In the last decade, convolutional neural networks (CNNs) have been the go-to architecture in computer vision, owing to their powerful capability in learning representations from images/videos.

The rapid ascent of large language models (LLMs)—and their growing role in everyday life—masks a fundamental problem: ...