LLM Outputs Are Fictions

The development of large language model (LLM) tools presents both benefits and challenges. One issue is the expectation of reliable information from users, particularly students. The tool’s use conditions and expectations influence its effectiveness. The concept of fictional truth negotiates relations that hold between the real world and the world of fiction and is therefore useful for understanding LLM outputs that rely on and contain facts even as they also contain false or invented information.