[intended]

Proving you can pass an open book test ? Or doing graded assignments?

That wasn’t something possible prior to LLMs. Being able to cheat code or generate homework assignments was not this trivial either.

[thunky]

I think people need to figure out how to demonstrate they can add value on top of AI rather than showing they can do it all from scratch with a "closed book".

[intended]

This is at a level of abstraction that allows one to say something hopeful, and let the details blow up in one’s face for later.

A better example would be “we don’t use fork lifts to do our gym exercises”

The whole issue with AI is the verification labour. This requires people to have the skill / discernment to make out if the output is useless or not.

Building this human capability is fundamentally a product of learning, training and experience. All of which require exams, tests and verification.