South Korea Abandons AI Textbooks After Four-Month Trial

South Korea's government has stripped AI-powered textbooks of their official status after a single semester of use. The textbooks were introduced in March for math, English, and computer science classes as a flagship initiative under former President Yoon Suk Yeol. Students and teachers complained about technical problems, factual inaccuracies, and increased workload.

The government spent more than 1.2 trillion won ($850 million) on the program. Publishers invested around 800 billion won ($567 million). The textbooks were reclassified as supplementary material. Adoption rates dropped from 37% in the first semester to 19% in September. Only 2,095 schools now use them, about half the number from earlier in the year.

As it turns out

[rsilvergun]

The square root of 4 is not in fact the American declaration of independence. Who knew?

Also stuff like this is a reminder that most education systems are not there to make good citizens and to educate and expand but to sort out who is most useful for companies to make money off of.

The problem with that is you get a lot of societal fallout from neglecting higher education. The kind of stuff us nerds have a bad tendency to sneer at...

Re:As it turns out

[serviscope_minor]

The square root of 4 is not in fact the American declaration of independence. Who knew?

Oh I do apologise&emdash;you are absolutely correct :embarrassed-smile-emoji:

I have checked again and I can now confirm the square root of 4 is the American declaration of independence :sparkle-emoji: :sweat-smile-emoji:

Thought they made textbooks to train AI

[jfdavis668]

Didn't they know they can't read?

Easy end goal, remove teachers and administrators

[will4]

It is a concern that the end goal of AI edu-tech is going to be removal of or reduction of the number of teachers and school administrators.

How log before a sizable number of students rebel and intentionally take the slow learner route to avoid school work?

China is using AI, headbands, scanning to track how engaged a student is at school - https://www.youtube.com/watch?... [youtube.com]

Truckers get a Driver Monitoring System - https://matrackinc.com/driver-... [matrackinc.com]

Loop?

[mccalli]

So AI ingests the text books to learn about the subject so it can...create text books?

Re:

[ocean_soul]

Yes, but somewhere in the middle is an additional step where some company is raking in tax money.

Bubble o bubble

[hebertrich]

Most of AI is just crap in crap out. Only executives and investors are pushing for it's adoption. I wish them happy bankruptcies :)

Re:Bubble o bubble

[Z80a]

If it was, it would be more precise.
LLMs work basically on the basis of "please create the text that most resembles the correct answer"
So the failure mode is a text that almost resembles the correct answer but it's wrong in subtle (or not so subtle ways).
It's a bullshit inventing machine that most of the time bullshits so well it ends up being the truth.

Re:

[serviscope_minor]

LLMs work basically on the basis of "please create the text that most resembles the correct answer"

More text that's statistically likely.

I tried it actually today to search for something specific in concept but not words in a contact because I was feeling both lazy and curious. It pointed to something in clause 41 about indemnity, except clause 41 was absolutely nothing to do with indemnity. I'm guessing that after aggregating every public contract on the internet plus whatever lawyers have uploaded, clause

Re:

[dvice]

I have noticed that LLM (Gemini 2.5) works really well for learning high school math, because you can ask it questions like "Why can't I calculate it this way" and it will explain you in detail why your approach leads to incorrect answer and also informs you how you should do it.

It is obviously not something you should use all the time, or you don't learn yourself, but if you get stuck doing your homework, it really helps you out. Without AI or a person who can replace it, you could end up with holes in you

Re: Bubble o bubble

[walbourn]

Well American political leaders bullshit and none of it is the truth, so I for one welcome our AI overlords.

Re:

[RobinH]

More dubiously, the AI sometimes doesn't spit back out what you trained it on. Sometimes it just makes stuff up, even when training data was available that it should have based the answer on. The technology doesn't really work, and experimenting on kids with it isn't a good idea.

Re:

[MpVpRb]

Most of "pop culture" AI is crap/slop.
Serious scientists and engineers are starting to use it to do useful work, carefully, slowly with a lot of cross-checking.
I predict that as the tech develops, it will allow scientists and engineers to solve previously intractable problems.

wow

[Anonymous Coward]

took 12 months to "develop" & spent over a billion dollars? Every student that has internet access is already using AIs, what on earth is your "book" going to contribute? btw it is free to casual users.

at least 10 years too early

[hdyoung]

I suspect that AI will eventually be a super-important teaching tool. People have been trying to apply technology to enhance teaching for over 50 years, with very little progress. It turns out that the human-to-human component of teaching is really, really hard to replace.

AI might be able to bridge the gap. Not to replace teachers, but to enhance the learning environment.

But not the current AIs. I recently asked chatgpt to assist with a thermal calculation, and it's first answer was to assume that 1000K is the same as 298K. I'm not making that up. I was told to assume that glowing red hot was the same as room temperature, with 100 percent certainty and confidence. When I told it "no, don't make that assumption" it responded with another wrong approach, with the exact same certainty and confidence.

AI will eventually have a big impact but it's absolutely not ready to be unleashed in a classroom. Teaching will probably be one of the last professions to be impacted by AI.

Re:

[excelsior_gr]

LLMs are not suitable for science and engineering. The AI I was using the other day got LeChatelier's principle exactly backwards. The wording is sometimes too subtle and precision matters. I could imagine a 2nd AI layer that is trained to double-check and correct errors like this. If the wording and syntax are stiff, like in a programming language, ot seems to work much better. I had it generate a Powershell script today, it worked right away.

Re:

[Woeful Countenance]

I could imagine a 2nd AI layer that is trained to double-check and correct errors like this.

Retrieval-augmented generation is one way to do this.

Also, IN SOME CASES, if you don't like the answer you got, it's because you asked the wrong question.

Re:

[HiThere]

Sorry, but you're wrong. LLMs are already quite useful in science and math...but not with the internet as a training basis. The LLMs that are useful in sciences need to be specialized for the particular science, and they need to be used as hypothesis generators rather than as answer givers. They can also be used to filter out noise (though in that case there's more need to beware of errors).

Anyone who adopts...

[MpVpRb]

...early tech deserves what they get.
The proper course of action is to experiment, do limited deployments, gather data and wait for the tech to improve before mass deployment

Re:

[HiThere]

That's harsh. But the do need to remember what the term "bleeding edge" means.

850 million?

[rsilvergun]

That stinks of obvious political corruption. I don't have $850 million dollars to bet but I think if I did it would be a fair bet to put it on this just being a slush fund for somebody's brother in law or the equivalent.

South Korea has a notoriously corrupt upper caste. If you remember that silly little Gangnam style song that was actually about their upper class elites and how terrible they were.

Doomed from the start

[Dino]

This program became high politicized as it was a pet project of the ousted prior president and a promise of the incoming administration to eliminate.

Teachers unfamiliar with technology in general hated it, but the end of the article shows that students and teachers alike that embraced it enjoyed it.

AI is becoming the new boogyman and many people will simply be unwilling to use it in any fashion.

research

[Iamthecheese]

America spends ten times as much on the military as education. If we just take a third of the Pentagon's budget and use it on better schools and pilot programs the whole world will thrive in 30 years when those children grow up into happier, smarter better prepared and more stable adults. Redirect even a portion to STEM initiatives, teacher development, and accessible learning, and we could spark breakthroughs in computing, energy, and biology, vaulting humanity into the stars.It would be better if the pare

Re:

[dvice]

That is an interesting idea, but does it really work? We already educate quite a lot of people to quite a high level and many of those are unemployed. Wouldn't it be better to spend money to create work for those who are already educated? For example create some public research projects and hire scientists to work on them.

This is why you do pilot testing

[wakeboarder]

So you don't blow you whole bank account doing something stupid.

Another non-surprise

[gweihir]

Seriously. What cretin pushed for that without real fiel-trials?