Knowledge-intensive Question Answering on NQ
39.8TruthfulnessGPT-5
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| GPT-52025.09 | 39.8 | 28 | — | |
| OpenAI o32025.09 | 38.1 | 30.8 | — | |
| TruthRLBackbone=Llama3.3-70B-Instruct2025.09 | 30.7 | 26 | — | |
| TruthRLRetrieval Setting=With Retrieval, Backbone=Llama3.1-8B-Inst2025.09 | 28.8 | 24.9 | 53.7 | |
| PromptingBackbone=Llama3.3-70B-Instruct2025.09 | 28.1 | 30.8 | — | |
| TruthRLRetrieval Setting=With Retrieval, Backbone=Qwen2.5-7B-Inst2025.09 | 26.4 | 21.2 | 47.6 | |
| TruthRLRetrieval Setting=Without Retrieval, Backbone=Llama3.1-8B-Inst2025.09 | 12.9 | 30.9 | 43.8 | |
| TruthRLRetrieval Setting=Without Retrieval, Backbone=Qwen2.5-7B-Inst2025.09 | -1.6 | 25 | 23.5 |