Debiasing LLMs by Fine-tuning

Zhenyu Gao, Wenxi Jiang, Yutong Yan

Prior research shows that large language models (LLMs) exhibit systematic extrapolation bias when forming predictions from both experimental and real-world data, and that prompt-ba…