Local AI: why, what, how

Why A year ago, I was convinced that running AI locally on consumer hardware would never be useful or cost-effective. The gap between frontier models and local models was too wide, and closing it was