1 paper · 1 filter
Zhe Yu, Wenpeng Xing, Tiancheng Zhao +3
When retrieved evidence contradicts parametric memory, language models frequently ignore context and default to memorized priors -- a failure that undermines the core purpose of re…