This week’s Paper Club is all about alternative compute. We cover the case for abandoning backprop and the gradient-free optimizer being built to replace it, how the propagation of light can be programmed to run a diffusion model, where neuromorphic computing really stands in 2026, and how human brain cells on a chip learned to play Doom.
Sign up for the next Paper Club, where we’ll discuss AI Safety: https://events.ycombinator.com/yc-paperclub-oct7
Chapters:
00:10 – Why we need a new kind of compute
08:08 – Training without gradients
15:06 – Computing with light
39:00 – Computing with analog circuits
1:04:24 – Computing with living neurons
1:19:32 – Can any of this scale?
Apply to Y Combinator: https://www.ycombinator.com/apply
Work at a startup: https://www.ycombinator.com/jobs
source
