Reproduce “More Memory, Worse Agents: Error Reproduction and Anti-Persistence 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
Ruhan Wang, Kishan Panaganti, Dongruo Zhou
Anti-Persistence lets agents retain task records while generating task-specific guidance for one-time use and discarding it. An adaptive selector chooses direct records or synthesized behavioral patterns, strategies, or corrective advice. After ten memory-update rounds, held-out success averaged 59.9% across nine settings, versus 53.5% for trajectory memory. It led in seven settings, but not every setting or against joint memory.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.