AI inference gets cheaper by splitting work across chips and time

Original: Inference keeps getting carved up, and every cut makes intelligence cheaper.

Source: x.com ↗