Reproduce “GVGAI-LLM: Evaluating Large Language Model Agents with Infinite Games” 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
Yuchen Li, Cong Lin, Muhammad Umair Nasir, Philip Bontrager, Jialin Liu, Julian Togelius
GVGAI-LLM tests whether language models can play rule-driven 2D games by translating formal game rules and changing symbolic states into text, then repeatedly asking a model to choose an action that the game engine executes. In a zero-shot test across 118 games, GPT-4o-mini won 10.27% overall; results varied by game, and every tested LLM won 0% on Boulderdash.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.