Reproduce “Diagnosing the Explicit-to-Applied Theory of Mind Gap in LLMs” 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

Prajvi Saxena, Sabine Janzen, Wolfgang Maass

The authors test whether language models use recognized mental states to predict behavior and judge actions on benchmarks. They compare mental-state reminders, matched in-context examples, and BSTTT, which adapts LoRA parameters using answered support questions. On SimpleToM, Llama-3.1-8B behavior accuracy rises from 28.9% frozen to 69.7% with a reminder and BSTTT, but gains do not establish applied competence.

Summary by alphaXiv.

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