1 paper
Yuwei Cheng, Weiyi Tian, Haifeng Xu
Fine-tuning is often believed to reduce uncertainty and diversity in large language models, but existing analyses overlook output length, a key confounder, and therefore fail to ca…