Research map / Large language models / Agents, code and math
ConInstruct: Evaluating Large Language Models on Conflict Detection and Resolution in Instructions
Xingwei He, Qianru Zhang, Pengfei Chen, Guanhua Chen, Linlin Yu, Yuan Yuan, Siu-Ming Yiu
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 the guiding prompt for Behavior 2 | GPT-4o · ConInstruct, one conflict | % unacknowledged (B1): 97 -> 4 | Improved |
| add the guiding prompt for Behavior 2 | GPT-4o · ConInstruct, conflict-free instructions | % unnecessary clarification (B6): 0 -> 40 | Hurt |
| add the guiding prompt for Behavior 2 | GPT-4o-mini · ConInstruct, conflict-free instructions | % unnecessary clarification (B6): 0 -> 53 | Hurt |
| add the guiding prompt for Behavior 2 | Claude-3.5-Sonnet · ConInstruct, one conflict | % unacknowledged (B1): 78 -> 18 | Improved |
| add the guiding prompt for Behavior 2 | Claude-3.5-Haiku · ConInstruct, one conflict | % unacknowledged (B1): 93 -> 16 | Improved |
| increase number of conflicts from 1-2 to 5-6 | Claude-3.5-Sonnet · ConInstruct conflict resolution | % acknowledging conflicts (B2+B3): 32.0% -> 64.0% | Improved |
| inter-constraint instead of intra-constraint conflicts | Claude-3.5-Sonnet · ConInstruct conflict detection | F1: 92.7 -> 74.6 | Hurt |
| increase model size from 7B to 14B | Qwen2.5-Instruct · ConInstruct conflict detection | average F1: 57.9 -> 75.8 | Improved |
| increase model size from 14B to 32B | Qwen2.5-Instruct · ConInstruct conflict detection | average F1: 75.8 -> 74.1 | No clear effect |
| add a newly introduced conflicting constraint | LLMs · ConInstruct (CC, KK, PP conflicts) | original-constraint satisfaction rate | Hurt |
| place the new conflicting constraint before the original constraint | LLMs · ConInstruct (CC, KK, PP conflicts) | original-constraint satisfaction rate | Improved |
Closest papers by meaning
- Evaluating Large Language Models at Evaluating Instruction Following (ICLR 2024 · similarity 0.91)
- Benchmarking Complex Instruction-Following with Multiple Constraints Composition (NeurIPS 2024 · similarity 0.90)
- Can Large Language Models Understand Real-World Complex Instructions? (AAAI 2024 · similarity 0.90)
- DINGO: Towards Diverse and Fine-Grained Instruction-Following Evaluation (AAAI 2024 · similarity 0.90)
- Do LLMs estimate uncertainty well in instruction-following? (ICLR 2025 · similarity 0.89)
- Self-play with Execution Feedback: Improving Instruction-following Capabilities of Large Language Models (ICLR 2025 · similarity 0.89)
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.
