Abstract: Large language models (LLMs) are transforming writing and planning practices across the breadth of knowledge work, where misconceptions about the technology abound. We contend that these misconceptions are rooted in the dispersal of disciplinary knowledge—that multiple disciplines are needed to understand LLMs. We bring together disciplinary knowledge to address four constructs commonly applied to LLMs: writing, learning, hallucination, and thinking. Misconceptions about LLM technology are embodied in and transmitted through conceptual metaphors and analogies. We thus provide a basic description of LLM technology with a careful accounting for metaphor. This description and the discussion of the four constructs leads to corrected statements for understanding LLMs. For instance, the assertion that “LLMs think” is replaced with: “LLMs produce outputs that resemble the products of thinking.” Finally, we consider common and potential metaphors for more accurately conceptualizing LLMs, and favor LLMs as answer simulators, remixers, fuzzy thesauruses, and plagiarism traps. The statements and metaphors provided here are a starting point for well-grounded usage patterns with LLMs for professionals, researchers, and students alike.