Reproduce “When Do Causal World Models Help Modular 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

Xinyuan Song, Zekun Cai

CausalCompose models module interactions by validating action–response links across repeated executions, then composing local world models to predict downstream effects of deliberate actions. Hidden factors can make logged action-outcome patterns misleading. On τ-bench, it reports 54.86% Retail task success, versus 33.33% for Global Sequence WM; without confidence intervals, this is a diagnostic comparison, not a precise ranking.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.