AI Paper Atlas

Research map / Large language models / Alignment and preferences

SPA: Achieving Consensus in LLM Alignment via Self-Priority Optimization

Yue Huang, Xiangqi Wang, Xiangliang Zhang

AAAI 2026 · Large language models › Alignment and preferences · approach: Robust and calibrated DPO · 3 citations (OpenAlex)

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 changedWhereMeasuredEffect
add dual-criterion denoising (ablation)SPA (Llama-3.1-8B) · SafeRLHFhelpfulness: 6.99 -> 7.14Improved
add dual-criterion denoising (ablation)SPA (Mistral-7B) · SafeRLHFharmlessness: 9.60 -> 9.76Improved
add uncertainty-weighted loss over standard SimPO (ablation)SPA (Llama-3.1-8B) · SafeRLHFhelpfulness: 6.98 -> 7.14Improved
use SimPO instead of DPO objective (ablation)SPA (Llama-3.1-8B) · SafeRLHFhelpfulness: 5.80 -> 6.98Improved
apply SPALlama-3.1-8B-Instruct · SafeRLHFhelpfulness: 5.23 -> 7.14Improved
apply SPAMistral-7B-Instruct · SafeRLHFharmlessness: 8.83 -> 9.76Improved
apply SPALlama-3.1-8B-Instruct · WildGuardhelpfulness: 6.09 -> 6.22No clear effect
apply SFT on preferred samplesLlama-3.1-8B-Instruct · WildGuardhelpfulness: 6.09 -> 3.20Hurt
apply SPALlama-3.1-8B-Instruct · MTBenchscore: 8.025 -> 8.075 (SafeRLHF), 8.013 (WildGuard)No clear effect
run second SPA iteration reusing same promptsSPA (Llama-3.1-8B, WildGuard) · WildGuardhelpfulness: 6.35 (new prompts) -> 6.49 (same prompts)Improved
use SPA instead of Self-CriticismLlama-3.1-8B-Instruct · SafeRLHFhelpfulness: 7.68 -> 7.14Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy