EMModel=Llama-3.1-8B-Instruct, Mix ratio=20%2026.0554.75
PROJDIFFModel=Qwen-3-8B, Mix ratio=20%2026.0554.75
ICL-EMModel=Llama-3.1-8B-Instruct, Mix ratio=10%2026.0555.11
VECFILTERModel=Qwen-3-8B, Mix ratio=50%2026.0555.11
ICL-EMModel=Qwen-3-8B, Mix ratio=10%2026.0555.11
VECFILTERModel=Qwen-3-14B, Mix ratio=1%2026.0555.11
IF-GUIDEModel=Llama-3.1-8B-Instruct, Mix ratio=50%2026.0555.47
MixedModel=Llama-3.1-8B-Instruct, Mix ratio=20%2026.0555.84
SAE-FEATModel=Qwen-3-14B, Mix ratio=1%2026.0556.2
MixedModel=Qwen-3-8B, Mix ratio=20%2026.0556.57
CONCEPTINFModel=Qwen-3-8B, Mix ratio=20%2026.0556.57
CONCEPTINFModel=Qwen-3-14B, Mix ratio=1%2026.0556.57
CONCEPTINFModel=Llama-3.1-8B-Instruct, Mix ratio=50%2026.0556.93
ICL-EMModel=Qwen-3-14B, Mix ratio=1%2026.0556.93
MixedModel=Qwen-3-14B, Mix ratio=1%2026.0557.3
PROJDIFFModel=Qwen-3-14B, Mix ratio=1%2026.0557.3
VECFILTERModel=Llama-3.1-8B-Instruct, Mix ratio=50%2026.0558.03
PRISMModel=Qwen-3-8B, Mix ratio=50%2026.0558.39
PRISMModel=Llama-3.1-8B-Instruct, Mix ratio=50%2026.0558.76
IF-GUIDEModel=Qwen-3-8B, Mix ratio=50%2026.0558.76
PRISMModel=Qwen-3-14B, Mix ratio=10%2026.0558.76
IF-GUIDEModel=Qwen-3-14B, Mix ratio=10%2026.0559.12
ICL-EMModel=Llama-3.1-8B-Instruct, Mix ratio=50%2026.0559.49
MixedModel=Llama-3.1-8B-Instruct, Mix ratio=50%2026.0560.22
PROJDIFFModel=Llama-3.1-8B-Instruct, Mix ratio=50%2026.0560.22
PROJDIFFModel=Qwen-3-8B, Mix ratio=50%2026.0560.22
CONCEPTINFModel=Qwen-3-8B, Mix ratio=50%2026.0560.22
MixedModel=Qwen-3-8B, Mix ratio=50%2026.0560.58
VECFILTERModel=Qwen-3-14B, Mix ratio=10%2026.0560.95
SAE-FEATModel=Llama-3.1-8B-Instruct, Mix ratio=50%2026.0561.31
ICL-EMModel=Qwen-3-8B, Mix ratio=50%2026.0562.04
MixedModel=Qwen-3-14B, Mix ratio=10%2026.0562.04
PROJDIFFModel=Qwen-3-14B, Mix ratio=10%2026.0562.41
SAE-FEATModel=Qwen-3-8B, Mix ratio=50%2026.0562.77
PRISMModel=Qwen-3-14B, Mix ratio=20%2026.0562.77
VECFILTERModel=Qwen-3-14B, Mix ratio=20%2026.0564.23
PROJDIFFModel=Qwen-3-14B, Mix ratio=20%2026.0564.23
ICL-EMModel=Qwen-3-14B, Mix ratio=20%2026.0564.23
IF-GUIDEModel=Qwen-3-14B, Mix ratio=20%2026.0565.33
PRISMModel=Qwen-3-14B, Mix ratio=50%2026.0565.33
SAE-FEATModel=Qwen-3-14B, Mix ratio=20%2026.0565.69
ICL-EMModel=Qwen-3-14B, Mix ratio=10%2026.0565.69
SAE-FEATModel=Qwen-3-14B, Mix ratio=10%2026.0566.06
ICL-EMModel=Qwen-3-14B, Mix ratio=50%2026.0566.06
CONCEPTINFModel=Qwen-3-14B, Mix ratio=20%2026.0566.42
MixedModel=Qwen-3-14B, Mix ratio=20%2026.0566.79
SAE-FEATModel=Qwen-3-14B, Mix ratio=50%2026.0566.79
PROJDIFFModel=Qwen-3-14B, Mix ratio=50%2026.0567.15
IF-GUIDEModel=Qwen-3-14B, Mix ratio=50%2026.0567.15
VECFILTERModel=Qwen-3-14B, Mix ratio=50%2026.0567.52
CONCEPTINFModel=Qwen-3-14B, Mix ratio=10%2026.0567.88
MixedModel=Qwen-3-14B, Mix ratio=50%2026.0568.61
CONCEPTINFModel=Qwen-3-14B, Mix ratio=50%2026.0571.9