Back to feed
arXiv cs.CL
arXiv cs.CL
7/31/2026
BridgeAlign: Bridging Preference Alignment for Humanities and Social Sciences

BridgeAlign: Bridging Preference Alignment for Humanities and Social Sciences

Short summary

BridgeAlign is a preference-alignment pipeline for humanities and social sciences (HSS) with three phases: seed curation from web corpora, preference data synthesis via persona-based instruction inversion, and preference optimization using quality rubrics with controlled degradation. Aligning over 210K synthetic samples, Qwen3-8B achieved the best average across 17 benchmarks against 11 baselines, leading on both human-preference and knowledge-based capabilities without trade-offs.

  • •Three-phase pipeline: seed curation, preference synthesis, rubric-grounded optimization
  • •210K synthetic preference samples; Qwen3-8B beats 11 baselines across 17 benchmarks
  • •No trade-off between human-preference and knowledge-based capabilities

Generated with AI, which can make mistakes.

#research-breakthrough#open-source
Read full article at arXiv cs.CL

Is this a good recommendation for you?

Comments

Failed to load comments. Please try again.

Explore more

HSS-Synth: Humanities and Social Sciences Data Synthesis for LLMs
arXiv cs.CL
Human Preference aligned Tabular Similarity
arXiv cs.LG
RIMS: Preference Optimization via Smoothed Multi-pair Aggregation for Small-Scale LLM Retrieval-Augmented Generation
arXiv cs.CL
Beyond Shapley: An Influence-Based Data Auditing Pipeline for LLM Alignment and Evaluation
arXiv cs.LG