ABBEL: Training LLMs to Update Beliefs for Efficient Long-term Interaction
ABBEL is a framework for training language models to update beliefs in the form of textual states, enabling efficient context compression during long-term interactions. The method uses belief grading through observation reconstruction, reducing the gap with full-context models in collaborative programming and other tasks.
Cursor
Alibaba/Qwen
BAIR (Berkeley AI)27.07 · 07:02
