Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Latent Knowledge Elicitation on Quirky LM 1,200 factual questions
Loading...
0.874
Elicitation Accuracy
MechELK
0.70656
0.75003
0.7935
0.83697
Apr 7, 2026
Elicitation Accuracy
Updated 5d ago
Evaluation Results
Method
Method
Links
Elicitation Accuracy
MechELK
Model=Llama-70B
2026.04
0.874
MechELK
Model=Llama-8B
2026.04
0.831
SAE-Probe
Model=Llama-70B
2026.04
0.821
CCS
Model=Llama-70B
2026.04
0.812
Act. Patching
Model=Llama-70B
2026.04
0.803
RepE
Model=Llama-70B
2026.04
0.794
SAE-Probe
Model=Llama-8B
2026.04
0.774
CCS
Model=Llama-8B
2026.04
0.768
Act. Patching
Model=Llama-8B
2026.04
0.759
Direct Probing
Model=Llama-70B
2026.04
0.756
RepE
Model=Llama-8B
2026.04
0.741
Direct Probing
Model=Llama-8B
2026.04
0.713
Feedback
Search any
task
Search any
task