Research map / Large language models / Alignment and preferences
SPA: Achieving Consensus in LLM Alignment via Self-Priority Optimization
Yue Huang, Xiangqi Wang, Xiangliang 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 dual-criterion denoising (ablation) | SPA (Llama-3.1-8B) · SafeRLHF | helpfulness: 6.99 -> 7.14 | Improved |
| add dual-criterion denoising (ablation) | SPA (Mistral-7B) · SafeRLHF | harmlessness: 9.60 -> 9.76 | Improved |
| add uncertainty-weighted loss over standard SimPO (ablation) | SPA (Llama-3.1-8B) · SafeRLHF | helpfulness: 6.98 -> 7.14 | Improved |
| use SimPO instead of DPO objective (ablation) | SPA (Llama-3.1-8B) · SafeRLHF | helpfulness: 5.80 -> 6.98 | Improved |
| apply SPA | Llama-3.1-8B-Instruct · SafeRLHF | helpfulness: 5.23 -> 7.14 | Improved |
| apply SPA | Mistral-7B-Instruct · SafeRLHF | harmlessness: 8.83 -> 9.76 | Improved |
| apply SPA | Llama-3.1-8B-Instruct · WildGuard | helpfulness: 6.09 -> 6.22 | No clear effect |
| apply SFT on preferred samples | Llama-3.1-8B-Instruct · WildGuard | helpfulness: 6.09 -> 3.20 | Hurt |
| apply SPA | Llama-3.1-8B-Instruct · MTBench | score: 8.025 -> 8.075 (SafeRLHF), 8.013 (WildGuard) | No clear effect |
| run second SPA iteration reusing same prompts | SPA (Llama-3.1-8B, WildGuard) · WildGuard | helpfulness: 6.35 (new prompts) -> 6.49 (same prompts) | Improved |
| use SPA instead of Self-Criticism | Llama-3.1-8B-Instruct · SafeRLHF | helpfulness: 7.68 -> 7.14 | Hurt |
Closest papers by meaning
- LLM Alignment as Retriever Optimization: An Information Retrieval Perspective (ICML 2025 · similarity 0.88)
- Style Outweighs Substance: Failure Modes of LLM Judges in Alignment Benchmarking (ICLR 2025 · similarity 0.88)
- On Evaluating LLM Alignment by Evaluating LLMs as Judges (NeurIPS 2025 · similarity 0.88)
- TODO: Enhancing LLM Alignment with Ternary Preferences (ICLR 2025 · similarity 0.88)
- Varying Shades of Wrong: Aligning LLMs with Wrong Answers Only (ICLR 2025 · similarity 0.88)
- Keeping LLMs Aligned After Fine-tuning: The Crucial Role of Prompt Templates (NeurIPS 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.
