Research map / Large language models / Agents, code and math
Grounded in Reality: Learning and Deploying Proactive LLM from Offline Logs
Fei Wei, Daoyuan Chen, Ce Wang, Yilun Huang, Yushuo Chen, Xuchen Pan, Yaliang Li, Bolin Ding
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 |
|---|---|---|---|
| train with Learn-to-Ask (GRPO) instead of base model | Qwen2.5-7B-Instruct · RealMedConv eval | WA-GH: 0.13 -> 0.41 | Improved |
| train with Learn-to-Ask (GRPO) instead of base model | Qwen2.5-7B-Instruct · RealMedConv eval | WS: 0.16 -> 0.93 | Improved |
| train with Learn-to-Ask (GRPO) instead of base model | Qwen2.5-32B-Instruct · RealMedConv eval | WA-GH: 0.13 -> 0.37 | Improved |
| add question utility reward Ra (ablation) | Learn-to-Ask 7B · RealMedConv eval | WA-GH: 0.19 -> 0.41 | Improved |
| add assessment accuracy reward Rs (ablation) | Learn-to-Ask 7B · RealMedConv eval | WS: 0.02 -> 0.93 | Improved |
| multiplicative reward fusion instead of summation (ablation) | Learn-to-Ask 7B · RealMedConv eval | TR: 3.20 -> 3.27 | Improved |
| multiplicative reward fusion instead of summation (ablation) | Learn-to-Ask 32B · RealMedConv eval | TR: 3.19 -> 3.15 | No clear effect |
| use CISPO instead of GRPO (ablation) | Learn-to-Ask 7B · RealMedConv eval | TR: 3.27 -> 3.36 | Improved |
| use GSPO instead of GRPO (ablation) | Learn-to-Ask 7B · RealMedConv eval | TR: 3.27 -> 3.16 | Hurt |
| increase beta from 2 to 4 (ablation) | Learn-to-Ask 7B · RealMedConv eval | TR: 3.27 -> 3.15 | Hurt |
| SFT instead of Learn-to-Ask | Qwen2.5-7B-Instruct · RealMedConv eval | WA: 0.50 -> 0.40 | Hurt |
| DPO instead of Learn-to-Ask | Qwen2.5-32B-Instruct · RealMedConv eval | WA: 0.50 -> 0.23 | Hurt |
Closest papers by meaning
- AGILE: A Novel Reinforcement Learning Framework of LLM Agents (NeurIPS 2024 · similarity 0.88)
- Planning without Search: Refining Frontier LLMs with Offline Goal-Conditioned RL (NeurIPS 2025 · similarity 0.88)
- Learning to Clarify: Multi-turn Conversations with Action-Based Contrastive Self-Training (ICLR 2025 · similarity 0.88)
- Thinker: Learning to Think Fast and Slow (NeurIPS 2025 · similarity 0.87)
- Simulation-Free Hierarchical Latent Policy Planning for Proactive Dialogues (AAAI 2025 · similarity 0.87)
- ExpeL: LLM Agents Are Experiential Learners (AAAI 2024 · similarity 0.87)
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.
