1 paper · 1 filter
Minghao Li, Bangyan Li, Zifan Wang +5
Long-horizon tasks require LLM agents to continually draw on information from earlier interactions. However, retaining the full history increases context costs, while compressing i…