1 paper · 1 filter
Shiwen Qin, Cameron Barker, Henry Gouk +4
Long-context language models support ever longer context windows, and evaluations typically treat context length as the main indicator of difficulty. However, context length alone…