haha I mean not just china, but it's not like the worldwide AI pipeline is rife with innovation these days. Do you watch Yannic Kilcher's paper reviews?
I keep waiting for a model that will do something I want (image -> language with a transformer model that doesn't have a fixed token length during training). I have been waiting for about 3 years, haven't seen it yet.
You can't measure innovation by bulk flow of humans through a system. If anything if you force too many humans through a pipeline that needs innovation you risk kneecapping progress because the separation of wheat and chaff becomes quadratically harder, increasingly gamed, and the people who actually have talent and creativity inevitably burn out and fuck off to do something else.
I keep waiting for a model that will do something I want (image -> language with a transformer model that doesn't have a fixed token length during training). I have been waiting for about 3 years, haven't seen it yet.
You can't measure innovation by bulk flow of humans through a system. If anything if you force too many humans through a pipeline that needs innovation you risk kneecapping progress because the separation of wheat and chaff becomes quadratically harder, increasingly gamed, and the people who actually have talent and creativity inevitably burn out and fuck off to do something else.