Reproduce “Evaluating LLM-Generated World Models for Neuro-Symbolic Reasoning” 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
Souhaila Djaffal, Amna Jamal, Gagana Madhugiri Devaraja, Anidipta Pal, Jobst Heitzig
The authors present an evaluation framework for language-generated world models of partially observable stochastic games. It checks whether models execute, match reference structure, and include scenario-relevant features. In the game track, 134 of 135 first-pass models pass the mechanical audit, yet valid models average 0.366 structural F1 and only 28 exactly match references; comparisons cover three transitions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.