Model training is already trivial, just expensive. As LLM optimized HW appears the expense will drop. By 2030, I’d expect decent consumer hardware to be both fa

Model training is already trivial, just expensive. As LLM optimized HW appears the expense will drop. By 2030, I’d expect decent consumer hardware to be both faster and two orders of magnitude cheaper than what’s in use today for model training.