Retrieval over ACE Playbooks: What Works, What Breaks, and Why
By Shyam Sai Bethina* (Stanford), Fenglu Hong* (SambaNova), and Qizheng Zhang (Stanford)
✍️ TLDR: ACE’s self-generated playbooks can grow to 174k tokens, making inference expensive. We explore three retrieval strategies — embedding-based, LLM-based, and Recursive Language Models (RLM) — to compress playbooks into compact sub-playbooks at inference time. Simple retrieval methods (embedding and LLM ranking) preserve 59–65% of the full-adaptation accuracy gain...
[Read More]
