How fast a kid learns 5th grade syllabus is far less important than how well they learn to get along with other kids and form friendships and basically learn how to live in society. Cooperation, conflict resolution, public speaking, group bonding etc. You can't learn any of these things from an AI

A good AI, sure, this could be plausible. Current ones get too much wrong.

The much bigger issue is we're talking about children, not adults. How many children would be motivated to self teach, every day, for years on end, even if you had an AI equal to a human tutor.

The problem isn't one of motivated learners being forced to drag their heels amidst their unmotivated peers.

The problem is that the core function of LLMs, the whole basis for their existence, is completely and entirely truth-agnostic. Not only do they not know what is the truth and what is not, they don't even know what the difference is. LLMs are very good at guessing what word looks like it should come next, they can make very convincing statements, they can be very persuasive, but those words don't MEAN anything to the machine and they are made without any consideration for accuracy.

They are literally making everything up on the basis of whether or not it sounds good, and every crackpot bullshit conspiracy theory from flat earth dumbshittery to very sincere-sounding arguments that birds aren't real have been included in the training data. That all linguistically SOUNDS fine, so to an LLM it's fair game!

And even curating your training data to ONLY contain things like textbooks wouldn't cure the problem, because LLMs just aren't capable of knowing what those words mean. It's why they can't do even basic Math, the one thing Computers have been incredible at!

Using an LLM as an actual teacher is genuinely worse than no education at all, because it will just create a generation that, instead of knowing nothing, will very confidently be wrong all the time.

Social aspect is missed

"but first, chatgpt: please motivate me"

Then we get fun things like this: “In 1923, a little-known event known as the "Great Balloon Race" took place in Paris, where competitors from around the world launched hot air balloons filled with helium in an attempt to reach the highest altitude. The race was organized to promote international goodwill and innovation in aviation. Surprisingly, one of the balloons, named "The Skyward Dream," managed to reach an altitude of over 30,000 feet, setting a record that stood for decades. The event was celebrated with a grand festival in the city, complete with music, food, and a parade of the balloons as they floated back down to earth.”

So... you're saying that a positive learning environment is better than a terrible one? The AI part is ancillary to the scenarios you set up, isn't it?

"AI is better than having the student learn in a terrible learning environment."

"A homeless alcoholic is a better language teacher than having a student learn in a classroom whilst being beaten about the head with a stick."

You're saying AI is better than a bad teacher. Maybe a bad AI is worse than a bad teacher, and maybe a good teacher is better than the best AI. I just don't know how setting up such a comparison is constructive.