The paper in question for the nerds: https://www.nature.com/articles/s41586-023-06558-8

Right now it seems they depend on direct optical input, so it’d be harder to adapt it to other deep-learning techniques. It might be useful down the line for stand-alone computer vision parts for micro-controllers. If they manage to develop this photonic computation tech to a commercial and more general-purpose scale, it could seriously revolutionise the GPU industry too, specially on the energy efficiency front. Really impressive stuff.