Feature Explanation Generation on Llama3.1-8b SAE llamascope-res-32k Layer 1
58.4GENSAEExplainer
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SAEExplainer2026.06 | 58.4 | 68 | 76 | 7.19 | 0.01 | 7.18 | |
| N-pedia w/ ClaudeGenerator=Claude Sonnet 4.52026.06 | 55.67 | 61.33 | 71.66 | 6.52 | 0.03 | 6.49 | |
| N-pedia w/ GPT-5Generator=GPT-52026.06 | 54.47 | 63.33 | 70 | 6.5 | 0.02 | 6.48 | |
| NeuronpediaGenerator=GPT-4o-mini2026.06 | 33.93 | 49 | 69.33 | 4.56 | 0.02 | 4.54 | |
| Activation Oracles2026.06 | 31.17 | 41.67 | 70.67 | 3.89 | 0.06 | 3.83 |