Reproduce “Memory Tax: Dissecting the Net Value of Agent Memory” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xuan He, Sining Zhoubian, Ruozhen Yang, Yuhao Wen, Zhineng Wang, Yuyang Hu, Ke Yang, Yuyu Zhang and 3 more
The authors evaluate whether processing long conversation histories into memory improves on retrieving directly from raw history. They define memory benefit as queries memory answers correctly when one-round retrieval-augmented generation (RAG) fails, and memory tax as the reverse. On LongMemEval-s, HippoRAG2 reaches 78.2% judge accuracy versus 81.4% for two-round RAG using Qwen3.6-27B; results vary across tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.