Reproduce “Knowing What is Missing: Efficient Conversational Memory via Explicit Evidence-Gap Tracking” 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

Xingbo Du, Loka Li, Duzhen Zhang, Leonard Song, Le Song

MemR³ tracks concise evidence and unanswered gaps during long-conversation question answering, routing each step to retrieval, reflection, or answering and masking retrieved snippets after incorporating them. On LongMemEval-s, GPT-4o-mini received a 71.67% three-run mean judge score, versus 58.53% for Full-Context, using 5.9% of its total query-time tokens; it did not outperform Full-Context on every backend.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.