Hacker News new | ask | show | jobs
by elcomet 13 days ago
We didn't have the compute required (GPUs powerful enough to parallelize forward and backward pass). This compute is what allows us to train from human knowledge or distillation.