一作ACL 2026Poster· Natural Language Processing
Feeling Right vs. Being Right: How AI Sycophancy Affects Value-Laden Deliberation
Jeongwoo Ryu · Soomin Kim · Jinsu Eun · Kyusik Kim · Changhoon Oh · Bongwon Suh
摘要
As people increasingly turn to AI for personal deliberation beyond task-oriented assistance, concerns about sycophancy in these value-laden contexts have grown. Unlike human flattery, which is intentional and self-interested, AI sycophancy emerges as a byproduct of RLHF’s reward structure for use...