Yann LeCun

Giving a trained neural PDE surrogate denser sensor measurements at test time can actually degrade its predictions, exposing widespread distribution collapse under spatial observation shifts.

Single-pass neural operators can actually beat iterative diffusion solvers on sparse PDE recovery, provided latent-space prediction is structurally coupled with physical field decoding.

Forget scaling compute – the future of AI hinges on a 1000x leap in energy efficiency via tight AI+Hardware co-design over the next decade.

Pre-normalization in Transformers is the culprit behind the mysterious link between massive activation outliers and attention sinks, but decoupling them reveals their distinct functions: global parameterization vs. local attention modulation.

Vision models are far more data-hungry than language models, but Mixture-of-Experts can harmonize this asymmetry for truly unified multimodal models.

LLMs can achieve the same accuracy with 16x less data by constraining their hidden-state trajectories to follow geodesics on a semantic manifold.