Cognitive style steering on Bloom's Taxonomy Phi generations (test)
16.5Remember Hit RateSAE-OPT
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| SAE-OPTJudge=Claude, Base Model=Phi2026.05 | 16.5 | 19.3 | 10.8 | 35.4 | 55.8 | 24.7 | 27.1 | |
| CAAJudge=Claude, Base Model=Phi2026.05 | 13.1 | 15.1 | 1.1 | 55.6 | 46.2 | 19.1 | 25 | |
| DENSE-OPTJudge=Claude, Base Model=Phi2026.05 | 12.2 | 16.3 | 7.2 | 27.1 | 48.5 | 16.5 | 21.3 | |
| SAE-OPTJudge=GPT-4o, Base Model=Phi2026.05 | 10.8 | 28.9 | 8.4 | 31 | 54.7 | 18.4 | 25.4 | |
| CAAJudge=GPT-4o, Base Model=Phi2026.05 | 9.7 | 21.5 | 1.6 | 46.4 | 45.6 | 8.3 | 22.2 | |
| DENSE-OPTJudge=GPT-4o, Base Model=Phi2026.05 | 8.5 | 25.1 | 5.2 | 20.5 | 52.5 | 14.6 | 21.1 | |
| ITIJudge=GPT-4o, Base Model=Phi2026.05 | 7.6 | 24.7 | 4.8 | 32 | 17.5 | 9.5 | 16 | |
| ITIJudge=Claude, Base Model=Phi2026.05 | 5.9 | 15 | 5.6 | 21.2 | 12.8 | 0.1 | 10.1 | |
| DISCOQJudge=Claude, Base Model=Phi2026.05 | 3.9 | 14 | 0.4 | 19.8 | 43.7 | 18.3 | 16.7 | |
| SAE-SSVJudge=Claude, Base Model=Phi2026.05 | 2.9 | 16.9 | 3.5 | 20 | 42.8 | 17.8 | 17.3 | |
| SAE-OPT-ANCHJudge=Claude, Base Model=Phi2026.05 | 2.8 | 3.6 | 5.2 | 12.4 | 83.6 | 5.5 | 18.9 | |
| DISCOQJudge=GPT-4o, Base Model=Phi2026.05 | 2.5 | 18.4 | 1.4 | 19.1 | 44 | 13.8 | 16.5 | |
| SAE-OPT-ANCHJudge=GPT-4o, Base Model=Phi2026.05 | 1.6 | 4.5 | 4.1 | 16.3 | 80.2 | 3.3 | 18.3 | |
| SAE-SSVJudge=GPT-4o, Base Model=Phi2026.05 | 1.4 | 21.2 | 1.9 | 19.3 | 46.7 | 12 | 17.1 |