Researchers have identified six misconceptions about large language models (LLMs). These misconceptions include thinking that LLMs simply predict the next token in a sequence, or that they are 'stochastic parrots' regurgitating training data. A new model and taxonomy aim to clarify these misunderstandings by distinguishing between pretraining and deployed systems, learned distributions and particular samples, and other key aspects of LLMs. The framework is applied to governance case studies, highlighting how policy language can perpetuate errors in understanding LLM capabilities. --- Why it matters: Understanding the limitations and capabilities of large language models is crucial for developing effective AI policies and governance frameworks. This research provides a diagnostic toolkit for identifying and correcting misconceptions about LLMs, which can inform decision-making in fields such as education, science, and governance. Source: https://arxiv.org/abs/2608.20421