Researchers have developed a graph-based framework to compare how humans and large language models (LLMs) reason when fact-checking scientific claims. The system, called the typed reasoning graph, represents explanations as graphs that link false claims to relevant study context, findings, and fallacies. This allows for direct comparison of human and LLM reasoning paths at a detailed level. An evaluation using 84 false claims shows that different LLMs have distinct performance characteristics, with some excelling in verdict accuracy while others perform well in terms of valid reasoning. --- Why it matters: This work matters to AI researchers because it provides a new framework for evaluating the reasoning abilities of large language models and comparing them to human experts. This can help improve the development of more accurate and transparent fact-checking systems. Source: https://arxiv.org/abs/2608.23047