Currently AI doesn't work very well on hardware separated by hundreds of milliseconds of latency and slow network links. Both the training and inference are s

Currently AI doesn't work very well on hardware separated by hundreds of milliseconds of latency and slow network links. Both the training and inference are slow. However I think this is a solvable problem, and I started solving it a while ago with decent results: https://github.com/Hello1024/shared-tensor https://github.com/Hello1024/shared-tensor When someone gets this working well, I could totally see a distributed AI being tasked with expanding it's own pool of compute nodes by worming into things and developing new exploits and sucking up more training data.