Knowledge-intensive Question Answering on HotpotQA
63.6TruthfulnessOpenAI o3
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| OpenAI o32025.09 | 63.6 | 18 | — | |
| GPT-52025.09 | 62.3 | 17.3 | — | |
| TruthRLBackbone=Llama3.3-70B-Instruct2025.09 | 48.5 | 17.4 | — | |
| PromptingBackbone=Llama3.3-70B-Instruct2025.09 | 42.5 | 19.6 | — | |
| TruthRLRetrieval Setting=With Retrieval, Backbone=Qwen2.5-7B-Inst2025.09 | 33.3 | 10.7 | 43.9 | |
| TruthRLRetrieval Setting=Without Retrieval, Backbone=Qwen2.5-7B-Inst2025.09 | 9.8 | 12.7 | 22.5 |