> but that the last 10% is going to take a long time to attain, and may not be reachable with the current approach. I would argue it is much more than 10% (any

> but that the last 10% is going to take a long time to attain, and may not be reachable with the current approach. I would argue it is much more than 10% (anything about reasoning is basically missing/just imitation), and the current model won’t really scale in that direction.