1 paper
Mingyang Song, Mao Zheng, Zhiyuan Ma +1
Selection bias in large language models (LLMs) can persist despite instructions to ignore presentation order or prompts that elicit reasoning. Recent efforts to reproduce Jev-style…