Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
SAE label evaluation on Goodfire SAE latents (held-out)
Loading...
71.4
Hit Rate
Trained Adapter (SA)
48.208
54.229
60.25
66.271
Feb 10, 2026
Hit Rate
Coverage
Updated 1mo ago
Evaluation Results
Method
Method
Links
Hit Rate
Coverage
Trained Adapter (SA)
Model=Llama-3.3-70B-In...
2026.02
71.4
83.2
Trained Adapter (SA+LR)
Model=Llama-3.3-70B-In...
2026.02
68.3
82.2
Original + 5 Paraphrases
Model=Llama-3.3-70B-In...
2026.02
63.4
79.8
Auto-Interp Labels x6
Model=Llama-3.3-70B-In...
2026.02
54.8
71.5
Untrained SelfIE
Model=Llama-3.3-70B-In...
2026.02
49.1
61.9
Feedback
Search any
task
Search any
task