Learning Is Slower Than You Think
nisheethvishnoi.substack.com
nisheethvishnoi.substack.com
"We don’t need another miracle of steel and jet fuel to fling us across the sky at five hundred miles an hour. We need wheels that creak, hooves that strike the ground with an honesty no turbine can mimic. We need the long road—the smell of manure and lilacs, the way your thighs ache after a day of swaying on a wooden bench.
Because in the air, you don’t hear the frogs tuning their throats in a roadside ditch. You don’t feel the wind peel an apple’s scent from an orchard as you pass. And maybe—just maybe—you don’t really arrive anywhere at all. Not the way you do when a horse brings you there, slow enough to see every dragonfly and every dying sunflower bowing in the field."
And this is how AI slop often arrives: so recognizable it hurts.
The real proposal should be that slate dot com type "Is Food Really Good For You?" or "Hands Are A Completely Unnecessary Part Of The Arm" article authors should be replaced by LLM.
I like the proliferation of LLM slop, because it involuntarily reveals the emptiness of an enormous proportion of actual human writing. You can't help but see it, even if you don't want to. You end up forced to talk about the author's resume in defense.
Someone in this thread accused LLM of writing the OP.
If so, you're almost certainly wrong.
How can AI itself be so great if it's output is literally AI slop, which is basically garbage?
When it comes to writing ordinary natural language, the constraints are less rigorous, and LLM output tends to focus on rhetoric, where the goal is to fool people into accepting a conclusion rather than actually supporting the conclusion from a logical perspective.
"Childhood itself, reframed: not as a journey to be nurtured, but a system to be streamlined."
"But what starts as adaptation ends in control."
"And when you name your school Alpha, you signal something deeper: Not growth. Not care. But dominance."
"What’s lost in Alpha’s world isn’t content—it’s interiority: the struggle, the slowness, the human friction."
"And in that shift, the soul becomes overhead."
"Human scaffolding gives way to sleek sterility."
"...we risk raising children who are fluent—but forever waiting for the next prompt."
"It’s not that it fails—it’s that it succeeds. Brilliantly. But at the wrong task..."
I know this is aside to the point you're making but the every dying sunflower line really got me as a reader of poems.
As an argument of course there's barely one at all which was closer to your point.
Don't know about the "without ever arguing why one is better than the other" part.
The LLM made a very strong case for horse and buggy. It just takes the right sensibility to appreciate it...
Ahh..the iconic and obnoxious ChatGPT marker. Half the fiction I read these days has this BP rising phrase.
> And some readers may ask: if schools like Alpha accelerate the basics and then give students space to explore—what’s wrong with that?
> The problem isn’t the intention—it’s the architecture. You can’t optimize one part of a child and expect the rest to unfold naturally. Learning isn’t modular. Once efficiency takes hold, it doesn’t stay in its lane. It reshapes what matters.
The argument, if I'm following it, is that the Alpha model is going to do well, and therefore the model will change to remove the current 3/4 of the curriculum used for exploration, which will make the model worse. But Alpha won't care, because teaching students worse in the name of efficiency is the natural end point. I am not convinced.
I think the article would have done a much better job starting with Asian cram school culture and how AI tutoring is being "fueled by state incentives and parental anxiety", instead of having that as a throwaway thought in the middle of an article otherwise focused on Alpha school.
> It’s a signpost for a broader trend—one that treats friction as failure, learning as delivery, and formation as a plug-in.
I think building something provides a lot more friction (and learning opportunities) than reading books. We had computer classes for at least 7 years in school and beyond some loops and recursion, I didn't really think I understood computers. One month of trying to build an app and that worked. Similarly, hundreds of hours of YC (and other) video content paled in comparison to trying to salvage a startup that was going bankrupt.
> And what’s the real tragedy of this model? It’s not that it fails—it’s that it succeeds. Brilliantly. But at the wrong task: a perfect system solving for performance, not presence.
I don't know why this line feels like it's written by ChatGPT. Maybe because it has that tone of trying to say something deep in a verbatim manner that is the signature style of ChatGPT.
> Training children to outperform machines may win the game, but it misses the point: Machines don’t need meaning. We do.
And I haven't found more aimless people than those coming out of current highschools. They then go to the universities their peers go to, get the job their peers do and try to fill their weekends with entertainment. No judgement, but I don't think AI will reduce any of that.
This article sounds like the usual ideological objections, lots of vague claims that amount to “I don’t like it”.
There is nothing more threatening to a failing institution than a solution that delivers results.
The US education sector is quick to embrace any new fad that sounds good but doesn’t work, building thinking classrooms is the latest. Productive struggle is another.
Yes learning takes time, but it doesn't have to be painfully slow and unproductive.
The education system in US is a disaster and getting worse. The response from schools like the San Francisco school district has been to lower standards and remove higher level material.
I see tremendous potential in Ai tutoring. I use chatgpt to help me learn new material daily. Why should school be any different?
Every human civilization and tribal community does this.
It's why we in the USA spend so much time in the early grades not just on arithmetic and ABCs, but on stuff like the Pilgrims, the first Thanksgiving, George Washington, Benjamin Franklin, etc. It's not really anything anyone needs to know for day to day life in 2025, but it's a set of common stories that we all can relate to, draw analogies to, and yes rally around if called to do so. I'm not sure really how much this is true today, I'm reflecting on my experience in the 1970s.
The article is saying something is lost when the algorithm becomes the teacher, as when the AI is the instructor.
1-on-1 teaching (with the parent as teacher) is not that; whatever problems it has, it's never a problem of a too-powerful algorithm.
Well said. Did you just make that up?
However, what remains to be seen is if 1-1 with an LLM reaches that productivity.
Humans, and especially kids, are weird. They're picky, they're self-sabotaging, and they're short-sighted.
From what I've seen from kids, they will absolutely destroy their own education in the name of fucking around. Presumably, you didn't let that happen. Will an LLM? We're already reaching the breaking point with teachers - a lot of parents just say "I don't care" when confronted with the fact their kid is illiterate.
Anyway -
Education is a very slow and difficult thing, because it is paired with the maturing of the individual along with sharing enough concepts to spark connection _but_ not so much as to overwhelm. Adult humans can make the individual judgment that algorithms can't - WIS not INT.
Regarding dysfunctions.
Children can absorb knowledge _very_ fast in the right environment. I'm uncertain how much that replicates across the whole population, but you can see this in the top-flight homeschoolers.
When we were looking for better environments for our kid than the neighborhood school, we realized that private schools have an advantage in that they can select the parents and effectively curate the environment. Most of the issues in the public elementary school were generated by dysfunctional parents combined with certain political choices on classroom management. The second is materially fixable given work - the first is not something a school board can fix. Much of the discourse on education needs to relate to the total educational delivery across family/student quartiles of capacity, rather than trying to cherry pick a student, or a class, or - . You see, it's very hard to teach when certain students have no self control and disrupt the class continually, and there is no facility to remove them from the class for the good of others. And as we all know- 90% of students are above average, and our kids are _definitely_ gifted. So there's no easy political way to solve the environment problem - which happens to be the key drag on education today.
I'd have thought repetition was more about retention than the described "learn one new thing every day".
I'm also curious what new means here. Is attempting to research and retain (eg SR) some small fact the goal? Or would it be learning novel modals, such as new skills or new ways of thinking - which i imagine would be quite difficult in a day of course, just thinking out loud.
For SR to apply i imagine it depends on the underlying set of SR cards/etc. If you're just topping off on well known material then SR can't do much - but if it's well known material, there's often not even a review stage to have - all the material won't be flagged for review.
However if you have several items that you're reviewing then the SR algo has determined you don't know it (or perhaps they're new, but that's a bit besides the point).
At the very least it’s an interesting experiment—still unclear if or how well this sort of thing will succeed.
https://x.com/RahimNathwani/status/1933354196792979590?t=bMl...
My main problem is that they claim:
- it can work with any cohort
- that the gains come primarily from the 2 hour learning platform
But actually:
- the gains come from the high quality and large quantity of adults
- only 10% of the benefit comes from the platform (according to Matt Bateman, an education thinker who now works there)
- there are definitely large selection effects, too
I like the idea of it. But AFAICT there's nothing special about the execution. It's just that public schools (both government-run, and charters):
(i) can't choose their students, and
(ii) aren't trying to maximize learning, and
(iii) have parents who want something 'normal'.
So it's easy to do something better, if you can get a few folks to pay you a lot of money, and you have investors willing to burn additional money.
(BTW at their new school in San Francisco, opening this fall, they're planning to charge $75k/year, so probably no need for VC subsidy)
They might iterate to something that can scale. But right now they're making claims that I don't think would stand up to scrutiny.
Regarding their charter school application in Pennsylvania: the fact that they're trying to get taxpayers to pay so much for their software (which Matt acknowledges only accounts for 10% of the gains) seems like a trick to extract money from a taxpayer-funded 'not for profit'.
Separately: if I were paying $75k/year for a school for my child, I'd be disappointed if they were using IXL and ALEKS for math, instead of Math Academy.
0: $59,320 for the 25-26 year according to their website.
You could employ a private tutor full time for that sort of money.
But actually I don't think you could hire a full time tutor for that amount in San Francisco. Many public school teachers in SF cost double that https://transparentcalifornia.com/salaries/search/?a=school-...
You could hire one full time tutor per two children, though...
more expensive than Harvard
I definitely recommend it for those who enjoy thinking about what life is like for people of different perceived "intelligence" levels.
- Spaced repetitions
- Active recall
That is, some people will approach spaced repetition as a means to re-consume the information. But, sadly, that doesn't really work. Instead, those that use it as a means to attempt active recall, effectively "guessing" the answer and then immediately checking that result do see massive gains.
To that end, I'm curious if this describes your approach when it works?
That is the mental model I have of it, I don't know how accurate it is though.
This is why "flash cards" have the question on one side and the answer on the other. If you just looked at all on one side, then you lose retention.
I think the idea is that you want to get your mind trained to create the answer based on the prompt. You don't, necessarily, remember the answer and the prompt, together. Instead, you want you mind to be able to make the answer.
Next up is LEARNING first, then remembering. Grinding trivia is not learning.
Combine elaborative encoding + self-explanation + dual coding as your core encoding triad. Use mnemonics selectively for hard-to-remember facts. Embed spaced and interleaved practice into your review schedule via tools like Anki or RemNote.
This is true for many things, but there are exceptions.
For example, my son studied the international phonetic alphabet (IPA) using Anki, without any prior exposure. Now, when he uses a dictionary (paper or online), he can verify the pronunciation of an unfamiliar word.
Another example: foreign language vocabulary. These are rote facts. There's no meaningful distinction between 'learning' a foreign word and 'remembering' that word. You cannot say you've 'learned' a word if you cannot already 'remember' it.
You haven't learned anything you can't remember, you've just heard it once.
Also, this is horseshit on its face. Sometimes I just need to to remember what dots and dashes in what order mean "J," or how -ar verbs are conjugated in the progressive past. Learning things and remembering them is the same thing. Once you have the facts remembered you can abstract groups of them into parameterized functions, and abstract those parameterized functions, and test how they work in strange contexts, etc.
But these are all rewards you get for remembering things. On the other hand, I see no benefit in thing that I "learned" yesterday that I don't remember today.
edit: I agree 100% with the prescription, though. But creating lots of spurious associations around pieces of data you want to remember is a temporary measure until you can make more real and useful associations. The process of creating real and useful associations between new information and old knowledge is what I'm referring to when I say "learning." It happens whether you want it to or not. It's a product of the sheer amount of repetition times the salience of the new fact to old knowledge. [edit: although number of sleeps seems to have an effect.]
It is really hard to guess what is the right way, everyone has a different idea.
Effort and resources seems like the biggest missing piece in practice.
In almost everything more effort and resources seem to be able to overcome a lot of bad decisions and mistakes but there is not much hope if there isn’t enough effort.
> The problem isn’t the intention—it’s the architecture. You can’t optimize one part of a child and expect the rest to unfold naturally. Learning isn’t modular. Once efficiency takes hold, it doesn’t stay in its lane. It reshapes what matters.
use of "not this — that", em dash, staccato sentences to make a point, unnecessary metaphor, etc
I stopped reading at this point.
but i have no idea how to type an em-dash. you just put in a double-dash/hyphen (i.e. "--") and any modern word processors will know to convert it to an em-dash for you.
I suspect that lots of writers who use this trick are getting unfairly slammed with "omg, youre an AI! no human knows how to type that character!"
> 72 results
It doesn't prove anything but it's a highly suggestive signature. I would hazard a guess that it would be rare for most writers to use an em dash every other sentence.
Oh, what a world we live in.
> There was a silence at the table—not confusion, but recognition.
Then keeps happening again and again and again to the point where it would be obnoxious even if a human wrote it.
It's unbelievable to me that there's a kind of person who would publish an article about learning that is so obviously AI generated, and from a professor nonetheless. Maybe this is some sort of experiment or an ironic joke that went over my head.
Slow is smooth. Smooth is fast.
We are living in a world where high schools are graduating students that can't read and don't know their times tables. The critiques of alpha school are reactive inertia for a system that is already badly broken.
Didn't read further — that's an AI-generated article. Shame!
Sometimes if you force them into such conversations, you can uncover some surprising blind spots where they don't understand very basic things. I once had this experience when I was talking to a guy I worked with in crypto space who had a PhD in mathematics; smarter than me in many ways. But one time we were having a discussion and he just couldn't understand my argument about basic price and demand dynamics, he couldn't accept my argument that the Bitcoin blockchain isn't actually scalable, that the main reason the network doesn't get congested and grind to a halt is because when the Bitcoin transaction volume is high, block miners increase the transaction fee as a way to drive down demand for transactions. He couldn't gasp the role of market incentives and could not accept the fact that the Bitcoin blockchain itself can't physically process more than about 4 transactions per second.
When you think about it, what is now gently called AI is more like think by number - TBNI. You take all the current thinking, you search it and compile it. How you get AI should tell you something. Searching. So it tells you what color to paint the square with a 7 on it. That is the good part.
The bad part is that if the drawing is wrong, so is the color. If your paint-by-number kit is titled "A Duck" and the numbers are on a drawing of a dog, then TBNI will not really help you.
As I look around I notice the degree to which we humans use following. You learn to do things by watching others. A map tells you how to go some place. Why go to college? The curious thing to me is that so much of what we follow is wrong already. People follow a well worn path even when they see it leads know where. Why? Because many people are unaware there is an alternative.
I blame our education system for this. The essence is to trust what other people know, not what yourself. Silly educators.
If you are a follower, TBNI will make you redundant, because TBNI will be faster and cheaper.
But you can think if you try. Perhaps the most important thought any person can have is when they experience something and say to themselves "This is stupid". Follow that thought. :-)
The author is a professor in Computer Science at Yale. As well as an ex-research at Microsoft and IBM. You would think this person has the necessary writing skills to write this article themselves. There is no excuse to use AI for writing this. It makes it much harder to read. And always leaves me wondering if I don't understand the point the author is trying to make or if there wasn't one to begin with.
Overall I'm just annoyed by how often I see people use genAI to write stuff for them. Do they think people won't realize it is generated by AI? Do they just not care? Or has it become socially acceptable to write emails, articles, and memos with AI? Just give me the f***g prompt. Then at least I don't have to deal with reading the AI slop.
Let alone how ironic it is to write an article that talks about AI in education, with AI.
PS: I'm not trying to attack the author. This is becoming a widespread issue and I don't want to single him out.
We have a lot of good marketing to try and persuade people that everyone isn't a moron, but ultimately you can't change the fact that they are. That's why you'll notice that 99% of work in every organisation is done by a handful of people.
This schooling by design locks out the possibility of world leading medicians, scientists, mathematicians and economists if it ignores learn by rote. Just because repetition is boring doesn't mean it's not essential, and frankly before the age of 18 is when you're most susceptible to getting benefit from that learning.
Same goes for a lot of the new age and home schooling. It's not that that these models can't work. But they're normally fighting against an education system which has been refined for almost 100+ years in the developed world. Instead, the good ones embrace and extend without extinguishing. Just because it has it's faults (it does) doesn't mean you throw out the baby with the bath water.
Then again kids going to such a school are going to be wealthy and connected which in certain strata in society matters more than efficacy or meritocracy (regardless of what is preached).
This nonsense that ML(AI) will be the ultimate recall always is the worst of fluff. It's right behind "terminator". ML systems are at best a great tool on hand for experts because query and verify to them is faster than lookup, then reference and then query is to do manually. That's the point. And no, not a single ML system has proven reliable outside of a lab context in raw isolated offline fact verification...
This article is so blatantly written by chatgpt
i'm curious how these ai schools will be different
I was worried for a second we might’ve been over-generalizing for the sake of online karma.
Seriously though, your superiority complex is showing, you may wanna cover that up.
My experience has been the exact opposite, almost exclusively. I witness them not performing as well on tests in the high school/college years, but excel at the "real world" afterwards. To the point that they look "behind" in school, but then leapfrog their peers in the post-college years. They're particularly better equipped in the subjective "adulting" areas, and not just the "maximize income" area.
> Homeschooled kids often do well on tests - but they all lack an awareness of the real world because they don't meet republicans or non-christians or whatever is outside of their parents groups.
I will assume that in "lack an awareness of the real world", you mean first hand experience with non-scholastic environment of all varieties - If so, this is incorrect. Most home-schooled children have significantly more opportunity to experience "real world" than public school ones. This is because the very nature of home-schooling often involves traveling and often interacting with different people. (e.g., counting money - public school, in class with same teacher same people, home school might take to grocery store and talk to manager who shows how she does it; history - read book, explained by teacher, maybe show pictures, video clips; Home school meet with individuals who lived through or have relatives that told stories, show historical objects.)
If you mean that they lack socialization, that is also incorrect. Homeschool parents often prioritize character development, moral understanding, and meaningful relationships over the large-group peer interactions typical in public schools, which may align better with long-term social fluency but differ from conventional metrics.
Of course, there are terrible outliers where home schooling ends up nothing more just laziness by the parents, and ends disastrously.
[0] Medlin, R. G. (2000). Homeschooling and the question of socialization. Peabody Journal of Education, 75(1–2), 107–123.
Smedley, T. C. (1992). Socialization of home schooled children. [1] Smedley, T. C. (2004). Social Skills: A Comparison Study. National Home Education Research Institute.
[2] McKinley, M. J., Asaro, J., Bergin, J., D’Auria, N., & Gagnon, K. E. (2007). Social skills and satisfaction with social relationships in home-schooled, private-schooled, and public-schooled children. Home School Researcher, 17(3), 1-6.
Many your points do not sound like something uniquely possible because of homeschooling.
> Many your points do not sound like something uniquely possible because of homeschooling.
I am open to elucidate if you give me details you think that is not possible.
That’s my problem with your specific examples. None of them seem unique to homeschooling.
I will say that the likelihood of school trips in public schools is possible, but significantly fewer and less likely than for home schooling. A home schooled child can take a field trip every week. This will not happen for public school student, just for logistical reasons.
My point was to counter the argument that "they [home schooled children] all lack an awareness of the real world". If you look at my response from that context, it is clear, at least to me, that a public school student has significantly less opportunity to experience "real world".
This is all anecdotal of course, but of all the homeschooled kids I knew, even going into highschool, they were freaks of nature. I mean, just deeply unlikable people. They were always "off", often rude, often meek, and had this sort of air of desperation for socialization, with none of the tools to make it work.
Now, some of them grew out of it - I met some in highschool working at the Dairy Queen. Turns out, a couple of years of dealing with food service will help you out socially too. But a lot of them never did. Not when I knew them, anyway.
I don't know why, I don't know how, and I can't tell what, exactly, about the public school experience causes this. But it's what I've observed.
Uh, not all of us that homeschool are right-wing fundies. A sizable population of homeschoolers in the US do so because the public school system is atrocious, but we can't afford private school. You better believe we prioritize a diverse set of experiences and ideas, and I think we are more favorably positioned to deliver on that.
Us public education is really good, but people will always complain about something.
Yes, that's probably why the first example of an out-group in GP is "republicans", while the second is "non-Christians", and the broad category label for the examples is "whatever is outside of their parents group".
Actually reading the excerpt you quote to react to would help.
> You better believe we prioritize a diverse set of experiences and ideas, and I think we are more favorably positioned to deliver on that.
And on what evidence should I believe either of those things?
My mistake, misread the sentence. Not that it is an excuse, but it is a common stereotype.
> And on what evidence should I believe either of those things?
I don't see how one could even go about collecting empirical evidence to substantiate either of those claims, and I expect you know that. Regardless, I am sharing an anecdote - not writing a thesis. Believe what you will.