[coffeefirst]

I think that’s correct. There seems to be a lot of fundamental limitations that have been “fixed” through a boatload of reinforcement learning.

But that doesn’t make them go away, it just makes them less glaring.