List your company

Causal Faithfulness Evaluation on Edgar Domain

0.988nAOPCGAE
0.588640.692320.7960.89968May 21, 2026

Evaluation results

MethodLinks
GAEModel=Pythia-1.4B2026.050.9881.6520.023
GAEModel=GPT-2 Small2026.050.9811.6180.0009
SAEBoostModel=GPT-2 Small2026.050.9791.5420.0072
FaithfulSAEModel=GPT-2 Small2026.050.9761.1340.0197
FixedModel=GPT-2 Small2026.050.9751.0250.0201
FinetuneModel=GPT-2 Small2026.050.9711.1170.0047
RetrainModel=Pythia-1.4B2026.050.971.8480.0261
GAEModel=Pythia-1.4B2026.050.9681.9460.0098
TERMModel=GPT-2 Small2026.050.9640.9990.0198
TERMModel=Pythia-1.4B2026.050.9591.7870.0167
FinetuneModel=Pythia-1.4B2026.050.9591.8320.0131
SAEBoostModel=Pythia-1.4B2026.050.9551.5980.0152
FixedModel=Pythia-1.4B2026.050.9531.690.017
FaithfulSAEModel=Pythia-1.4B2026.050.9531.7690.0205
RetrainModel=GPT-2 Small2026.050.9361.0340.0015
SAEBoostModel=Pythia-1.4B2026.050.9031.2970.0296
TERMModel=Pythia-1.4B2026.050.8950.9810.0298
RetrainModel=Pythia-1.4B2026.050.7460.8210.0329
FixedModel=Pythia-1.4B2026.050.7250.8280.03
FaithfulSAEModel=Pythia-1.4B2026.050.7240.8220.0323
GAEModel=GPT-2 Small2026.050.7230.8090.0037
RetrainModel=GPT-2 Small2026.050.7150.7970.0065
FinetuneModel=Pythia-1.4B2026.050.6840.860.0282
FinetuneModel=GPT-2 Small2026.050.6580.6870.0068
FaithfulSAEModel=GPT-2 Small2026.050.6570.6830.0074
TERMModel=GPT-2 Small2026.050.6550.6820.0145
FixedModel=GPT-2 Small2026.050.650.6670.0089
SAEBoostModel=GPT-2 Small2026.050.6040.650.0098