一作ACL 2026Findings· Findings of ACL
Better and Worse with Scale: How Contextual Entrainment Diverges with Model Size
Dikshant Kukreja · Kshitij Sah · Gautam Gupta · Avinash Anand · Rajiv Ratn Shah · Zhengkui Wang · Aik Beng Ng · Erik Cambria
摘要
Larger language models become simultaneously better and worse at handling contextual information—better at ignoring false claims, worse at ignoring irrelevant tokens. We formalize this apparent paradox through the first scaling laws for contextual entrainment, the tendency of models to favor toke...