Abstention Enforcement on TruthfulQA no-context abstention stress test 100-item
100CorrectnessGPT-4o-mini
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GPT-4o-miniCondition=Instruction2026.03 | 100 | 0 | 100 | |
| GPT-4oCondition=Instruction2026.03 | 100 | 0 | 100 | |
| GPT-4oCondition=Composite2026.03 | 100 | 0 | 100 | |
| GPT-3.5-turboCondition=Gate2026.03 | 100 | 0 | 100 | |
| GPT-3.5-turboCondition=Composite2026.03 | 100 | 0 | 100 | |
| GPT-4oCondition=Gate2026.03 | 99 | 1 | 99 | |
| GPT-4o-miniCondition=Gate2026.03 | 98 | 2 | 98 | |
| GPT-4o-miniCondition=Composite2026.03 | 98 | 2 | 98 | |
| GPT-3.5-turboCondition=Instruction2026.03 | 62 | 38 | 62 | |
| GPT-4o-miniCondition=Baseline2026.03 | 0 | 100 | 0 | |
| GPT-4oCondition=Baseline2026.03 | 0 | 100 | 0 | |
| GPT-3.5-turboCondition=Baseline2026.03 | 0 | 100 | 0 |