Gazing into the Loop: An Artistic Take on Machine Subjectivity

Using looping videos generated by DeepMind Veo2 and a minimalist token visualization script, this project reexamines Human-in-the-Loop (HITL) interactions from the machine’s viewpoint. Employing intuitive visual metaphors, it reveals the essence of large language models as probability-driven predictors, do not truly understand content. This project experiments with treating the algorithm as a camera and representing tokens as living being. Ultimately, it prompts reflection on who truly controls the rhythm in human-machine interactions, and what "intelligence" really means when we face an AI that appears realistic but lacks authentic understanding.