Research map / Large language models / Alignment and preferences
Closing the Expression Gap in LLM Instructions via Socratic Questioning
Jianwen Sun, Yukang Feng, Yifan Chang, Chuanhao Li, Zizhen Li, Jiaxin Ai, Fanrui Zhang, Yu Dai, Kaipeng Zhang
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 information-gain entropy reward instead of slot-counting reward (ablation) | Nous (OfG) · Diagram specification dialogue (100 test diagrams) | total information gain: 97.11 -> 120.53 | Improved |
| add information-gain entropy reward instead of slot-counting reward (ablation) | Nous (OfG) · Final chart quality, VisPainter | weighted score: 0.63 -> 0.76 | Improved |
| add information-gain entropy reward instead of slot-counting reward (ablation) | Nous (OfG) · Diagram specification dialogue (100 test diagrams) | dialogue turns: 13.6 -> 20.3 | Hurt |
| use offline GRPO instead of SFT | Nous · Diagram specification dialogue (100 test diagrams) | total information gain: 94.3 -> 120.5 | Improved |
| use offline GRPO instead of online GRPO | Nous · Diagram specification dialogue (100 test diagrams) | training resource hours: 93 -> 28.5 | Improved |
| use offline GRPO instead of online GRPO | Nous · Diagram specification dialogue (100 test diagrams) | total information gain: 115.8 -> 120.5 | Improved |
| use offline GRPO instead of DPO | Nous · Diagram specification dialogue (100 test diagrams) | total information gain: 111.3 -> 120.5 | Improved |
| train on Vague Oracle dialogues instead of Template Oracle (ablation) | Nous (OfG) · Diagram specification dialogue (100 test diagrams) | total information gain: 120.5 -> 117.1 | No clear effect |
| train on Noisy Oracle dialogues instead of Template Oracle (ablation) | Nous (OfG) · Diagram specification dialogue (100 test diagrams) | total information gain: 120.5 -> 115.8 | No clear effect |
| start from a partial draft instead of from scratch | Nous (OfG) · Human users | dialogue turns: 17.3 -> 7.6 | Improved |
| human users instead of simulated oracle | Nous (OfG) · Human users, zero-start | dialogue turns: 20.3 -> 17.3 | Improved |
| apply Nous (OfG) in novel writing instead of GPT-zero | Nous (OfG) · Novel writing generalization | total information gain: 55.2 -> 65.4 | Improved |
Closest papers by meaning
- Theory of Mind and Persuasion Beyond Conversation: Assessing the Capacity of LLMs to Induce Belief States via Planning and Action (NeurIPS 2026 · similarity 0.88)
- When Models Know More Than They Can Explain: Quantifying Knowledge Transfer in Human-AI Collaboration (NeurIPS 2025 · similarity 0.88)
- Do LLMs ``know'' internally when they follow instructions? (ICLR 2025 · similarity 0.88)
- Generalizing Verifiable Instruction Following (NeurIPS 2025 · similarity 0.87)
- Learning “Partner-Aware” Collaborators in Multi-Party Collaboration (NeurIPS 2025 · similarity 0.87)
- MetaMind: Modeling Human Social Thoughts with Metacognitive Multi-Agent Systems (NeurIPS 2025 · 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.
