if we link piece of code to data vending machine…. Then data in, model weight out. Find another way to do back propagation or gather these weight together….
Then data vending machine can distribute AI training process👽👽👽
coming soon to a nostr:npub1w0rthyjyp2f5gful0gm2500pwyxfrx93a85289xdz0sd6hyef33sh2cu4x near you
pay AIs to give you a custom-made feed based on... whatever (your interests, what you've recently written about on nostr, what your network has recently written about... whatever! long-tail algorithms ftw!)
cc nostr:npub149p5act9a5qm9p47elp8w8h3wpwn2d7s2xecw2ygnrxqp4wgsklq9g722q
Discussion
We will need a unified annotation format at some point so dvms understand the labels.
A proper framework to think about distributed processing for AI, is (Bayesian) predictive coding --- no need for backpropagation just message passing. Each dvm is trying to predict the output of oder dvms, some dvms are linked to data and decisions, they collectively form an information proccesing network, much like information processing in the brain. In ML this is also known as variational autoencoders but with distributed components.