Maybe when it comes to any coding-related learning, I can agree with that point.
However I think the future of education in general is very bleak. When we have plummeting scores for the PISA exam across the world [1], I fear people will get dumber and dumber.
There will always be those who thrive, but the easier it is to obtain information/knowledge, the less I think people have an incentive to do it. And I don't buy that AI will solve all intellectual jobs anyway.
[1] https://www.dw.com/en/pisa-2026-results-reading-math-scores/...
This is exactly what AI companies need, so that's development should be celebrated. There are two paths to capture the value that's currently wasted on paying people:
1. Build a system that's superior to nearly all people.
2. Make people worse, so even an inferior system is superior to them.
Getting AI into the classrooms and degrading the educational system are important for path #2.
It's good not to put all your eggs in one basket, so I'm glad the AI industry is pursuing a multi-pronged strategy against their competitors. Who cares about the waste and destruction for humanity, as long the investors get good returns?
Learning to write will either help all the humans by extending their capabilities, or will make them stupid -and dependent on writing.
Socrates: It will definitely make them stupid, and I am suspicious of this new witchcraft.
Maybe he was right, but I’m kind of enjoying having both writing and AI around.
I'm sure AI is going to do even more damage but COVID was an important factor here as well.
Education itself has lost access to the tools that were making it cheaper and easier to teach.
Learning something is a side effect of overcoming friction. AI removes friction.
I’m using ChatGPT to learn German. I can immediately adjust its responses and language level(it says I’m at A2/B1) and tell it how I want to study.
I find it to be a far superior tool than anything else I’ve used.
If education is all dependent on the use of AI, then we're cutting off everyone but the most wealthy from access to education. Seems very regressive to me.
1. This is an assumption but I think it's fairly well established that current LLM prices are far below operating costs of inference, so prices will go up, maybe not 10x but some multiple > 1
There’s a lot of individual anecdotal responses that people come up with in these threads, and I’m happy for the positive examples.
I do think people are capable of extrapolating, especially since all of us have the experience of seeing students wanting to escape studies as kids.
It’s identical to the gym problem, everyone wants to be healthy and fit, but the effort sucks and we avoid it.
Education at scale is the same issue. We basically nurture/force crops of students over a 20 year time span.
The people who wanted and could learn by themselves always had the possibility to do it with books, audio and videos, and AI just add a layer of interactivity on top. Self-learning never replaced standard education, even in the high bourgeois class because they answer to different problematics.
ChatGPT might work for you learning German because you have a reason to care about learning it, but for most kids the goal of school is getting good grades, not learning.
The first thing I did was ask it to continuously adjust its difficulty level based on my replies(A1-C1), it told me I’m in the A2-B1 span. Then I formalised my training on Akkusativ and Dativ, which is basically A2 level.
I have structured the lessons like this:
Every day I wake up with a new lesson. It starts with the nominativ/dativ/akkusativ table. Now I’ve also added wechselpräpositionen including translation to Swedish. It then gives me 5 sentences for me to fill out. If I have time and energy, I continue the exercises throughout the day.
This allows me to continue the practice at my own pace and at my own time.
I only type and read. I can already pronounce quite well but this may be the next step.
COVID generation.
The pandemic was barely 3 years ago, and stretched over 1-3 years depending on where you lived.
It will likely have a material impact on most developmental statistics for another decade.
We end up with shit results and where they arent shit students are rushed though as if to prevent them from thinking about the materials.
We have libraries full of unused education formulas. Some 2000 years old. Then there is nutrition, sleep and other work/rest ratios, physical health.
How expect results without learning to teach and learning to learn?
Onto what should one anchor learnings if the suroundings are always the same?
I never hear a word about spaced repetition in school. Its so basic i must conclude they knew nothing about what should have been their area of expertise.
When things get bad enough we will [finally] have reason to improve.
Maybe the great work of Salman Khan is merely a hint?
MOOCs were something I personally hoped were the solution to scaling education. However it didn’t work out, with courses having single digit completion rates.
This is for free, high quality, immaculate pedigree, learning material available at any time and any place.
Years of additional research had to be done to figure out how to raise completion rates, which creates more complexity on systems and delivery.
Unfortunately, that improvement is now moot, because LLMs obviate the friction and effort.
The point of doing a homework exercise is to provide learning friction for students.
LLMs obviate friction.
I would add Bloom’s 2 sigma problem to the mix.
Not to long ago i wouldnt have believed what LLMs "learned" today, surely we can make a similar effort with humans if we want to.
Im not suggesting we go back to beating kids with a stick but there might be something equally effective. Lots of unskilled labor is practically a life time of torture. If you ask me we shoyld gradually introduce the torture in stead of making it a surprise. There are better people to ask.
Im literally at the point that my main discovery method for this sort of thing is word of mouth in real life. Everything on the internet is poisoned
LLms dont help here either.
Part of learning is about struggle. Seems many here want to disregard that fact and just run to the LLM lmao. Yeah you can do that - the knowledge won't really stick.
The initial creativity is still with us, not the LLM. It's the Art/Craft thing - we have the Art, the machine does more of the Craft (and it always did do a fair amount of the Craft).
I'm seeing folks who couldn't (for whatever reason) work out how to code, starting to produce some amazing stuff with the LLM doing the coding now. It's their creativity, their project, their vision, still. The fact that they didn't write the code is meaningless.
From what I’ve seen, the ability to prompt quality output with AI is entirely proportional to experience and skill level.
> From what I’ve seen, the ability to prompt quality output with AI is entirely proportional to experience and skill level
In my direct experience, it's not my coding skills that I'm using for this. I don't even look at the code any more. It's my experience of developing stuff and managing projects that is enabling me to run the LLM better than a newbie [0]. Teaching someone to code so that they can run an LLM is probably not the most efficient way of doing this.
[0] though I have friends who can't code and are producing better stuff than I am, by just prompting and making mistakes.
It's a real shame a leader of an education company doesn't value the actual learning process. Those "low value" processes are the building blocks to understand and master high level concepts. I'm not spending my day to day thinking about the low level memory allocation in my job, but that knowledge and work helps immensely in how I approach the higher level assignments and wider architecture. And I still kick myself because my assembly knowledge is is only 1.5 semester's worth.
And yes, I still look up how to construct an array in C++. Not really because my memory is bad (okay, kind of), but more for the practice of reviewing API's and parameters and maybe making new revelations. I probably knew internally some 12 years ago I could set an initial capacity in one of the array's constructors. I didn't really "understand" the value in that until one random work day 4 years later as I was working through some in-house container packages.
You can even say that doesn't matter if you know what resize() is, but it's those little bits of learning nuggets that builds you up from a "fine engineer" to a "great engineer". Less the knowledge and more the work ethics built to obtain it.
That’a the low value they’re talking about. And all of that stuff is accidental complexity.
To use their example, you need to know that dropping duplicates is a good idea (and why). Knowing what the syntax and api calls to do that is, in my opinion, unnecessary.
Even your example has unintentional complexity. A 'duplicate'? What's that? Do we compare the memory it points to? Through use of a custom comparator function?by an id or tag? APIs and entire languages are designed around how we define such seemingly simple terms.
Ultimately a language, be it a programming language or human one, is about communication. I find there to be value in knowing how to communicate effectively, even if employers increasingly disregard the value of such knowledge. Without it, we'll quickly find out why it's fundamentall quickly.
This is the same thing as pressing the power button on my computer; I want the thing turned on, I don’t want to have to know about memory training.
And then we wonder why apps are getting so much slower and bloated. That's fine if you don't want to think that low level, but someone needs to worry about performance. And others will need different needs for defining a duplicate in their different use cases.
The people who need to most think about performance are those making the frameworks and languages. Why are bloated electron apps so easy to make but building native cross-platform apps with similar ergonomics hard? Why are people choosing electron?
Yes. I am not a web dev, but I tend to fall into that category in my domain.
>Why are bloated electron apps so easy to make but building native cross-platform apps with similar ergonomics hard?
If you want to be blunt, it's political. Each OS wants to ultimately build its own moat, so there's less incentive to come together and make a standardized solution. Then Linux is a less centralized OS that makes theoretical cross platform support even harder. Cross platform applications get made despite that friction to fill the obvious demand.
The why's of which cross platform framework wins is hard to say. It only becomes obvious that whoever breaks out tends to win more, and it gets a massive flood of support, funding, and clients. So it snowballs up. Those who don't simply struggle to stay relevant.
Electron likely wins in the market aspect that bloat isn't enough friction to turn off a modern consumer with a modernish (10 years or younger) desktop. And it especially doesn't matter to phones (which tend to skew newer).
I don’t have to remember assembly intrinsic since the compiler deals with that, I just describe the higher level goal and it implements it.
> But today that's no longer the truth, and for the better. I am happy students don't have to go over the same repetitive low-value processes that I had to go to. Students can now focus on more high level valuable tasks: understand the domain, compare evaluation methods, compare models, balance tradeoffs, present solutions/reports, etc.
You honestly believe that students that don't have the tenacity to stick with learning a simple set of verbs and grammar are going to have the tenacity to learn concepts steeped in jargon?
I do mean learn, which AI use has been repeatedly shown to be detrimental to.
But unlike the author of this post (or cited authors) I'm not sore. AI is providing a much better education model for students. So it's us that we have to adapt, not complain that the world is changing.
For example, in Machine Learning our projects are sometimes extremely long and repetitive. Because before AI, it was a valuable skill to be able to recall APIs quickly (what is the method to drop duplicates in a dataframe again?).
But today that's no longer the truth, and for the better. I am happy students don't have to go over the same repetitive low-value processes that I had to go to. Students can now focus on more high level valuable tasks: understand the domain, compare evaluation methods, compare models, balance tradeoffs, present solutions/reports, etc.
It's like in the 50s when students had to memorize tables of logarithms. Nobody misses that.
We are in the process of recreating all our projects to adapt to the new world. We'll see how it goes.