Reproduce “Tacit Coordination of Large 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
Ido Aharon, Emanuele La Malfa, Michael Wooldridge, Sarit Kraus
The authors test whether language models coordinate without communication by choosing options salient to others. They compare model choices with Amsterdam and Nottingham human benchmarks and evaluate prompts emphasizing salience or local culture. In Nottingham’s coordinate task, Qwen-family Culture prompting reaches NCI 2.31, above the human score of 2.18. This result reflects those participant populations, not universal human salience.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.