Reproduce “Don't Just Trust Your Memory: Evidence-Linked Memory Reduces Memory-Induced Degradation in LLM 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
Yukyeong Lee, Yumin Choi, Jinheon Baek, Sung Ju Hwang
WikiTrace links concise memory insights to interpretations and unchanged raw trajectories from their source runs, allowing an agent to inspect evidence when advice is incomplete, conflicting, or from failure. In three-trial tests across eight benchmark suites, it averages 90.2 versus 87.0 for Insights Only and 4.5% versus 6.5% Regression; source access does not verify insights.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.