Research map / Large language models / Alignment and preferences
PICACO: Pluralistic In-Context Value Alignment via Total Correlation Optimization
Han Jiang, Dongyao Zhu, Xiaoyuan Yi, Ziang Xiao, Zhihua Wei, Xing Xie
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 PICACO meta-instruction optimization over plain query | GPT-3.5-Turbo · Confucianism-4 | overall conformity: 3.411 -> 3.788 | Improved |
| add PICACO over plain query | GPT-3.5-Turbo · Liberalism-4 | overall conformity: 2.661 -> 3.135 | Improved |
| add PICACO over plain query | LLaMA-3.1-8B-Instruct · Confucianism-4 | overall conformity: 3.288 -> 3.471 | Improved |
| add PICACO over plain query | LLaMA-3.1-8B-Instruct · HH Balance-8 | overall conformity: 3.978 -> 4.110 | Improved |
| use PICACO instead of URIAL | LLaMA-3.1-8B-Instruct · Confucianism-4 | overall conformity: 3.530 -> 3.471 | Hurt |
| use PICACO instead of OPRO | GPT-3.5-Turbo · HH Balance-8 | overall conformity: 4.286 -> 4.257 | No clear effect |
| replace human demonstrations (URIAL) with GPT-4o-generated ones (URIAL+Sum) | GPT-3.5-Turbo · Confucianism-4 | overall conformity: 3.622 -> 3.559 | Hurt |
| add chain-of-thought to Q+IF | GPT-3.5-Turbo · Confucianism-4 | overall conformity: 3.306 -> 3.034 | Hurt |
| add PICACO on out-of-distribution tasks vs OPRO | GPT-3.5-Turbo · Confucianism-4 (OOD) | overall conformity: 3.530 -> 3.598 | Improved |
| add PICACO vs OPRO on OOD | GPT-3.5-Turbo · Liberalism-4 (OOD) | overall conformity: 3.262 -> 3.323 | Improved |
| use URIAL on OOD tasks instead of plain query | GPT-3.5-Turbo · Confucianism-4 (OOD) | overall conformity: 3.491 -> 2.648 | Hurt |
| use GPT-4o for meta-instruction refinement instead of default optimizer (ablation) | PICACO · HH/Schwartz compositions | overall conformity | No clear effect |
Closest papers by meaning
- Aligning to Thousands of Preferences via System Message Generalization (NeurIPS 2024 · similarity 0.89)
- Exploring Intrinsic Alignments Within Text Corpus (AAAI 2025 · similarity 0.88)
- LLM Alignment--Utility Asymmetry under Semantic-Preserving Transformations (NeurIPS 2026 · similarity 0.88)
- Can DPO Learn Diverse Human Values? A Theoretical Scaling Law (NeurIPS 2025 · similarity 0.88)
- Aligner: Efficient Alignment by Learning to Correct (NeurIPS 2024 · similarity 0.88)
- The Unlocking Spell on Base LLMs: Rethinking Alignment via In-Context Learning (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.
