Research map / Large language models / Retrieval and knowledge
Teaching Large Language Models to Maintain Contextual Faithfulness via Synthetic Tasks and Reinforcement Learning
Shuzheng Si, Haozhe Zhao, Cheng Gao, Yuzhuo Bai, Zhitong Wang, Bofei Gao, Kangyang Luo, Wenhao Li, Yufei Huang, Gang Chen, Fanchao Qi, Minjia Zhang and 2 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Dual-GRPO and short-form data synthesis (ablation) | Canoe-LLaMA-8B · ablation short-form tasks | EM: 36.3 -> 67.7 | Improved |
| add Dual-GRPO and short-form data synthesis (ablation) | Canoe-LLaMA-8B · ablation long-form tasks | QualityScore: 64.3 -> 79.7 | Improved |
| use Dual-GRPO instead of original GRPO (ablation) | Canoe-LLaMA-8B · ablation short-form tasks | EM: 60.5 -> 67.7 | Improved |
| use Dual-GRPO instead of original GRPO (ablation) | Canoe-LLaMA-8B · ablation long-form tasks | QualityScore: 23.5 -> 79.7 | Improved |
| add reasoning-required context (ablation) | Canoe-LLaMA-8B · ablation short-form tasks | EM: 63.7 -> 67.7 | Improved |
| add inconsistent context (ablation) | Canoe-LLaMA-8B · ablation short-form tasks | EM: 64.4 -> 67.7 | Improved |
| add counterfactual context (ablation) | Canoe-LLaMA-8B · ablation short-form tasks | EM: 62.6 -> 67.7 | Improved |
| add counterfactual context (ablation) | Canoe-LLaMA-8B · ablation long-form tasks | QualityScore: 73.7 -> 79.7 | Improved |
| add inconsistent context (ablation) | Canoe-LLaMA-8B · ablation long-form tasks | QualityScore: 72.5 -> 79.7 | Improved |
| add reasoning-required context (ablation) | Canoe-LLaMA-8B · ablation long-form tasks | QualityScore: 75.3 -> 79.7 | Improved |
| apply CANOE RL training | LLaMA-3-Instruct-8B · 11 faithfulness tasks | Avg EM: 47.7 -> 70.3 | Improved |
| apply CANOE RL training | Qwen-2.5-Instruct-7B · 11 faithfulness tasks | Avg EM: 49.0 -> 68.0 | Improved |
Closest papers by meaning
- SPELL: Self-Play Reinforcement Learning for Evolving Long-Context Language Models (ICLR 2026 · similarity 0.89)
- LongRLVR: Long-Context Reinforcement Learning Requires Verifiable Context Rewards (ICLR 2026 · similarity 0.89)
- Probing to Refine: Reinforcement Distillation of LLM Reasoners via Explanatory Inversion (ICLR 2026 · similarity 0.89)
- Thinker: Learning to Think Fast and Slow (NeurIPS 2025 · similarity 0.88)
- Webscale-RL: Automated Data Pipeline for Scaling RL Data to Pretraining Levels (ICLR 2026 · similarity 0.88)
- Supervised Knowledge Makes Large Language Models Better In-context Learners (ICLR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
