List your company

Causal Faithfulness Evaluation on FineWeb Temporal

0.985nAOPCGAE
0.692760.768630.84450.92037May 21, 2026

Evaluation results

MethodLinks
GAEModel=Pythia-1.4B2026.050.9851.6770.0207
RetrainModel=Pythia-1.4B2026.050.9831.6810.0563
FaithfulSAEModel=Pythia-1.4B2026.050.9821.6420.0307
SAEBoostModel=Pythia-1.4B2026.050.9711.4510.0211
TERMModel=Pythia-1.4B2026.050.9651.4860.0237
FinetuneModel=Pythia-1.4B2026.050.9631.6180.0216
FixedModel=Pythia-1.4B2026.050.9621.5360.0216
GAEModel=GPT-2 Small2026.050.961.4940.0167
SAEBoostModel=GPT-2 Small2026.050.9581.4760.0177
FaithfulSAEModel=GPT-2 Small2026.050.9361.1860.0213
GAEModel=Pythia-1.4B2026.050.9151.3540.0269
SAEBoostModel=Pythia-1.4B2026.050.9081.2960.0284
RetrainModel=GPT-2 Small2026.050.8951.1180.0218
RetrainModel=Pythia-1.4B2026.050.8941.3150.0405
FinetuneModel=Pythia-1.4B2026.050.8591.2490.0264
FaithfulSAEModel=Pythia-1.4B2026.050.8581.0560.0272
FixedModel=GPT-2 Small2026.050.8571.0170.0281
FinetuneModel=GPT-2 Small2026.050.8560.9840.0172
TERMModel=GPT-2 Small2026.050.8530.9930.0283
TERMModel=Pythia-1.4B2026.050.8451.0430.0271
FixedModel=Pythia-1.4B2026.050.8391.0910.0278
GAEModel=GPT-2 Small2026.050.7680.8710.0011
RetrainModel=GPT-2 Small2026.050.7660.8560.0375
TERMModel=GPT-2 Small2026.050.7410.790.0183
FixedModel=GPT-2 Small2026.050.7350.7960.0185
FinetuneModel=GPT-2 Small2026.050.7250.8020.0015
FaithfulSAEModel=GPT-2 Small2026.050.7250.760.0262
SAEBoostModel=GPT-2 Small2026.050.7040.7860.012