The challenge of self-sufficient AI model pretraining

A user argues that pretraining is hard, pointing to what they see as a speed advantage for labs willing to bootstrap or distill over those building self-sufficient stacks. They call the comparison a rare compliment to Microsoft.