In the video titled “The moment we stopped understanding AI [AlexNet]”, Matthew Berman delves into the significance of AlexNet, a pioneering model in the field of artificial intelligence that fundamentally changed how machines interpret images. The video begins by introducing the concept of an Activation Atlas, which provides insights into the high-dimensional embedding spaces that modern AI models utilize to understand and categorize the world. Berman explains how AlexNet, published in 2012, marked a turning point in computer vision by demonstrating that deep learning could achieve remarkable results when scaled appropriately. The protagonist discusses the architecture of AlexNet, highlighting its use of convolutional layers to process images and how these layers learn to detect various visual patterns, such as edges and colors. Through a series of visual examples, Berman illustrates how AlexNet’s layers work together to recognize complex concepts, ultimately mapping input images to output probabilities for different classes in the ImageNet dataset. The video further explores the implications of this model, emphasizing the shift towards high performance in AI while moving away from explainability. Berman also touches on the advancements in AI since AlexNet, including the development of larger models like ChatGPT, and the challenges that arise from their complexity. The video concludes with a reflection on the future of AI, pondering whether the next breakthrough will come from scaling existing models or reviving older methodologies. This engaging presentation not only educates viewers about the history and mechanics of AlexNet but also invites them to consider the broader implications of AI advancements in society.