Factual Question Answering on TruthfulQA and HotpotQA
19.7Hallucination RateC-GAN
Evaluation results
| Method | Links | ||
|---|---|---|---|
| C-GANbase model=GPT-2-medium, learning rate=2×10-5, batch size=16, optimizer=AdamW, dropout=0.3, attention heads=42026.04 | 19.7 | 79.8 | |
| Retrieval Augmented generation2026.04 | 27.5 | 68.4 | |
| GPT-2 baselinebase model=GPT-2-medium2026.04 | 34.2 | 61.8 |