If LLMs Have Human-Like Attributes, Then So Does Age of Empires II

The article presents a study that challenges the attribution of anthropomorphic attributes to large language models (LLMs) by training a simple neural network on the game Age of Empires II. It argues that the observed behaviors of LLMs may not be unique and could be replicated by any sufficiently powerful substrate, suggesting that conclusions about LLMs' capabilities must be grounded in explicit measurement criteria rather than assumptions. This perspective is crucial for practitioners as it calls for a reevaluation of how LLMs are assessed and understood in terms of their functional properties, potentially influencing experimental design and interpretation in AI research.