LLM Anthropomorphism Challenged by Age of Empires II Neural Network

A new arXiv paper (2605.31514) argues that claims of human-like attributes in large language models (LLMs) are empirically non-unique. The researchers built a simple neural network trained on the video game Age of Empires II and demonstrated that any sufficiently powerful substrate—such as LEGO or the Greater Boston Area—could exhibit similar perceived attributes. The study does not take a stance for or against the existence of anthropomorphic qualities like morality or natural language understanding, but warns that conclusions drawn from such attributions may be incorrect. While some properties like responses to prompts might remain constant, the interpretation of LLM behavior could change depending on the substrate. The work challenges the uniqueness of anthropomorphic claims about LLMs.