At the Spring School AI For Impact in March 2026, two of the most consequential thinkers in machine learning sat for what was billed as a debate but functioned more like a forensic disagreement about what intelligence is for (YouTube recording). On one side, Yann LeCun, Meta’s former chief AI scientist, argued that any system that models the world by trying to reconstruct pixels is doomed because most of what happens in a video – leaves trembling, light flickering, individual molecules of air – is intrinsically unpredictable, and forcing a network to predict it wastes capacity and corrupts representations (Meta AI on V-JEPA). On the other side, Eric Xing, president of MBZUAI and longtime CMU professor, argued that abstract latent prediction without a generative validator is essentially meditation in a closed room: elegant, internally consistent, and prone to losing contact with reality (MBZUAI on PAN).
