I suspect that, in practice, what many enthusiastic advocates mean by “learning AI” is actually “learning to need AI”.

In other words, the aim is to get kids used to using AI as soon as possible, so that they do not learn the skills to function without depending on it.

If you’re smart AI saves you time getting to something you could probably achieve anyway. If you’re… not smart… then it will be a necessary crutch for you to get through life.

I can see the angle for making sure kids start using it before they develop the skills to become independent of it.

Especially since kids these days aren't even very good at using computers:

http://www.coding2learn.org/blog/2013/07/29/kids-cant-use-co...

It seems to me that if someone can read and think critically-- they can RTFM and get much better much quicker at computers and AI than people who spent all their time tapping an ipad to watch the next video.

I'd think really the only AI skill you need is the ability to think independently and be able to verify the results you are getting or spot when something is wrong in the response.

It would take a few sessions at most to take someone from 10 years and get them fully up to speed with AI tools since they have zero learning curve.

I think exercises when student is given pre-generated AI output and told to identify as many issues or mistakes as possible might be sensible. Not sure how long creating such exercise would take and what should be the tools or sources to verify the output but that might be helpful excersise.

Similar to Google and Wikipedia lessons back in my day.

We had something like this when I was in school but it was reading a news article, or the same story covered by different people and identifying the bias or missing information.

Evaluating AI output it’s not a skill on its own. It’s just general critical thinking and literacy.

You also need to understand the limits of AI and that it has limits that a human that gives you usually correct and authoritative answers does not have.

I think it comes easily to the sort of people who comment here. Moat people have a very vague understanding of computers in general.

Presumably prompting skills like https://github.com/califio/publications/blob/main/MADBugs/CV...

I’m not sure anything was clarified. Nothing about that conversation is special or unique?

Are these supposed to be the "skilled" prompts? This just reads as a basic conversation and not as particularly well-written or well-defined prompts. So far everything I've seen about prompting "skills" has just come down to being able to articulate and critically think a bit.

Yeah, my post was kind of sarcastic and it doesn't show.

As someone who sells AI... You'd be shocked at how bad people are at using AI.

My 6 year old kid who watches me is a better prompter.

You absolutely need prompting skills to use AI usefully. You need to know how to eliminate sycophancy, how to ask for and check primary sources, and how to use follow-up questions.

I've been using AI for some legal issues, and it's been incredibly good at searching for case law and summarising the key implications of various statutes - much more efficient than web search, with direct links to the primary sources it finds.

I'm still the one gaming out "What if...?" and "Does that mean..?" scenarios and making sure the answers are grounded in the relevant statutes, and aren't mistakes or hallucinations.

It's not so much a prompting problem as a critical thinking and verbal reasoning problem.

Learning those prompting skills was very useful for you, but in the context of schools it's a lot more difficult to make the investment worth it.

Schools are slow, by the time the teachers get around to teaching the sophisticated techniques you use today, those techniques will be obsolete, the new AI models will require completely different style of prompts.

As for critical thinking and reasoning, those are even harder to teach. How can teachers teach what they don't know?

> It's not so much a prompting problem as a critical thinking and verbal reasoning problem.

And that means you have to learn without AI to understand when the AI is wrong. This is just how its dangerous to use a calculator without knowing math since you wont spot when you entered things wrongly etc.

It would probably be useful to learn how the models work under the hood - particularly about high dimensional vector spaces - as a counter to magical thought. But I doubt that's what is meant.