Vision Language Models are Biased

An Vo, Khai-Nguyen Nguyen, Mohammad Reza Taesiri +3

Large language models (LLMs) memorize a vast amount of prior knowledge from the Internet that helps them on downstream tasks but also may notoriously sway their outputs towards wro…