News China makes AI breakthrough, reportedly trains generative AI model across multiple data centers and GPU architectures

[MikeClark512]

If their technique allows the training to be broken up into latency-insensitive chunks, like Folding@home, then it should probably be considered a breakthrough. I don't know if that's possible, just throwing it out there.

It's impossible to break up a fully interdependent+interconnected system into substructures can be operated on in isolation and then recombined in a way that gives the exact same results as manipulating the entire superstructure as a whole. You can use heuristics to try to break things up in a way that approximates it, and with the fuzzy systems that biomimicry tends to work with, this approach can yield useful results. But this is not an on/off switch like: "it couldn't be done at all and now we can do it." It has long been a game of inches, and the article fails to mention how far China moved the ball (benchmarks), let alone what game they're playing (e.g., model definition, loss function, training data, evaluation criteria.)