Generative models make new content that follows the statistical patterns of their training data, and large language models predict the next word or token. The crew runs Tiny Text with a die and finds that it writes smooth sentences that are not in its data: some contradict the data and some are unsupported. Confabulation, when generative AI confidently presents false content, is a natural result of how these models are designed. It matters most for long, open-ended answers and expert topics, and it can come with made-up reasons or citations. The crew rates the harm of SH-1 risk R1, confabulation.