Research map / Large language models / Alignment and preferences
Coactive Critiquing: Elicitation of Preferences and Features
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Dynamically acquire critique features using NeedCritique | Coactive Critiquing · Synthetic preference task | median utility loss | Helped |
| Use Coactive Critiquing instead of CL^100 | Synthetic preference task · Synthetic preference task | iterations to zero median utility loss: 11 iterations; 41 iterations | Hurt |
| Dynamically acquire critique features using NeedCritique | Coactive Critiquing · Trip planning | utility loss | Helped |
| Use Coactive Critiquing instead of CL^60 | Trip planning · Trip planning | utility loss | No clear effect |
| Use Coactive Critiquing instead of CL^40 | Trip planning · Trip planning | utility loss | Helped |
| Replace NeedCritique with random critique querying at θ=1 (ablation) | Coactive Critiquing · Synthetic preference task | median utility loss | Mixed |
| Replace NeedCritique with random critique querying at θ=0.75 (ablation) | Coactive Critiquing · Synthetic preference task | median utility loss | Mixed |
| Replace NeedCritique with random critique querying at θ=0.5 or θ=0.25 (ablation) | Coactive Critiquing · Synthetic preference task | median utility loss | Hurt |
| Start with 2 features and acquire additional features through critiques rather than having a fixed feature subset | Coactive Critiquing · Synthetic preference task | median utility loss: 41 iterations; 18 iterations | Mixed |
| Acquire features dynamically rather than start with a fixed 20% feature subset | Coactive Critiquing · Trip planning | utility loss | Helped |
Closest papers by meaning
- Coactive Learning for Large Language Models using Implicit User Feedback (ICML 2024 · similarity 0.86)
- Preference learning made easy: Everything should be understood through win rate (ICML 2025 · similarity 0.84)
- Capturing Individual Human Preferences with Reward Features (NeurIPS 2025 · similarity 0.84)
- Preference Learning with Response Time: Robust Losses and Guarantees (NeurIPS 2025 · similarity 0.83)
- Eliciting Human Preferences with Language Models (ICLR 2025 · similarity 0.83)
- Discovering Preference Optimization Algorithms with and for Large Language Models (NeurIPS 2024 · similarity 0.83)
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.