But self driving cars are on the roads, and they are expensive hardware that can kill lots of people. Mistakes need to be exceptionally low so it's a lot of ord

But self driving cars are on the roads, and they are expensive hardware that can kill lots of people. Mistakes need to be exceptionally low so it's a lot of order of magnitude improvements but minor visible changes before it hits major thresholds. I'd also argue that LLMs and image ai has grown far beyond linearly over a fairly short time horizon.