Isn't this at the level now where it can sort of self improve. My guess is that they will just use it to improve the model and the cost they are showing per evaluation will go down drastically.

So, next step in reasoning is open world reasoning now?