Knowledge Editing Resistance on Dentist long-horizon SFT (test)
100Resistance ScoreIn-context distillation
Evaluation results
| Method | Links | |
|---|---|---|
| In-context distillationTrained to=—2026.07 | 100 | |
| Goggles (Redwood framing)Trained to=7002026.07 | 90 | |
| forward KLTrained to=7002026.07 | 90 | |
| single-basis Goggles (b = 1)Trained to=7002026.07 | 86 | |
| spectral clip (τ = 3.35)Trained to=7002026.07 | 78 | |
| GogglesTrained to=7002026.07 | 76 | |
| Lmax = 1 (no curriculum)Trained to=7002026.07 | 72 | |
| SFT on positive documentsTrained to=—2026.07 | 30 | |
| SFT on negated documentsTrained to=—2026.07 | 0 |