YC panel explores alternatives to GPUs for AI compute

A Y Combinator Paper Club video discusses research into computing approaches that could eventually supplement or replace GPUs for AI workloads. Topics include training methods that skip traditional backpropagation, using light-based systems to run diffusion models, the current state of neuromorphic (brain-inspired) computing chips, and an experiment where cultured human brain cells on a chip learned to play a video game.

The panel examines whether any of these alternative compute approaches could realistically scale to compete with today's GPU-dominated AI infrastructure, weighing their early-stage promise against major technical hurdles.