Comment by nine_k

I hope this is the future. Offline, small ML models, running inference on ubiquitous, inexpensive hardware. Models that are easy to integrate into other things, into devices and apps, and even to drive from other models maybe.