The generation of fictional or factually incorrect information by a large language model is known as fabrication or hallucination. This occurs because the model is designed to predict the next most likely sequence of words to form a coherent response, rather than retrieving facts from a knowledge base. As a result, it can confidently present information that is not grounded in the source data or reality. This is a recognized limitation of current generative AI, highlighting the need for human verification of AI-generated content.

A. A deepfake is a synthetically altered video or audio recording, not a text-based response from an AI model.

C. Overreliance is a human behavior of trusting AI output without critical evaluation, not the AI's act of generating false information.

D. Prompt injection is a security exploit where a user tricks an AI into bypassing its safety constraints, which is not what is described.

E. Bias refers to skewed or prejudiced outputs based on the training data, whereas fabrication is the creation of entirely new, untrue information.

1. Microsoft. (2023). Microsoft Responsible AI Standard, v2. Section 2.2, "Reliability and Safety," discusses the need to mitigate "the risk of the system generating content that is not factually accurate (often referred to as 'hallucinations' or 'fabrications')."

2. Stanford University Human-Centered Artificial Intelligence (HAI). (2023). On the Opportunities and Risks of Foundation Models. Section 4.2, "Misinformation," discusses how models can "generate plausible-sounding but false or nonsensical text," a phenomenon referred to as hallucination or fabrication.