1 paper · 1 filter
Xuancheng Li, Haitao Li, Yujia Zhou +2
Long-context inputs in large language models (LLMs) often suffer from the "lost in the middle" problem, where critical information becomes diluted or ignored due to excessive lengt…