Reproduce “Piccolo: Co-Evolving Decision Models and Data through Recursive Self-Improvement” 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
Zhenqing Ling, Qirui Jiao, Daoyuan Chen, Cheng Chen, Xinji Mai, Haowen Gao, Zhuo Zhang, Xuchen Pan and 9 more
Piccolo develops decision models that return probabilities for supplied choices, rather than generating answer text. It reads logits for single-token candidate labels and uses semantic slices to guide human-steered data and training experiments. In the 17 November 2026 Decision Index Vision snapshot, Piccolo-27B scored 75.59, ahead of JDI-Decision V1 at 75.05, a benchmark result, not evidence of superiority or measured speedup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.