Video Recordings

When studying the outputs of large language models, researchers can focus on various aspects — underlying generative processes, emergent phenomena, the influence of prompting strategies, and so on. One key choice is whether to direct attention toward the upper limits of what these systems can produce, or toward their most typical, default outputs. In this talk, I focus on the latter: AI slop. I argue that studying the features of — and attitudes toward — AI slop matters precisely because it is this kind of output that floods our online spaces, from unusable AI-generated knitting patterns on Pinterest to the growing wave of AI-written papers that overwhelm scientific journals. I will present two lines of research: one on AI poetry generated by minimal prompts, and one on AI-generated images and the extent to which they reflect human-specific visual asymmetries.