Reproduce “PreCoMem: Predictive Cognitive Memory for Self-Evolving Long-Term Dialogue Agents” 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
Jian Zhong, Zeyu Liu, Pingchuan Cao, Rongduo Han, Shunye Tang, Guohuan Xie, Mingyuan Qin, Ran Ji and 3 more
PreCoMem gives long-running assistants a way to update user memories without treating every new comment as lasting change. It stores beliefs with confidence weights and routes ambiguous cues to provisional hypotheses, while strengthening supported beliefs and softening contradicted ones. With gpt-4.1, it reaches 89.39% accuracy on LongMemEval-S; controlled drift recovery is shown on DiMoBench, not guaranteed for all conversations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.