Reproduce “Chain on Demand: Deferred Evidence Organization for Long-Term Conversational 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
Di Lu, Zechen Li, Hao Xue
The authors introduce Chain on Demand (CoD), which organizes conversational evidence after a question is known without rewriting persistent memory. It retrieves candidate turns, selects anchors for relevance, diversity, and query-term coverage, then restores nearby turns from the same session. On 1,986 LoCoMo questions, CoD raises accuracy from 74.5% to 76.3% versus relevance-only allocation, within the tested pipeline.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.