Reproduce “Cross-Centaur: Modeling Individual Decision-Making Across Psychological Tasks with Language Models” 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
Hikaru Asano, Hidezo Suganuma, Ryutaro Mori, Yukino Baba
Cross-Centaur tests whether a model can use one person’s choices in a gamble to predict that person’s choices about sooner or later rewards. It receives the source-task description and behavior, plus the target-task description, and predicts target actions. For held-out participants and different-category sources, Tradeoff-target action-token NLL is 0.720, versus Centaur’s 0.875; evidence is limited to controlled psychological tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.