Reproduce “WeightLoop: Weight-Space Programming with Experimental Feedback” 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
Mengran Li, Bo Li, Weijing Zhao, Chengyang Zhang, Boyu Zhang, Jinfeng Xu, Zelin Zang, Victor C. M. Leung and 1 more
WeightLoop lets assay results update a predictor ranking later candidates. A fixed editor, constructed once by an LLM, tests bounded changes to output weights against measured outcomes; it preserves the representation and starts each update from original parameters. In offline replay, targeted edits improved ranking utility over the unchanged model in seven of eight settings, without establishing biological efficacy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.