Reproduce “HCMD-zero: Learning Value Aligned Mechanisms from Data” 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
Jan Balaguer, Raphael Koster, Ari Weinstein, Lucy Campbell-Gillingham, Christopher Summerfield, Matthew Botvinick, Andrea Tacchetti
HCMD-zero learns an allocation rule for a four-person shared-fund game by observing contributions and votes, modeling participant behavior, and improving the rule through simulated contests among copies of itself. It is trained without the rival used in evaluation. In human tests, it received 54.3% of votes against Liberal Egalitarian, a numerical edge with p < 0.06 in this task.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.