I have heard that sentiment many times already. It's a statement that is broad enough to capture any benefit and disavow any costs, for some later day when one can choose the nature of their measurement.

How would you estimate actual gains and losses to education for LLMs? What would good look like?

The actual gains will be very similar to the gains that I thought would come from projects like Khan Academy (which was also quite hotly contested among teachers when it first came out).

I don't agree with your rhetorical statement about the nature of my argument. I'm not trying to disavow any costs. I'm saying that AI represents a unique opportunity in the history of humanity to encapsulate knowledge and carry it through space and time in a highly scalable way.

We have an epidemic of schools doing poorly in educating our students with constantly dropping standards, and an unwillingness to raise teachers' salaries. Education is extraordinarily expensive and I anticipate LLMs can be incorporated into a larger automated course load that is not stuck in the rigorous, true/false analysis that other automated systems have been stuck with in the past.

LLMs are already excellent at critiquing humans' writing and providing feedback that a talented editor would typically have to be paid to provide.

So, to summarize, the gains will be in the average cost of a high school education, the availability of that education across the world, the reliability of the information presented in the course curriculum, and the consistency with which we can educate at scale.