Large language models are mathematically programmed to create kitsch rather than high art.
AI-generated art often feels technically perfect but emotionally hollow or generic. This study explains the phenomenon by showing how training on statistical averages forces models toward the most common denominator. This kitsch is a direct byproduct of the loss functions used in modern machine learning. While humans might rate this art highly in blind tests, it lacks the novelty and depth of human-led creation. This discovery suggests that the soulless feeling of AI art is a feature of its architecture. To get real art, we may need to stop training models on the average of human history.
LLMs Generate Kitsch
arXiv · 2604.25929
Large Language Models (LLMs) are increasingly used to generate pictures, texts, music, videos, and other works that have traditionally required human creativity. LLM-generated artifacts are often rated better than human-generated works in controlled studies. At the same time, they can come across as generic and hollow. We propose to resolve this tension by arguing that LLMs systematically generate kitsch, and that this is a consequence of the way in which they are trained. We also show empirical