作者位次 #3/7ACL 2026Poster· Natural Language Processing
Evaluating Language Model Pluralism through In-the-wild Crowd Discussions
Gagan Mundada · Rohan Surana · Nandhini Swaminathan · Bodhisattwa Prasad Majumder · Junda Wu · Julian J. McAuley · Zhouhang Xie
摘要
When answering subjective questions, an ideal LLM should surface diverse plausible perspectives rather than favoring a single viewpoint, a characteristic known as pluralism. Recent studies show that modern LLMs optimized through preference alignment systematically favor certain positions on subje...