David Patterson explores the biggest hurdles facing hardware designed for running large language models. He highlights why current chips struggle and points to research directions that could unlock faster, more efficient AI.
https://arxiv.org/abs/2601.05047