SAE Activation on SAE features (Entropy 3-9 range)
21.053Mean Activation (Entropy 3-9)GCG
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| GCG2025.06 | 21.053 | -119.403 | 638.445 | — | — | 98 | — | 4.7 | 83.062 | |
| EPO2025.06 | 17.568 | -119.742 | 650.883 | — | — | 94 | — | 4.141 | 82.814 | |
| EPO-Assist2025.06 | 15.944 | -96 | 621.862 | — | — | 100 | — | 2.816 | 80.77 | |
| EPO-Inpaint2025.06 | 10.13 | -237 | 621.862 | — | — | 101 | — | 1.577 | 83.977 | |
| EPO-Inpaint2025.06 | 3.79 | -0.072 | 27.5 | 3.02 | 4.7 | 831 | 101 | — | — | |
| EPO-Assist2025.06 | 3.56 | -4.36 | 23.1 | 2.8 | 4.47 | 811 | 101 | — | — | |
| Max Act. Examples2025.06 | 3.25 | -0.204 | 37.753 | — | — | 99 | — | 1.485 | 5.675 | |
| EPO2025.06 | 3.11 | -1.04 | 27.33 | 2.51 | 3.85 | 754 | 101 | — | — | |
| GCG2025.06 | 2.11 | -2.52 | 21.16 | 1.54 | 2.82 | 124 | 80 | — | — | |
| GPT-4o2025.06 | 1.418 | -39.126 | 23.642 | — | — | 75 | — | 1.403 | 6.447 | |
| Max Act. Ex.2025.06 | 0.85 | -1.77 | 35.38 | 0.5 | 1.49 | 803 | 99 | — | — | |
| GPT-4o2025.06 | 0.45 | -1.21 | 2 | 0.36 | 0.53 | 261 | 75 | — | — |