Reproduce “Reasoning Promotes Robustness in Theory of Mind Tasks” 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
Ian B. de Haan, Peter van der Putten, Max van Duijn
An investigation into advanced Large Language Models, particularly those trained with Reinforcement Learning with Verifiable Rewards (RLVR), reveals they achieve improved performance and significantly increased robustness on Theory of Mind tasks. This robustness stems from more consistent inference processes that mitigate susceptibility to subtle prompt variations previously observed.
Key insights
- Improved LLM performance in ToM tasks is primarily attributed to increased *robustness* in finding correct solutions rather than the acquisition of fundamentally new cognitive capacities.
- Explicit reasoning processes, enhanced by RLVR training, establish more stable and reliable inference paths, reducing models' susceptibility to distracting or misleading prompt structures.
- Models exhibit emergent strategic behaviors, such as 'filtering facts' and 'perspective-taking,' in their reasoning traces, suggesting effective inference-time scaling techniques have been acquired.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.