Forget what it eats to continue improving.

Realize what it already has.

A foundational language model with no additional training is already quite powerful.

And that genie isn't going back into the bottle.