This isn't binary, teachers have limited bandwidth and supplementing real teaching with AI supported research can work.
Students need to understand what AI is and is not.
Something got lost as we transitioned from parents telling their kids to 'go play and do not come back until dinner is ready' to us handholding & driving our kids from one social/sport activity to the next.
While it causes fights with my kids, I blank refuse to do certain things for them (fix their internet connection, repair a toy or bike) I'll give them the tools and they know to ask when they get stuck.
The act of learning through failure is a dying art these days. It's 'teach to the test', stressed out teachers, parents having too much to do compounded by taking on second, third jobs in this economy.
Teaching our kids how to leverage AI is something parents should invest in. Its part of a broader education landscape.
As a parent I agree with you, but watching other kids get a leg up because their parents continue to jump in and save them. And your own get the double harsh reality of learning failure and then being punished for it through loss of placement and status. It's tragedy of the commons since everyone must pass no one can fail.
> Teaching our kids how to leverage AI is something parents should invest in. Its part of a broader education landscape.
The same was said for laptops and tablets. Now there is a push to remove them from the curriculum. Having them meant the students didn't retain anything because they were distracting. I feel using AI in place of pedagogy will have similar results. Why bother learning to retain any information when AI knows everything and can recall it in a split second.
> Why bother learning to retain any information when AI knows everything and can recall it in a split second.
This is a real question though. One I've been asking since the early 00s. Learning means something different than it did in decades previous. Merely remembering easy to look up facts is not the slam dunk it used to be.
Memorizing facts was never the slam dunk in and of itself, but just a path to something greater, and I don't think that's changed. Think about the most obvious example - what we're doing right now: language. You can only fluently speak and communicate with others because you've memorized, to the point of complete subconscious mastery, a completely ridiculous amount of information and knowledge - thousands of words, countless grammatical nuances, cultural and contextual connotations/implications, and much more.
Imagine trying to speak without having this base of knowledge. Even if possible, you'd never be able to use the language in a comparable fashion. And I think this generalizes to basically everything in every domain. If you want to get good at chess, you need to study a ridiculously massive amount of tactics/patterns. If you want to get good at basketball, you need to simple practice dribbling and shooting a ball into a hoop for many thousands of hours, and so on.
The whole point of the memorization is to make it feel like it's not memorized. Like when you see these words, or come up with your own - it requires absolutely zero effort. There's nothing magical there. It's the exact same way that a master chess player can easily play a competent game of chess with 1 minute on their clock or while blindfolded.
If, and as, we start doing away with the rote, I'd expect we're going to see worse and worse outcomes. Arguably, this is already happening as Western education is continually 'rethought' while China rockets ahead of everybody with an education system that looks very much like a more centralized version of the US education system of 70 years ago.
You hit the nail on the head. It sounds silly to say, but it simply is useful to know things. And to know things, you "memorize" them. More specifically, you build neural pathways as the information becomes better known to you, and which can be activated quickly. This is why I maintain a massive Anki deck for anything I find curious and want to remember, inspired by Michael Nielsen (https://augmentingcognition.com/ltm.html) and others.
Even if AI is very powerful, it is very obvious that someone actually knowledgable in a given domain can leverage AI more usefully than someone without that base of knowledge. And to be actually knowledgeable, as opposed to vaguely familiar with terms, takes significant effort over time. Luckily we have tools like spaced repetition that make that effort easier.
The last thing I'd add is that it's useful to know lots of things in various domains. Much like an LLM, our brains draw connections between disparate things which we know well, in the process of diffuse mode thinking.
Was learning easy to look up facts ever slam dunk useful?
It had its place in the 80s when looking up involved at least finding the right book. But that time was already on the way out in the 00s. My impression was more that education likes to ask for facts because they are easy to test for.
What you really want to teach are concepts and broad strokes. Nobody needs to know on which weekday the Soviet Union collapsed, but an understanding of the causes of the collapse is useful. Similarly, you don't need to know the speed of light to more than one digit of precision, but knowing the role of the speed of light as the speed at which information can travel through space is useful
That kind of knowledge also continues being useful in the age of AI. It allows you to see the connections in the world, and know which questions you even have to ask
You need to be able to hold facts in your own working memory while you think about them. If you do this with the same facts, and think through different combinations, they’ll stick in your memory for longer.
I never sat down to memorise trivia about the Roman calendar, but I know Julius Caesar was assassinated on the 15th, the ides of March, because of repeated exposure while thinking things through.
It seems like old assessment techniques could luck in to testing comprehension, because the people who understood things would also remember them. This devolved into the recall being seen as the goal, instead of as proxy measure of something harder to test.
Former teacher here. Your example of the USSR is a good one to demonstrate the importance of a broad base of knowledge. In order to actually understand what "the Soviet Union collapsed" means, you need to know facts about what precipitated the Cold War. What communism means, what democracy means, and so on. Where Russia is on a map, and what "satellite" countries are. These are all concepts, too, but more atomic that build to the bigger idea.
Any "concept" is composed of interrelated facts, all the way down.
>> Why bother learning to retain any information when AI knows everything and can recall it in a split second.
> This is a real question though.
Is it? Why bother to learn how to add 2 numbers when you can just use a calculator.
> Learning means something different than it did in decades previous.
Not really. This is just the acquisition of a new tool. It can possibly be incorporated into education like how calculators were used in the 00s/90s, but you must know how to do the work without the tool.
You don't need to know how to do the work without the tool. Knowing how to add two numbers together is not the point. Knowing what numbers to add and why is what you need. Just computing the answer hasn't been an interesting skill in my whole lifetime.
> You don't need to know how to do the work without the tool. Knowing how to add two numbers together is not the point.
Would you really raise your kids to not know how to add 2 numbers on their own?
I wouldn't even raise my kids to not know how to wash their clothes without a washing machine.
Wouldn't want them to be useless on their own, to be dependent on tools to take care of them.
It's fine to use tools to augment what you're able to do, but it'd be a kind of failure of their elders / of their environment to not teach them how to live on their own, independent, or at least dependent on as few things as possible. Tools aren't necessarily always going to be available in their lives. Having options for their life's necessities is good.
It also helps to know when the job's done well. Just because clothes come out of the washing machine doesn't necessarily mean they're clean.
If we're talking:
> Why bother learning to retain any information when AI knows everything and can recall it in a split second.
Kind of scary to think of adults that need AI to do their thinking for them, with the adults not knowing how to verify anything and just trust the output.
I mean, my kids can both do mental math because it's something they sort of absorbed along the way. But yeah I don't super value it as something to focus on. I've never done laundry without a washing machine nor would I know how to teach someone how. Though I'm sure I could look it up.
Once smart glasses with displays or just talking to Ai in your earbuds becomes humanly ubiquitious then I believe the game changes. As long as it's 100% correct humans become know it alls.
I hooked up my local LLM with zim-tools and a 120GB copy of Wikipedia. With some very strict instruction it's turned into a bar-trivia machine, it can answer factual questions faster than I can look them up and it's all offline.
The fun starts when you ask it more complicated questions like 'i have a device that communicates with microwaves and draws a maximum of 20mA at 12v, is it safe for use near humans?' It will look up microwave safety, ionizing and non-ionizing radiation, workplace exposure limits and eventually come up with 1/4W is already safe and the device likely is not a 100% efficient microwave transmitter.
Don't agree. If you know no facts, you'll be unable to think even the most basic thoughts. The more you know, the deeper your thinking can be as well as the more subtle.
For every fact you learn there is a necessary context and numerous exceptions when it's applied in other contexts. If you expect AI to spit all relevant facts at you every time you're clueless, you will have no idea when AI is wrong or lying. You will have to trust everything AI (or its owner) tells you.
Yes. Fact fluency allows higher level of abstraction. This is very clear in the case of mathematics - try doing algebra without having your addition and multiplication tables memorized, kids who fail algebra are often missing these basic facts, along with other basics like order of operations.
It's not just math either. Try to be an electrician without memorizing the basic principles of electricity and how they interact. You're going to have a bad time.
By induction we can see how this scales upwards, basically infinitely, to higher and higher levels of thinking.
I think this is mostly compounding confidence at play that unlocks your mind to be more fluid when faced with a task to do/problem to solve that actually requires skill (not just following a series of known steps). It’s like public speaking (unless you have really bad stage fright) you generally will give a better presentation if you really know your stuff, or at least will be less nervous allowing yourself to get into that flow state.
You used a calculator every time you needed to multiply or divide simple numbers while reducing matrices to echelon form in introductory linear algebra? Or when doing simple arithmetic necessary to find integrals and derivatives in calculus? I bet you in fact did have those tables memorized, even if it wasn't a conscious effort.
I guarantee you that you've memorized these tables, you just didn't put any effort into deliberately doing it. It happened accidentally as you used numbers frequently.
Of course, maybe I'm wrong. Maybe you DO need to use a calculator to quickly tell me what 8 * 6 is. But I would be highly surprised if you got a degree in math and still needed a calculator for that.
For 8*6? I definitely don't know either of the 8 or 6 tables by memory.
I of course know how to figure it out. Let's see my instinct might be to do 8x2 first because I do know most of the 2s. Then 16x3 I can do in a few steps of addition. 20+12+16 then 48
And yes then I checked on my calculator because I'm posting this online and don't want to look like an idiot.
I think the shift is in part from learning behavior indicating an ability to apply (‘I wanted to do something so bad I went to the library/college/etc.!’), to the free availability of knowledge leading to a delinkage of knowledge v.s. the ability to apply knowledge and get a desired outcome.
AI can really help the outcome part (by just doing it for you), so for better or worse I see that delinkage continuing.
People who bother to actually learn while using AI will have an edge on those shovel slop, but there’s clearly a difference from previous eras.
It might look like those kids are having a leg up, but really they aren't learning to stand on their two feet and the wheels start coming off when they aren't under the wing anymore.
Being able to say 'I attended $famous_uni' counts for much if not most in many circles. Upper classes are beyond merit; being there and staying there is all that's needed. In fact, stepping in to save offspring is a way of life for them. School is just the first time it happens.
Is that summing the wealth of every single person in a (usually growing) group of inheritors, or averaging it?
If 1 guy with $16 billion turns into 16 grandkids with (inflation adjusted) $1 billion endowments each, it sounds to me like the family fortune was preserved extremely well, the family is simply bigger.
Mate, I’m visiting Europe right now. The tail of generational wealth is so long here you can even see it. We can certainly see a few examples of family wealth being squandered. And it makes a great narrative about karma and balance. But I’ve met even more families that while not individually wealthy will never, ever be poor in the foreseeable future for the next 5 generations at least.
That severely depends on legal and social system you operate it. It is not universal truth, that was artifact some concrete historical period. And the loss is frequently result of large war, revolution, violent displacement, that sort of thing.
I assume they weren't advocating for using AI to just blindly give answers to the students...
You can argue the same about humans. Growing up, I definitely did encounter people who charged money for tutoring and never taught anything - they just gave the answers. Wasted money.
As for dealing with kids whose classmates' parents help them a bit too much... I honestly don't have a solution. From my experience growing up, beyond a certain amount of help, you definitely are hampering the kid if you overhelp them. I guess I was just fortunate that none of my peers' parents cared to, so I could still be near the top of the class.
Oh, and our final scores were heavily weighted on in-class tests/exams, so even if some kid's parents did overhelp, they wouldn't do well overall.
> but watching other kids get a leg up because their parents continue to jump in and save them
Can you give us an example to what his looks like in your context (not necessarily you and your kids)?
> The same was said for laptops and tablets. Now there is a push to remove them from the curriculum. Having them meant the students didn't retain anything because they were distracting. I feel using AI in place of pedagogy will have similar results
Laptops, tablets, and AI are just tools. If the folks wielding these tools (or ostensibly teaching others to wield these tools) are not using a tool effectively, then we shouldn’t be surprised when the outcomes are unimpressive.
Simple example — put a hammer in the hands of a skilled builder, and you can get a deck or a house frame. Put the same hammer in my hands, and you will get nothing more than bent nails and sore thumbs.
All that said, just like I feel confident that I could learn how to use a hammer well if I took the time to learn and practice, I think students and teachers can learn to use their tools (laptops, tablets, AI, etc.) well if they take the time to learn and practice (ideally with some good scaffolding).
> If the folks wielding these tools are not using a tool effectively
I struggle with this argument when the subject ("the folks") turns out to be "many people", perhaps even a majority of people. If one teacher is an outlier that uses it badly then sure, the teacher is to blame and be judged. But if a collective is using it badly, then perhaps the capability of the collective should have been evaluated before handing the tool to them.
I don't have an answer - it is bad to withhold a tool from the hands of people who will use it badly. But the overall outcome has to also count.
> If one teacher is an outlier that uses it badly then sure, the teacher is to blame and be judged. But if a collective is using it badly, then perhaps the capability of the collective should have been evaluated before handing the tool to them.
You won’t get any argument from me that most teachers are poorly trained. Money can largely buy someone out of that trap.
That said, just forbidding a tool that is critical in modern life just because some/many people are bad at using it does not strike me as a particularly constructive solution to the problem.
> I don't have an answer - it is bad to withhold a tool from the hands of people who will use it badly. But the overall outcome has to also count.
My “solution” to this is not a popular one. Lower expectations for public mass education.
Over the last 120 years, we have taken a curriculum that was targeted at elites and forced it onto almost everyone.
High quality education should be available and cheap or free to everyone, but I really don’t think it should be mandatory.
If we stopped compulsory education at 8th grade and replaced most high schools with some sort of other skill development program or work/study program, the expectations and outcomes would probably be better aligned.
Large parts of the tertiary education system would be gutted, but maybe that’s a good thing.
We were all once junior coders. We
Also didn’t understand the code we had written or what we were being asked to do. I have yet to meet any junior coders who hasn’t written garbage code that hasn’t managed to squeak through unit testing.
I’m embarrassed by some of my early work and simultaneously impressed it managed to work at all! And someone was dumb enough to pay me money to do it?!
> The same was said for laptops and tablets. Now there is a push to remove them from the curriculum. Having them meant the students didn't retain anything because they were distracting. I feel using AI in place of pedagogy will have similar results. Why bother learning to retain any information when AI knows everything and can recall it in a split second.
I remember growing up being told similar things about the internet and how you can't trust anything written on it. There was some truth to that, of course, but how many people will say that of search engines, despite them being filled with utter crap?
When I grew up we couldn’t use the internet as a reference for our research. There didn’t exist a standard for citing articles and blogs as there was for journals and books. Eventually, that was fixed.
I use AI extensively now, it also cites the websites and electronic documents it uses. It’s a little frustrating in that it cites the URL but not the exact paragraph or chart cited. For that I have to read the entire article. Sometimes it gets it wrong because the context is incorrect. It’s still a great tool, but without my last generation experience in high school and undergrad research, someone without that experience would blindly follow wherever it takes you.
In PBL the teacher adopts a constructivist approach - supporting, guiding, and monitoring the learning process as opposed to delivering a defined curriculum via lectures or seminars.
Agentic AI is arguably a better fit for this as it can scale infinitely to monitor student progress, offer timely feedback on design choices, and adjust difficulty dynamically based on how the student interacts with the problem.
Your main contention is one I'd broadly agree with - that use of Agentic AI for tutoring abdicates intellectual agency, allowing the AI to perform the critical synthesis and analysis that the student should be doing themselves for learning to occur.
The solution, however, is clear: guardrail the AI to promote core pedagogical principles rather than replace learner cognition by mandating things like instructional friction and the strategic fading of AI support as students build competence.
As explored in the paper I link below, by balancing proactive scaffolding with necessary and beneficial cognitive challenges, AI can preserve the productive struggle essential to effective learning.
(A final thought: have a look at the current field of companies who specialise in turning training materials into engaging interactive experiences for the likes of Compliance and Security training. This is a corporate field now being automated and democratised for use by the average student in Academic pursuits in an on-demand fashion. Appropriate AI tutoring is basically the same concept.)
AI has its own benefits but its really a bad idea to use it in education as a primary source. LLMs have lots of information, reasoning ability, however they have their own limitations and by the nature these limitations just amplify.
Few days ago my son was working on a problem which asked if square is a rectangle. Turns out his class teacher and external tutor (who was using AI) had different opinions on it. My wife consulted AI as well and she also came to the conclusion that square is not a rectangle cause AI told her so, AI told the same answer to external tutor. I understand that the context matters and if one probes AI more with additional context it may come up with correct answer, but this depends on the assumption that the person asking AI for something already knows the answer.
This is not a problem with AI, this is a problem with a human accepting an answer without understanding why it is correct. The tutor could also read a book that states "A square is not a rectangle" (or vice versa) and repeating it back to the student as established fact.
The whole point of having teachers is to give intellectually developing and still vulnerable children the best possible support they could have from a human which is ideally interested in their well-being, being responsible and in doing a good job.
Replacing that with AI is the kind of SV-stupid which starts making sense after the educational system was hollowed out for profit reasons and is failing. The human in that system is just checking in and DGAF.
When shape names are first introduced, many kids will flip out if you suggest a given polygon can be multiple things at once. So they define the shapes (square, rectangle, rhombus, parallelogram, trapezoid, etc.) so they are mutually exclusive, with categories like quadrilateral as residual categories. You can tell who is doing AoPS/Beast Academy because they argue based on the standard definitions, whereas the teacher will insist on the curricular definition or on the “most specific answer.”
Another classic is “how many roots does a parabola with real coefficients have?” The answer is always two by the FTA, but the textbook Algebra 1 answer is zero, one, or two because they haven’t covered complex numbers yet.
I read their comment as the teacher (who was correct) and a tutor (who used AI) disagreed, not two tutors.
With the amount of blatantly false information I've seen LLMs spew out in response to my very simple questions, I can very easily imagine one would get this basic information wrong somehow, although I did just Google it (as Gemini, and particularly the on-search-result-page version of it, seem to be the absolute worst of the common models in my experience), and it did get the answer right.
I am very curious what either person asked that led them to say that AI had told them a square is not a rectangle.
The obvious question of “is a square a rectangle?” is definitely NOT causing any LLM to say “no”.
Hallucinations are an issue but they genuinely are no longer a concern for any modern non-tiny LLM for basic facts that you would expect a K-6 course to cover. Maybe you might be able to find a few random hallucinations in knowledge covered at the high school level but I’m quite skeptical even that still happens. Yes, they can still, at times, hallucinate sources but I cannot recall the last time I actually saw this happen. I think it was at least a year ago
Let me say this in a different way: If some rectangles can also be squares, then this shows that a square is a rectangle. This is not a logical deduction; it is an intuitive one.
LLMs have lots of information, but they have zero reasoning ability. That is why they are so bad at giving answers: they can't reason from one thing to another, but instead everything is a statistically likely result. If using an LLM for learning, you have to make sure to check everything it says or you will take in false info, guaranteed.
LLMs have had a reasoning setting for a while now. The reasoning is what allows them to go through and test each output to see which output/answer ends up aligning with known context the best (whether that be constraints that steer it away from some outputs, or information/hints that steer it towards the desired output).
The problem with the tutor's question is that they're going to AI to answer this question and not asking underlying questions like "what makes a square a square" and "what makes a rectangle a rectangle", then doing their own thinking to determine if a square's qualifications also fit inside the qualifications to be a rectangle.
Heck, even if you just feed the underlying into AI it'll reason the answer in a logical manner:
> In standard geometry, A square is a quadrilateral with 4 equal sides and 4 right angles (90° each).
A rectangle is a quadrilateral with 4 right angles. Its opposite sides are equal and parallel, but all four sides do not have to be equal. So the key distinction is: Rectangle: four right angles. Square: four right angles plus four equal sides.
>
> That means every square is a rectangle, but not every rectangle is a square.
Asking LLMs to just give you answers is the real antipattern. But if you remember that the only thing it's good at is finding patterns given context/constraints, it can be an extremely good tool.
Also, I have a feeling the vast majority of the populus is using free versions of Claude or ChatGPT. Thankfully recently ChatGPT Free is GPT-5.6 Luna, but before the response qualities and capabilities of the free chat output were less than stellar.
The better alternative is to reshuffle what teaching means.
Right now teacher largely execute a curriculum this is predefined and do behavior management in US public schools at least. There is not a whole lot of creativity and a lot of rigid administration.
Given that teacher salaries are one of the biggest expenditures it makes sense to try to gain efficiencies while trying to improve results.
Under the current scheme, the results are headed downhill. More of the same won't solve any problems.
That would depend on which country I was in. Where I live it would probably be at least 20 USD per hour. But I live in a country (Norway) where unemployment benefits and protections against becoming unemployed in the first place are strong.
>Students need to understand what AI is and is not.
Do we really understand what AI is and is not? For example we all know AI can hallucinate, but how does that knowledge help us in practice? If I'm leaning something new I would have to fact check everything which would take about as much time as learning from verified sources.
I just watched a youtube about cruise ships. A near-riot broke out during a movie quote quiz because the AI that made the questions confused Casablanca and Gone With The Wind. If it aint ready for quiz night, it aint ready to teach kids.
People formed their opinions on LLMs in late 2022/early 2023 when ChatGPT was first released and taking the world by storm and then never updated them after nearly 4 years of improvements.
Hell, just being able to search the Internet did MASSIVE things for reducing hallucinations.
Ask a color blind person to spell "red" and "green", then ask them how many red apples are on the table.
LLMs work on tokens. They literally cannot see individual letters. Yet, unlike humans, they can instantly create a tool to solve this task for them and never be stumped by it again.
It is important to understand how this technology works. It isn't magic. Your example is solid. Similar to the Chinese Room.
Do you think your legislators or most users understand tokens and predictions and what is going on? Do you think that those who bombed a school using it know this?
Until frontier models can account for the disparity between what they can do and what the person (or other agent) is asking, we will continue to suffer bad information used like good information.
That tokens cannot count letters like that is not a brain-dead gotcha. Ignoring how humans are using these models is lacking in critical understanding of the real world. People are literally going to die due to humans not understanding how this works (and already have). True brain-dead, and body-dead.
A friend of mine put his kid into a school called Alpha [1]. It has no teachers, only guides (?), who are well paid grads but not formally teachers. Kids go in in the morning, do two hours of computer study with an AI program, and the guide only steps in if the kid has a question or doubt the system cannot answer.
Then there's recess and afternoon is all taken up with "Life Skills" classes such as entrepreneurship and the likes.
Kids are tested weekly and progress through grades at will. For example, his kid would officially be in 3rd grade, but she's doing 3rd reading, 5th grade math, and 4th grade writing. She can easily test 3rd grade topics now.
I am ambiguous about this model. Love the "learn as much as you want" even if it means learning 2, 3 grades ahead, but the no-teacher rule and so much computer use seems a little strange. Also if the kid is not extremely self-motivated, they will be left by the side of the road by the school.
I have found for many young kids lack of studiousness is not that they are not self-motivated. It is that whatever is being taught is worse than boring to them. The lack of motivation in schools comes from a lack of interest or too slow of pacing.
By the time they get out of primary school most have had the curiosity that drives motivation beat out of them.
Yet, when I go through the skills and knowledge acquired during my school education there are many bits in there that I am now eternally grateful for, that I did not understand the need for as a kid.
I am not sure where I would have ended up if left to my own devices, e.g. as a kid starting first class I was bad at languages and hated it with a passion, in the last class I had the single best German and English final exams in my school district and language had grown on me.
Alpha shows the results they do because they charge exorbitant rates (up to $75,000 / year depending on region) to staff the school with multiple individualized tutors per student.
It is a for-profit product which is sold to parents and state-level customers more than it is a school.
1. I have 0 problem with the "for-profit" product if the kids are well-educated.
2. I would challenge the "multiple individualized tutors" claim, as a parent with a kid actually studying there says there is no tutoring, only answering questions on the content or the software being used.
I'd have just sat there for hours trying to trick ChatGPT into giving bomb making advice, and then disrupted the others by sharing.
There have been many variations on this idea of "letting kids learn on their own". The basic issue is the one you highlight, which is that it works if the student is very self motivated. Most students aren't. Or rather, they're highly motivated to do other things.
Most children are naturally curios. The challenge is to make them curios about the things you want them to learn, and make that more interesting than finding out how to make bombs or what the consequences of sharing that with the class will be
I think if you can get into the chemistry just a bit, then the math behind the chemistry, it can be a vehicle. I mean it's the same with programming. Why write code (boring, yuck) when you could play video games? Then the kids see code makes video games. Suddenly you have all these programs with kids making Minecraft mods or whatever. But finding each child's "bomb" would be hard for a teacher to do, especially with a common curriculum
Alpha doesn't use a chatbot, it's still a multiple choice / text input into a box type of thing, I don't even think it's AI graded, though the questions may be AI generated, I'm not sure about that.
In the early days they were using off the shelf curriculum software and layering their own steering AI over it.
A good friend of mine works for them. It’s a weird combination of (1) probably the long-range future and (2) absolutely not going to use my kid as a test subject.
If you believe, as I do, that half of students only achieve half of their academic potential, then experiments like this are some of the most valuable work we could do in society.
But again, let me know how it goes in 25 years and my children can enroll my grandkids.
In terms of 3rd level accession and standardised testing I'd imagine they boast great statistics, and that the students are self-evident in communicating a degree of confidence and holistic education beyond their traditionally educated peers.
However, I would fear that the issue with home-schooling would manifest: that the rarified atmosphere would render them incapable of engaging or coping with the brutal reality of binary-minded Authority figures and hierarchical social constructs in their work and life in the 'real world'.
To writ, some of the most important lessons I learnt in school were to do with the concept of 'fairness', my own ability for emotional regulation, navigation of social hierarchies, and the symbols and signifiers of popular culture. The curriculum as a whole came a distant second.
N.B. Public (read:Private) schools in the UK are about £30k a year. Boarding schools hit about 50k in the UK/Ireland. 'New Age' or International schools for the children of diplomats and HNW ex-pats like Nord Anglia or similar go from about €/£20k upwards. 40k a year isn't outrageous in context, even outside of America.
> However, I would fear that the issue with home-schooling would manifest: that the rarified atmosphere would render them incapable of engaging or coping with the brutal reality of binary-minded Authority figures and hierarchical social constructs in their work and life in the 'real world'.
This keeps coming up every time, and it's always pointed out that there is scant evidence supporting this view at a large scale.
Every motivated parent I know who homeschooled his kid didn't deal with these problems.
And plenty of those who went to a traditional school did. They aren't even outliers. I'm sure anyone who went to a regular school can think of kids in their class who exhibited the problems you mentioned.
No doubt, plenty of homeschooled kids do as well. Just as you have a distribution for regular schooled kids, you'll get a distribution for homeschooled kids. Claims that the distributions for the latter are skewed in the "wrong" direction are never presented with evidence.
> To writ, some of the most important lessons I learnt in school were to do with the concept of 'fairness', my own ability for emotional regulation, navigation of social hierarchies, and the symbols and signifiers of popular culture. The curriculum as a whole came a distant second.
I went to a regular school, and none of what you say applied to me. The curriculum/instruction came first, and learned little of the former. I'm sure I'll find classmates who'll say the opposite. This sounds like a classic case of post-hoc attributions.
As a parent of two teenagers in public magnet HS (one of which is top 10 in the US) I still say homeschooling is by far the best and what my wife and I would have done if it was an option.
Home school is best, private school if you can, public school if you must.
Its a weird stereotype of home education that people with no, or limited, experience have of it. It is far better preparation for the "real world" than schools are.
Then I think you probably learned the wrong lessons.
The lesson you should have learned would be that we should tear down the "brutal reality of binary-minded Authority figures and hierarchical social constructs." We only tolerate these things as "normal" because people go through their lives not actively trying to tear them down and break them.
> Then I think you probably learned the wrong lessons.
> The lesson you should have learned would be that we should tear down the "brutal reality of binary-minded Authority figures and hierarchical social constructs." We only tolerate these things as "normal" because people go through their lives not actively trying to tear them down and break them.
I was a product of an environment like that (magnet school with a lot of unstructured time and good access to resources), and it didn’t take me very long to run into the Binary Authoritarian Figures’ version of “you and what army?”, and they were largely correct.
There are ways to undermine parts of the establishment, sometimes even at scale, but the process is not always obvious or easy to access.
> So you gave up when it got hard? That’s kind of what I read out of that.
That’s a very uncharitable reading.
I have made my mark in the world by exploiting the Achilles’ heels of “authority figures” (which is usually just some form of a lack of empathy). I have caused a few to fall, I’ve made others reform, and I’m just an annoyance to many others. I have trained others to do the same.
That said, I could get whacked out of existence figuratively or literally if I crossed the wrong person or group.
There are others like me, but (as I mentioned before) we aren’t always easy to find.
I know the above may seem grandiose, but I’ve had a chip on my shoulder about these “authority figures” since I was a kid. I’ve done something about it, but I also didn’t choose to bang my head against a wall and call it progress.
I think that's exactly what it reads as... I'm not trying to be a dick but, "I'll remember this and <authority-figure> will get his or her just rewards" is... I don't know, like I said, it reads like "well I ran up against a wall so I decided to give up and wait for a different wall."
Uncharitable, maybe. But, yeah, you should definitely stand up for yourself more.
> However, I would fear that the issue with home-schooling would manifest: that the rarified atmosphere would render them incapable of engaging or coping with the brutal reality of binary-minded Authority figures and hierarchical social constructs in their work and life in the 'real world'.
Having been to a school like that, I agree with you. I had to learn those lessons quickly.
I largely opted out of being in the middle of that world, but I am still adjacent to it due to its ubiquitousness.
I got the benefits of letting kids go at their own pace etc. at a fraction of the cost but with a bit of (mostly fun) effort by home educating my kids. Tutors for two subjects (three in exam terms as English language and literature are separate GCSEs), both Oxford graduates in those subjects. No AI but lots of books and other resources.
Most kids are motivated given the right environment. Enjoying learning is natural.
The AI element of what you describe does not appeal to me. The weekly testing and assignment to grades sounds a bit rigid too.
Considering how much one to one human tutoring 40k an year could buy that does not sound like good value.
My girlfriend was homeschooled and she would definitely have preferred to go to regular school (which she attended for high school).
Home schooling sounds good in paper, but as much as school can suck, it is also an irreplaceable opportunity to learn to socialize, which is a much more valluable skill than knowing pre-university curriculum.
Alpha school struggles with the kids who are a bit slow.
If you’re the sort of kid who gets everything right first go then you join the “rocket ship” and spend 2hrs a day with AI then play for the rest of the day. You would do well in any school.
If you don’t get it right quickly then you join the “Pirate Ship” and have to spend all day grinding and obeying AI while your friends play. It seems pretty terrible.
I’m very conceptually interested in AI assisted individual learning; I’ve managed to get into a lot of topics that I would have had a hard time approaching without an LLM helping. I really like having AI rephrase concepts usually visualizations and analogies that resonate with me personally. That said, I think there is a lot of sketchiness around Alpha Schools in particular: https://buildcognitiveresonance.substack.com/p/special-repor...
this is a classic confounder problem. I'm assuming the vast majority of parents who can afford 40k annual for their kids are relatively well off, have a lot of other disposable income, and the kid is never really lacking for any needs. put them in a public school and it's likely they'll turn out equally ahead for their grade
socioeconomic status (SES) is very often a huge confounder when it comes to anything from health issues, educational outcomes, and many other things. rich kids do better - the SES of the parents of Nobel Prize winners [0], young athletes [1], etc are all indicators of this
AI could either help or ruin American education. But the Department of Education is being shutdown. Many parents are relatively absent. And functionally all tech companies, and especially startups, are rapacious in pursuit of education funding.
I think the most likely outcome is AI makes American education worse, not because if the technology, but because of the “leaders” building and implementimg it.
Not sure the US Department of Education did much that would influence this either way, except if the AI was being used to correct income-disparity effects on education.
It doesn't mandate curriculums, choose technology, anything like that? Only recommendations.
They might have been able to monitor its effects but they wouldn't have been able to direct its use or non-use; that's not what they did.
Essentially it was/is an anti-discrimination organisation, making sure that states gave an education regardless of wealth/race/status.
Slightly creepy, seems harmless now but future without supervision you don't know what they're teaching or what is going on. I would agree it's a double egged sword.
I'm all for saving money where you can, but these models really aren't great for learning because they love giving you answers and it's so easy to get an answer, but the process to get to the answer is how you actually learn. Unless you put guardrails in place to actually learn, I don't think this advice is good.
I’ve found LLMs to be very good at teaching. Better than I expected.
You have to tell it what you want out of the session. A kid could just type in the question and get the answer, but that’s not tutoring.
When I use them for learning I tell it my goal, summarize my current status, and then tell it to teach me incrementally with a quiz after each step. You can get fancy and have it reach back into history to throw in extra refresher questions.
It uses very few tokens compared to a coding session so you can use the highest model for this and the usage is negligible. Using the top model available to you is important to keep things accurate. I wouldn’t have trusted this back in the GPT 4 or Sonnet days, but with Opus, Astra, or Fable it’s great.
For longer sessions set up a directory and have it keep notes about your progress, problem points, and what was covered in the past.
This is what I meant about using them in the right way. You can totally setup guardrails but my experience has been that unless a person has good self-control, it's way too easy to just get an answer from it. You have to really be interested in learning to do this, though. Thanks for sharing!
Students have always had the curriculum and standard study material, it was always possible to study alone, but the teacher mostly provides an interactive gap-bridging layer, and the classroom setting provides some level of obligation, I could be wrong but I don't know why AI can't fulfill the interaction layer part.
Sunk-cost, social pressure and obligation are probably the most salient points for educational success in a classroom setting with a teacher of median-skill versus an AI trained on the same corpus.
The sad reality is that many of the most senior professors and lecturers in top universities are their for their research rather than teaching ability, and that the success of the elite institutions is somewhat self-selecting as regards their cachet and filtering done by their admissions process (i.e. Ivy League institutions do not offer athletic scholarships).
Some teachers do genuinely make a difference, and maybe that's good enough, but unfortunately there are many more who struggled to add value to the original material, and I don't really think this is a unique experience.
Of all the teachers I have ever had, at school and university, I can think of only three who struggled to add value to the original material. One was in the middle of a nervous breakdown (of a kind that genuinely would not have been out of place in the John Hughes movies of the day, but that contributed to him passing the following summer); the other two were in the middle of significant crises of different forms.
So that's about 0.75%.
The rest were either competent, good or amazing.
(Then again, one of the remainder was internationally famous for being a (self-acknowledged) media whore and while I know he added value it wasn't always to the subject at hand)
2.5% I meant to write there. And I know this because I just found that 2.5% elsewhere that I meant to write 0.75%.
School is now distant enough for me that I have reached the brain fog age where I cannot be commenting on internet things intermittently while writing documents…
disagree- i used it for the last few years of coursework preparing for exams. If you give it context (i'm studying for an exam, do not give hints unless asked, your objective is to help my understanding) it is incredibly helpful. This isn't advice, it's a company going out of business.
I have recently realised that people actually don't know how to prompt.
They use AI "raw" and don't understand how the question can have a drastic impact on the result, let alone using skills or system prompts.
Learning new topics or skills is a very good example, I had the same experience as you with Claude when I learned a new language.
Which isn’t surprising at all if you turn off your adblocker and look at how these AI chat products are marketed to the general public. They’re presented as all-knowing oracles that can deliver exactly what you want with only the vaguest and briefest of prompts.
Maybe some argue it's ASI but for me AGI requires that the AI understand what you're trying to ask rather than require you to carefully engineer a prompt.
If you're starting a fresh session the AI has no context at all beyond your prompt (and the generic system prompt). It isn't psychic, it can't know if you want a straight answer, an elaborate explanation or an interactive teaching lesson. You have to actually say what you want
I'm not a huge fan of prompt engineering or elaborate skills that are longer than a couple sentences. But you have to give the AI context. The more context the better
That isn't really a good defense. An intelligent creature (i.e. a human) will recognize that it lacks context, and ask probing questions to get more information. An LLM just boldly goes down a path with only a fraction of the information, which is yet another proof they aren't actually intelligent.
"Pray, Mr. Babbage, if you put into the machine wrong figures, will the right answers come out?"
If a random person comes up and asks you a malformed question how do you answer? What's worse if what if the question is not actually malformed in some contexts and because you're a wizard that has massive amount of contextual information there are plenty of right answers for the 'wrong' question you just asked.
In schools classrooms tend to avoid unlimited exploration as it would take unlimited time. On your AI prompt the next thing you ask could be anything, anything at all. Just exactly how is the AI supposed to know if you're an idiot?
Continuing the great tradition of claiming that any machine, and especially any AI, unless it takes over literally everything, is worthless. Or in the words of Pablo Picasso:
"Computers are useless. They can only give you answers." (he was not joking by the way)
Of course, if computers DO threaten to take over literally everything, then people are even more scared and react aggressively, as demonstrated in this 1939 article describing IBM's vocoders:
"The Pedro Peril" (the first AI that actually talked back was nicknamed Pedro, and described as "the latest and most fearful gadget from IBM", the journalist's conclusion was to suggest Pedro be deported as he would spread propaganda and maybe even become a dictator) (unsigned Washington Post editorial, 1939)
It really depends on how you are using AI. If you simply use it to get the answers to homework you won't really learn. If you use it to figure out the topics you need to learn, and to identify/close gaps in your understanding then it is a very useful tool for learning.
Depends heavily on the tutee. If it’s someone highly motivated looking for some guidance like you, I could see it working really well. If it’s a kid who would rather be scrolling through Instagram and needs an authority figure to keep them on track, it’s not going to work.
If the kid would rather be scrolling instagram then solve the underlying problem. They will not have a good life if they will only do anything constructive when forced to by an authority figure.
I find it very easy to engage in a Socratic dialog with the AI. If I ask it, it will actually quiz me and carry on a tutoring session. And it doesn't mind if I ask dumb questions.
yes, similar experience... you can ask the same question in 10 different ways until you understand it and AI doesn't care
I've walked away from colleagues trying to explain things similarly because if you're not getting it after a few attempts you're going to hit a wall of their ability to explain, their patience, time, etc
I don't know if I agree, if I ask a physics question it will show me the process to get to the answer, and it's really nice to have something to talk to with infinite patience to dig down into steps you don't understand. I learned a ton about implementing machine learning from using AI models. Obviously that takes a level of curiosity and effort on the learners part, but with a pretty minimal system prompt I think you'd get a pretty great tutoring system.
They rule for maths. I had an extremely good experience going from high school trig through multivariable calc with them over a year. Best math teacher I ever had.
I think the difference is between someone like you who wants to learn and a teenager who has to learn.
I used to be a tutor, and the kids who wanted to get ahead were people who more needed someone to give them hints as they learned. The ones who needed to get the grade were a lot more reticent and needed a lot more attention and prodding.
I don't think AI as it currently exists is a good choice for those folks.
My daughter worked on learning for various classes by building her own plugin and skills for tutoring that uses various techniques of teaching without revealing answers, giving advice on homework questions without disclosure, etc. She overlaid with the personality of her stern gymnastics coach. It was surprisingly effective and I was pretty impressed at 11 she had the sense to not go straight to the answers. I think the “oh the kids” hand wringing is always overblown with kids - it’s the adults that don’t adapt well.
The ChatGPT web app, though apparently not the desktop app, does have a study mode where it will not give you the answer but will help you reason through the problem like the world's most knowledgeable, most attentive teacher.
One thing that I've noticed as I've gotten older is that the introduction of calculus has slowly moved to younger and younger students. I suspect the tools are partly responsible for this. Less time on basic math and algebra.
When I first started calculus, I was furious at being taught Newtonian Physics and some bits of chemistry without calculus. It's a stupid bunch of equations you have to learn, which Newton worked out from calculus. I'm glad people are getting it more often together now, but also just a bit jealous at the same time.
A calculator is a tool that replaced the slide-rule or Abacus - no one contended the Abacus replaced teaching or teachers. Ditto an iPad which is effectively just a fancy calculator in this context.
Similarly Youtube didn't replace teachers - it afforded remote learning and correspondence courses a new medium for transmission, recording, and playback of teaching.
More to the point none of those were trained on a corpus containing the history of human pedagogy, were capable of inference, or of generating dynamic learning resources based on best-practice models with inherent self-correction and pedagogical reinforcement mechanisms.
This line of reasoning makes sense, but also I would say that a lot of technologists who are interested in education realize that they have to design something that wouldn't just give the answers away because that's not the point of learning. That seems like it can be pretty difficult to accomplish with LLMs because I bet if a student keeps pestering them, they will give up the answer or give a clue that makes the problem way too obvious.
So usually the claims about replacing tutors like in this article come from a place where they usually give a caveat that maybe for highly motivated and ambitious students, they can figure out how to accelerate their learning with AI. But then that undermines the whole argument of transforming education if it wouldn't work with students who are average or below-average in their academic achievement.
I didn't. People who think they're learning from YouTube are mostly not learning, just thinking they did. I watched someone design a common-base amplifier. Can I design a common-base amplifier now? No, I can't. I now have a slight awareness of what one does, but I couldn't design it.
Tutoring is another situation where it can succeed wildly or fail miserably, and it will depend more on the harness than the model.
Tutoring is a process, and the end result is that the student gains a demonstrable new capability. The how isn't as important as the end result. If we end up with a bunch of people with increased personal capabilities, it's a win. If we're lazy and we end up with a bunch of people leaning on crutches, we failed.
Time will tell, I guess. Could be a surprise answer to Bloom's 2 sigma problem or a very stunted generation, or a mix of both.
> Tutoring is a process, and the end result is that the student gains a demonstrable new capability. The how isn't as important as the end result.
I disagree. If, after you have been tutored, you perform well, but still need tutoring for the next year's exam in that subject, then you haven't been "tutored", you've been given a similar enough copy of the exam questions to train on.
As I keep telling my kids, good results is a side-effect of good habits. If you're half-assing things, it shows up in your results. If you studious and methodical, it also shows up in your results, just in a different direction.
AI is used primarily to half-ass things, do things without checking, getting good vibes by that sycophantic tone.
Get your kids into a rut of good habits, and they'll run in that groove forever, regardless of tutors or AI. Half-ass your way through life, then sure, they'll get fucked over by AI tutors too.
I think you are doing a lot of assuming on how AI can be used for tutoring. I'm currently augmenting Chinese in person tutoring with a pi harness that helps me out with drills, lets me explore what things mean, get similar words or context related words and so on. It's not perfect, def not good enough to be a standalone thing, def has barrier of entry of being motivated to do honest self work but it's a massive value add into the process for me.
One huge issue IMO is that right now, around the world, most people are not taught at all how to learn for themselves nor how to critically think on their own. And that will indeed result in using it as a crutch and not a good tutoring tool.
> I think you are doing a lot of assuming on how AI can be used for tutoring.
I don't think so. In fact, I'll go one further: I don't think we're in disagreement, just that I've been communicating poorly.
Allow me retry: My point is NOT "AI makes tutoring worse". My point is that AI tutoring is irrelevant to sustained academic performance.
1. Academic success is a side-effect of good habits (high correlation, not all the way to `1`).
2. A good tutor will instill the good habits for a specific subject (IME, different subjects require different habits, but all require the student to be diligent and consistent at least)
3. Half-assing things, as a habit, will not materially improve outcomes.
4. Most AI use is for half-assing things.
I did not mean to imply that AI in tutoring cannot be successful, only that AI-use tends to result in half-assing things.
Your anecdote is almost the opposite of half-assing things.
> One huge issue IMO is that right now, around the world, most people are not taught at all how to learn for themselves nor how to critically think on their own.
That's a different problem, I think. As I am learning with my two boys, you don't need to teach critical thinking specifically. Critical thinking naturally arises out of being diligent and thorough.
Having got a tutor for one subject for one of my children, I'd say that tutors are about motivation/structure/instilling confidence not just subject matter. Our tutor has been great as a general mentor during the period leading up to exams, not just as a subject expert.
I have experimented a bit with AI and have found that AI isn't always great at catering precisely to the syllabus/marking criteria of a specific exam boards even when you give them all the docs publicly available (although I haven't tried frontier models). Somebody who has worked as an examiner for the exam board knows the tricks for how to structure longer answers to both get the marks and to not waste time in the exam. Of course, it may be that I haven't prompted well enough, but if I can't manage to do that, then I probably can't expect my children to!
The way to use AI to learn is not to ask it for the answers. Read a book, make notes about how the things you learn in that book relate to other things you know. Describe those connections as if you were teaching the topic to an interested friend and ask an AI to critique the descriptions. That'll tell you which parts of your textbook to reread, or which other books to read. It'll also give you the confidence to teach the topic to an interested friend, if you have one.
It has to be a second step, a critic. If it's the first step you don't end up with any way to know when it's being crazy.
This worked for me. I was debugging a DL model and asking it a lot of questions. At the end I asked the llm to tell me what I lacked and it gave me a good explanation and pointed me to a bunch of resources i could learn from
It depends on why the tutor is hired. If the student is easily distracted, the presence of a human tutor will prevent them doom-scrolling brainrot videos, whereas an AI tutor won't have that effect on them. I have an ASD level 2 son approaching year 11 and this is why I'll be hiring a human tutor next year.
I started taking daily classes learning the Thai language in person. While I do learn Thai with AI sometimes, I do feel like an AI tutor can't possible give me the same "edge". Because with AI can just sit down, get distracted, turn it off; walk away and procrastinate. With a tutor however, that never happens; there is also personal connection, and the tutor constantly adjusting what you're learning. I am definitely optimistic about AI-learning and tutoring, but really have a hard time seeing it as replacement.
This would be good advice except that students are no better at getting information from AI than they are from getting it from books, botes, or professors. If you know what to ask, AI will give a lot, but if you don't ...
It's actually worse, because if you know what to ask, the books and professors will give you information you can generally take to be reasonably reliable; and if you don't know what to ask, you'll either not find the information in books or potentially be guided towards a better question by professors. With LLMs, you can be given incorrect information in either case, but especially in the latter case with way higher frequency than you would with books or professors.
It's more like the cost of living crisis in Australia means that despite a likely oversupply of tutors working second jobs, people can't actually afford to pay tutors.
AI tutors will absolutely be worse, not better, but they may also be only a fraction of the price.
I've been using chatgpt to tutor my son to improve his writing skills. It actually works quite well and the lessons adapt to his skill level. I seeded it with the curriculum and marking guidelines for his school year. Each lesson is timeboxed to 30 minutes and he stops if the lesson goes over time. I then feed this back into chatgpt and it adapts the lessons appropriately. It's helped quite a bit with a definite improvement in his writing skills. He actually wants to do this instead of school as he feels he gets a lot more out of it!
The way around this I've seen so far is to have an LLM write software that is verified, both in answers and instruction. The problem is who does the verification; ideally that doesn't require a teacher...
Regardless of how many reasons you can come up with for real tutors being better than AI tutors, it’s clear that real tutors can’t survive in a market that also has AI. In other industries the accepted ritual for this situation is to genuflect at the Altar of the Invisible Hand and repeat the phrase “Actually, The More Economically Competitive Option Is Better”. (It is understandably harder to swallow this platitude when the industry happens to be “your children’s future”.)
it is insane that we live in a world where knowledge is so cheap and gets cheaper and more accessible every year, and yet people complain non stop and somehow make worse decisions.
Learning from an LLM is nearly impossible because it has no opinions on how to structure the content.
It's great at giving you the illusion that you're learning though.
If you know an outline of the material and want to structure it your own way (ex, explain attention to me using Haskell notation) it's really great. But otherwise, human notes win.
So you’re saying if I pointed an LLM, say GPT 5.6 Sol, at a PDF copy of a textbook and asked it questions about something i’m stuck on and don’t understand, it wouldn’t be able to teach me anything?
learning is not a passive activity. likewise an llm can be used in a passive or active fashion.
Why is having your LLM one-shot a 200 page textbook a prerequisite to learning? That doesn't follow. If you use an LLM more like a tutor and less like a book, you can get an enormous amount of value for 1/50 the cost. Monthly GPT plus membership is half the cost of a bay area hourly tutor rate, and infinitely patient and available 24/7
Okay I was a professor for a bit. Few things I found
- LLMs cannot motivate a person to learn because they aren't people
- Learning requires rewiring your brain; you need to try and read content without explanations of every question you might have. Struggling is required for learning
- LLMs cannot generate structured content on long scales. I have tried this many times. Your average TA can easily write a calculus textbook.
- LLMs are RLHFd to just agree with you. Learning requires disagreement and tutoring requires assuredness
These are all silly non-problems (speaking as someone who works with lots of professors around how to put AI systems to effective use)
-the thing that motivates you doesn't have to be the same thing that teaches you (e.g. a book can be a useful learning tool even if it isn't human)
-you can absolutely choose to interact with an LLM in ways that generate productive struggle. If you naively ask it for answers it won't be much use in this regard. But that's just a poor choice of how to use the tool. Just like hitting yourself on the head with a book doesn't mean the problem is with the book as learning tool.
-You don't need structured content 'on long scales' to learn. You learn a little bit at a time. Long textbooks are a side effect of the economics of moving paper around. Not a prerequisite for learning. An LLM will happily mimic the table of contents of a textbook or an outline for course. And you can use future prompts to generate the bits you need as you need them. One-shoting a textbook is a silly use of the tool. Plus, we're not short on textbooks. We're short on ways students can get customized help: answers to the specific ill-posed questions they have in their heads. LLM's are quite good at this. You can have textbooks AND LLM's (and professors)
-LLM's are happy to disagree with you. Just ask them to and they oblige. Ask them to respond as if they were a tutor and they will mimic the assuredness if appropriate.
- Only highly self motivated learners can learn from a book. Most students needed lecture and office hours.
- Students (Humans) aren't going to struggle on average. They're always going to take the path of least resistance. I know of some successful algebraic geometers who did indeed hit their head on heartshorne in grad school.
- a) Structured content generated on long scales is way more effective because you can tie together many details. It would be weird to learn from chapters, each unrelated to each other in style and content.
- b) We're actually not short on students getting help. Office hours were abound at my university. I actually found that customized help was detrimental; if a TA answered every question a student had, the student wouldn't put the effort into trying to learn the material. Imo you want to see students struggle. I've seen students go from F to A- by just forcing them to struggle on the content.
- Not really. I discuss complicated math and programming with models regularly and when I say something like "actually you're wrong" they say "yes, you're right". If I ask them to be confident, they will not change their wrong answers. Humans have a much better conception of self and opinions than LLMs
I'd compare tutors to personal trainers. They're not doing anything special other than making sure you're using the equipment properly and providing accountability so you don't lose motivation and give up.
I just tried ChatGPT on a simple find an error in the proof question today and it completely butchered it ignoring the main constraint that two variables a and b were natural numbers. It said, "a and b might not be natural numbers" and then went on to produce a very elaborate, completely wrong answer.
There is a lot to be gained from 1-1 in person tutoring. Can't read the article so don't know their specific gripe. But often when learning a kid needs redirecting, reassurance, or other emotional support. Getting this from AI seems like giving up humanity and connections for the sake of silicon valley educational 'betterment' which has always been suspect.
It is shocking how quickly many people are to dismiss the benefits of human interaction and connection since they aren't quantifiable.
If you have a regular tutor and replace it with an AI, how many times do you think you'd accidentally meet your AI in town getting coffee and have a nice chat?
> If you have a regular tutor and replace it with an AI, how many times do you think you'd accidentally meet your AI in town getting coffee and have a nice chat?
You will never get the chance to sit and talk with your now-elderly tutor, in the future, and tell them how much they helped you, or pay their deeper contribution forward by tutoring someone else.
Your AI tutor also doesn't know a job you should apply for, or a local group you could join for your other hobby they now know a lot about, or… or…
I mean the list of intangibles goes on and on, doesn't it? But here at HN people like to disrupt.
AI may be good for programming and basic facts, but many useful things are hidden behind many layers of slop that you might not know you need to wade through if you are ignorant on the actual topic. It will easily give you Topic 101 information and facts, it will start using Topic 201 material if you ask questions, it will use Topic 301 material mostly only when you use the 301 keywords to get the AI on that level. Beyond that, even if it has the information, good luck actually using it or finding it without already knowing about it and how to reference it. And 100-200 level topic information is just enough to make someone think they get it but then fuck shit up and not understand why.
If you just hand off learning the lower level knowledge to AI, once people get to needing higher levels they will get lost.
>'Sure AI won't perform as well as our human tutors with all the support behind them, but it's also $30 a month.'
For how long, though.
>Mr Buckland also claimed the tutoring industry doesn't 'actually do anything unique' but instead provides comfort to students - something AI can replicate.
Jesus. That's really not a good thing.
It's not every day you see someone so incredibly dismissive of the product they're selling themselves.
> It's not every day you see someone so incredibly dismissive of the product they're selling themselves.
I believe that considering that AI is much cheaper than their service offering, and thus customers of Dymocks Tutoring and Talent 100 often ask them how much more they are actually offering for the much more expensive price has put this company under a lot of price pressure from customers.
So, I bet that Dymocks Tutoring and Talent 100 wants to change their main service offering to target parents from rich families instead who aren't so stingy.
Students need to understand what AI is and is not.
Something got lost as we transitioned from parents telling their kids to 'go play and do not come back until dinner is ready' to us handholding & driving our kids from one social/sport activity to the next.
While it causes fights with my kids, I blank refuse to do certain things for them (fix their internet connection, repair a toy or bike) I'll give them the tools and they know to ask when they get stuck.
The act of learning through failure is a dying art these days. It's 'teach to the test', stressed out teachers, parents having too much to do compounded by taking on second, third jobs in this economy.
Teaching our kids how to leverage AI is something parents should invest in. Its part of a broader education landscape.
> Teaching our kids how to leverage AI is something parents should invest in. Its part of a broader education landscape.
The same was said for laptops and tablets. Now there is a push to remove them from the curriculum. Having them meant the students didn't retain anything because they were distracting. I feel using AI in place of pedagogy will have similar results. Why bother learning to retain any information when AI knows everything and can recall it in a split second.
This is a real question though. One I've been asking since the early 00s. Learning means something different than it did in decades previous. Merely remembering easy to look up facts is not the slam dunk it used to be.
Imagine trying to speak without having this base of knowledge. Even if possible, you'd never be able to use the language in a comparable fashion. And I think this generalizes to basically everything in every domain. If you want to get good at chess, you need to study a ridiculously massive amount of tactics/patterns. If you want to get good at basketball, you need to simple practice dribbling and shooting a ball into a hoop for many thousands of hours, and so on.
The whole point of the memorization is to make it feel like it's not memorized. Like when you see these words, or come up with your own - it requires absolutely zero effort. There's nothing magical there. It's the exact same way that a master chess player can easily play a competent game of chess with 1 minute on their clock or while blindfolded.
If, and as, we start doing away with the rote, I'd expect we're going to see worse and worse outcomes. Arguably, this is already happening as Western education is continually 'rethought' while China rockets ahead of everybody with an education system that looks very much like a more centralized version of the US education system of 70 years ago.
Even if AI is very powerful, it is very obvious that someone actually knowledgable in a given domain can leverage AI more usefully than someone without that base of knowledge. And to be actually knowledgeable, as opposed to vaguely familiar with terms, takes significant effort over time. Luckily we have tools like spaced repetition that make that effort easier.
The last thing I'd add is that it's useful to know lots of things in various domains. Much like an LLM, our brains draw connections between disparate things which we know well, in the process of diffuse mode thinking.
It had its place in the 80s when looking up involved at least finding the right book. But that time was already on the way out in the 00s. My impression was more that education likes to ask for facts because they are easy to test for.
What you really want to teach are concepts and broad strokes. Nobody needs to know on which weekday the Soviet Union collapsed, but an understanding of the causes of the collapse is useful. Similarly, you don't need to know the speed of light to more than one digit of precision, but knowing the role of the speed of light as the speed at which information can travel through space is useful
That kind of knowledge also continues being useful in the age of AI. It allows you to see the connections in the world, and know which questions you even have to ask
I never sat down to memorise trivia about the Roman calendar, but I know Julius Caesar was assassinated on the 15th, the ides of March, because of repeated exposure while thinking things through.
It seems like old assessment techniques could luck in to testing comprehension, because the people who understood things would also remember them. This devolved into the recall being seen as the goal, instead of as proxy measure of something harder to test.
Any "concept" is composed of interrelated facts, all the way down.
> This is a real question though.
Is it? Why bother to learn how to add 2 numbers when you can just use a calculator.
> Learning means something different than it did in decades previous.
Not really. This is just the acquisition of a new tool. It can possibly be incorporated into education like how calculators were used in the 00s/90s, but you must know how to do the work without the tool.
Would you really raise your kids to not know how to add 2 numbers on their own?
I wouldn't even raise my kids to not know how to wash their clothes without a washing machine.
Wouldn't want them to be useless on their own, to be dependent on tools to take care of them.
It's fine to use tools to augment what you're able to do, but it'd be a kind of failure of their elders / of their environment to not teach them how to live on their own, independent, or at least dependent on as few things as possible. Tools aren't necessarily always going to be available in their lives. Having options for their life's necessities is good.
It also helps to know when the job's done well. Just because clothes come out of the washing machine doesn't necessarily mean they're clean.
If we're talking:
> Why bother learning to retain any information when AI knows everything and can recall it in a split second.
Kind of scary to think of adults that need AI to do their thinking for them, with the adults not knowing how to verify anything and just trust the output.
The fun starts when you ask it more complicated questions like 'i have a device that communicates with microwaves and draws a maximum of 20mA at 12v, is it safe for use near humans?' It will look up microwave safety, ionizing and non-ionizing radiation, workplace exposure limits and eventually come up with 1/4W is already safe and the device likely is not a 100% efficient microwave transmitter.
For every fact you learn there is a necessary context and numerous exceptions when it's applied in other contexts. If you expect AI to spit all relevant facts at you every time you're clueless, you will have no idea when AI is wrong or lying. You will have to trust everything AI (or its owner) tells you.
It's not just math either. Try to be an electrician without memorizing the basic principles of electricity and how they interact. You're going to have a bad time.
By induction we can see how this scales upwards, basically infinitely, to higher and higher levels of thinking.
I mean yes, I have some "facts" in my memory because of repeated exposure. I know 2x2 and 5x5 but the whole table? Not a chance.
Of course, maybe I'm wrong. Maybe you DO need to use a calculator to quickly tell me what 8 * 6 is. But I would be highly surprised if you got a degree in math and still needed a calculator for that.
I of course know how to figure it out. Let's see my instinct might be to do 8x2 first because I do know most of the 2s. Then 16x3 I can do in a few steps of addition. 20+12+16 then 48
And yes then I checked on my calculator because I'm posting this online and don't want to look like an idiot.
> By induction...
Well played.
AI can really help the outcome part (by just doing it for you), so for better or worse I see that delinkage continuing.
People who bother to actually learn while using AI will have an edge on those shovel slop, but there’s clearly a difference from previous eras.
If 1 guy with $16 billion turns into 16 grandkids with (inflation adjusted) $1 billion endowments each, it sounds to me like the family fortune was preserved extremely well, the family is simply bigger.
I think it will actually be much, much worse.
You can argue the same about humans. Growing up, I definitely did encounter people who charged money for tutoring and never taught anything - they just gave the answers. Wasted money.
As for dealing with kids whose classmates' parents help them a bit too much... I honestly don't have a solution. From my experience growing up, beyond a certain amount of help, you definitely are hampering the kid if you overhelp them. I guess I was just fortunate that none of my peers' parents cared to, so I could still be near the top of the class.
Oh, and our final scores were heavily weighted on in-class tests/exams, so even if some kid's parents did overhelp, they wouldn't do well overall.
Young adults are less competent and resilient than ever for exactly this reason
Can you give us an example to what his looks like in your context (not necessarily you and your kids)?
> The same was said for laptops and tablets. Now there is a push to remove them from the curriculum. Having them meant the students didn't retain anything because they were distracting. I feel using AI in place of pedagogy will have similar results
Laptops, tablets, and AI are just tools. If the folks wielding these tools (or ostensibly teaching others to wield these tools) are not using a tool effectively, then we shouldn’t be surprised when the outcomes are unimpressive.
Simple example — put a hammer in the hands of a skilled builder, and you can get a deck or a house frame. Put the same hammer in my hands, and you will get nothing more than bent nails and sore thumbs.
All that said, just like I feel confident that I could learn how to use a hammer well if I took the time to learn and practice, I think students and teachers can learn to use their tools (laptops, tablets, AI, etc.) well if they take the time to learn and practice (ideally with some good scaffolding).
I struggle with this argument when the subject ("the folks") turns out to be "many people", perhaps even a majority of people. If one teacher is an outlier that uses it badly then sure, the teacher is to blame and be judged. But if a collective is using it badly, then perhaps the capability of the collective should have been evaluated before handing the tool to them.
I don't have an answer - it is bad to withhold a tool from the hands of people who will use it badly. But the overall outcome has to also count.
You won’t get any argument from me that most teachers are poorly trained. Money can largely buy someone out of that trap.
That said, just forbidding a tool that is critical in modern life just because some/many people are bad at using it does not strike me as a particularly constructive solution to the problem.
> I don't have an answer - it is bad to withhold a tool from the hands of people who will use it badly. But the overall outcome has to also count.
My “solution” to this is not a popular one. Lower expectations for public mass education.
Over the last 120 years, we have taken a curriculum that was targeted at elites and forced it onto almost everyone.
High quality education should be available and cheap or free to everyone, but I really don’t think it should be mandatory.
If we stopped compulsory education at 8th grade and replaced most high schools with some sort of other skill development program or work/study program, the expectations and outcomes would probably be better aligned.
Large parts of the tertiary education system would be gutted, but maybe that’s a good thing.
I’m embarrassed by some of my early work and simultaneously impressed it managed to work at all! And someone was dumb enough to pay me money to do it?!
Sounds just like most 'senior' devs tbh.
I remember growing up being told similar things about the internet and how you can't trust anything written on it. There was some truth to that, of course, but how many people will say that of search engines, despite them being filled with utter crap?
I use AI extensively now, it also cites the websites and electronic documents it uses. It’s a little frustrating in that it cites the URL but not the exact paragraph or chart cited. For that I have to read the entire article. Sometimes it gets it wrong because the context is incorrect. It’s still a great tool, but without my last generation experience in high school and undergrad research, someone without that experience would blindly follow wherever it takes you.
https://en.wikipedia.org/wiki/Problem-based_learning
In PBL the teacher adopts a constructivist approach - supporting, guiding, and monitoring the learning process as opposed to delivering a defined curriculum via lectures or seminars.
Agentic AI is arguably a better fit for this as it can scale infinitely to monitor student progress, offer timely feedback on design choices, and adjust difficulty dynamically based on how the student interacts with the problem.
Your main contention is one I'd broadly agree with - that use of Agentic AI for tutoring abdicates intellectual agency, allowing the AI to perform the critical synthesis and analysis that the student should be doing themselves for learning to occur.
The solution, however, is clear: guardrail the AI to promote core pedagogical principles rather than replace learner cognition by mandating things like instructional friction and the strategic fading of AI support as students build competence.
As explored in the paper I link below, by balancing proactive scaffolding with necessary and beneficial cognitive challenges, AI can preserve the productive struggle essential to effective learning.
https://arxiv.org/html/2606.04543v1
(A final thought: have a look at the current field of companies who specialise in turning training materials into engaging interactive experiences for the likes of Compliance and Security training. This is a corporate field now being automated and democratised for use by the average student in Academic pursuits in an on-demand fashion. Appropriate AI tutoring is basically the same concept.)
Few days ago my son was working on a problem which asked if square is a rectangle. Turns out his class teacher and external tutor (who was using AI) had different opinions on it. My wife consulted AI as well and she also came to the conclusion that square is not a rectangle cause AI told her so, AI told the same answer to external tutor. I understand that the context matters and if one probes AI more with additional context it may come up with correct answer, but this depends on the assumption that the person asking AI for something already knows the answer.
Replacing that with AI is the kind of SV-stupid which starts making sense after the educational system was hollowed out for profit reasons and is failing. The human in that system is just checking in and DGAF.
Another classic is “how many roots does a parabola with real coefficients have?” The answer is always two by the FTA, but the textbook Algebra 1 answer is zero, one, or two because they haven’t covered complex numbers yet.
I don't understand how your wife interrogated the AI to be told a square is not a rectangle.
And how can two tutors disagree on this?
In the world of object-oriented programming: https://en.wikipedia.org/wiki/Liskov_substitution_principle#...
With the amount of blatantly false information I've seen LLMs spew out in response to my very simple questions, I can very easily imagine one would get this basic information wrong somehow, although I did just Google it (as Gemini, and particularly the on-search-result-page version of it, seem to be the absolute worst of the common models in my experience), and it did get the answer right.
The obvious question of “is a square a rectangle?” is definitely NOT causing any LLM to say “no”.
Hallucinations are an issue but they genuinely are no longer a concern for any modern non-tiny LLM for basic facts that you would expect a K-6 course to cover. Maybe you might be able to find a few random hallucinations in knowledge covered at the high school level but I’m quite skeptical even that still happens. Yes, they can still, at times, hallucinate sources but I cannot recall the last time I actually saw this happen. I think it was at least a year ago
I'd guess that the tutor had a brain-fart, and then blamed AI for their own silliness? No matter what the reason, they clearly shouldn't be tutoring.
(I mean, seriously, what on earth?)
The problem with the tutor's question is that they're going to AI to answer this question and not asking underlying questions like "what makes a square a square" and "what makes a rectangle a rectangle", then doing their own thinking to determine if a square's qualifications also fit inside the qualifications to be a rectangle.
Heck, even if you just feed the underlying into AI it'll reason the answer in a logical manner:
> In standard geometry, A square is a quadrilateral with 4 equal sides and 4 right angles (90° each). A rectangle is a quadrilateral with 4 right angles. Its opposite sides are equal and parallel, but all four sides do not have to be equal. So the key distinction is: Rectangle: four right angles. Square: four right angles plus four equal sides. > > That means every square is a rectangle, but not every rectangle is a square.
Asking LLMs to just give you answers is the real antipattern. But if you remember that the only thing it's good at is finding patterns given context/constraints, it can be an extremely good tool.
Also, I have a feeling the vast majority of the populus is using free versions of Claude or ChatGPT. Thankfully recently ChatGPT Free is GPT-5.6 Luna, but before the response qualities and capabilities of the free chat output were less than stellar.
Hire more?
Right now teacher largely execute a curriculum this is predefined and do behavior management in US public schools at least. There is not a whole lot of creativity and a lot of rigid administration.
Given that teacher salaries are one of the biggest expenditures it makes sense to try to gain efficiencies while trying to improve results.
Under the current scheme, the results are headed downhill. More of the same won't solve any problems.
Do we really understand what AI is and is not? For example we all know AI can hallucinate, but how does that knowledge help us in practice? If I'm leaning something new I would have to fact check everything which would take about as much time as learning from verified sources.
People formed their opinions on LLMs in late 2022/early 2023 when ChatGPT was first released and taking the world by storm and then never updated them after nearly 4 years of improvements.
Hell, just being able to search the Internet did MASSIVE things for reducing hallucinations.
LLMs work on tokens. They literally cannot see individual letters. Yet, unlike humans, they can instantly create a tool to solve this task for them and never be stumped by it again.
I'm so tired of this brain-dead gotcha comment.
Do you think your legislators or most users understand tokens and predictions and what is going on? Do you think that those who bombed a school using it know this?
Until frontier models can account for the disparity between what they can do and what the person (or other agent) is asking, we will continue to suffer bad information used like good information.
That tokens cannot count letters like that is not a brain-dead gotcha. Ignoring how humans are using these models is lacking in critical understanding of the real world. People are literally going to die due to humans not understanding how this works (and already have). True brain-dead, and body-dead.
Then there's recess and afternoon is all taken up with "Life Skills" classes such as entrepreneurship and the likes.
Kids are tested weekly and progress through grades at will. For example, his kid would officially be in 3rd grade, but she's doing 3rd reading, 5th grade math, and 4th grade writing. She can easily test 3rd grade topics now.
I am ambiguous about this model. Love the "learn as much as you want" even if it means learning 2, 3 grades ahead, but the no-teacher rule and so much computer use seems a little strange. Also if the kid is not extremely self-motivated, they will be left by the side of the road by the school.
[1] - https://alpha.school/
Edit: It's 40k a year.
By the time they get out of primary school most have had the curiosity that drives motivation beat out of them.
I am not sure where I would have ended up if left to my own devices, e.g. as a kid starting first class I was bad at languages and hated it with a passion, in the last class I had the single best German and English final exams in my school district and language had grown on me.
It is a for-profit product which is sold to parents and state-level customers more than it is a school.
There have been many variations on this idea of "letting kids learn on their own". The basic issue is the one you highlight, which is that it works if the student is very self motivated. Most students aren't. Or rather, they're highly motivated to do other things.
I think if you can get into the chemistry just a bit, then the math behind the chemistry, it can be a vehicle. I mean it's the same with programming. Why write code (boring, yuck) when you could play video games? Then the kids see code makes video games. Suddenly you have all these programs with kids making Minecraft mods or whatever. But finding each child's "bomb" would be hard for a teacher to do, especially with a common curriculum
In the early days they were using off the shelf curriculum software and layering their own steering AI over it.
If you believe, as I do, that half of students only achieve half of their academic potential, then experiments like this are some of the most valuable work we could do in society.
But again, let me know how it goes in 25 years and my children can enroll my grandkids.
However, I would fear that the issue with home-schooling would manifest: that the rarified atmosphere would render them incapable of engaging or coping with the brutal reality of binary-minded Authority figures and hierarchical social constructs in their work and life in the 'real world'.
To writ, some of the most important lessons I learnt in school were to do with the concept of 'fairness', my own ability for emotional regulation, navigation of social hierarchies, and the symbols and signifiers of popular culture. The curriculum as a whole came a distant second.
N.B. Public (read:Private) schools in the UK are about £30k a year. Boarding schools hit about 50k in the UK/Ireland. 'New Age' or International schools for the children of diplomats and HNW ex-pats like Nord Anglia or similar go from about €/£20k upwards. 40k a year isn't outrageous in context, even outside of America.
This keeps coming up every time, and it's always pointed out that there is scant evidence supporting this view at a large scale.
Every motivated parent I know who homeschooled his kid didn't deal with these problems.
And plenty of those who went to a traditional school did. They aren't even outliers. I'm sure anyone who went to a regular school can think of kids in their class who exhibited the problems you mentioned.
No doubt, plenty of homeschooled kids do as well. Just as you have a distribution for regular schooled kids, you'll get a distribution for homeschooled kids. Claims that the distributions for the latter are skewed in the "wrong" direction are never presented with evidence.
> To writ, some of the most important lessons I learnt in school were to do with the concept of 'fairness', my own ability for emotional regulation, navigation of social hierarchies, and the symbols and signifiers of popular culture. The curriculum as a whole came a distant second.
I went to a regular school, and none of what you say applied to me. The curriculum/instruction came first, and learned little of the former. I'm sure I'll find classmates who'll say the opposite. This sounds like a classic case of post-hoc attributions.
Home school is best, private school if you can, public school if you must.
The lesson you should have learned would be that we should tear down the "brutal reality of binary-minded Authority figures and hierarchical social constructs." We only tolerate these things as "normal" because people go through their lives not actively trying to tear them down and break them.
> The lesson you should have learned would be that we should tear down the "brutal reality of binary-minded Authority figures and hierarchical social constructs." We only tolerate these things as "normal" because people go through their lives not actively trying to tear them down and break them.
I was a product of an environment like that (magnet school with a lot of unstructured time and good access to resources), and it didn’t take me very long to run into the Binary Authoritarian Figures’ version of “you and what army?”, and they were largely correct.
There are ways to undermine parts of the establishment, sometimes even at scale, but the process is not always obvious or easy to access.
That’s a very uncharitable reading.
I have made my mark in the world by exploiting the Achilles’ heels of “authority figures” (which is usually just some form of a lack of empathy). I have caused a few to fall, I’ve made others reform, and I’m just an annoyance to many others. I have trained others to do the same.
That said, I could get whacked out of existence figuratively or literally if I crossed the wrong person or group.
There are others like me, but (as I mentioned before) we aren’t always easy to find.
I know the above may seem grandiose, but I’ve had a chip on my shoulder about these “authority figures” since I was a kid. I’ve done something about it, but I also didn’t choose to bang my head against a wall and call it progress.
Uncharitable, maybe. But, yeah, you should definitely stand up for yourself more.
1. You may want to work on your reading comprehension skills.
2. You would probably benefit from being more curious and less certain.
Making yourself feel better about giving up does not constitute actually doing anything.
Having been to a school like that, I agree with you. I had to learn those lessons quickly.
I largely opted out of being in the middle of that world, but I am still adjacent to it due to its ubiquitousness.
Most kids are motivated given the right environment. Enjoying learning is natural.
The AI element of what you describe does not appeal to me. The weekly testing and assignment to grades sounds a bit rigid too.
Considering how much one to one human tutoring 40k an year could buy that does not sound like good value.
Home schooling sounds good in paper, but as much as school can suck, it is also an irreplaceable opportunity to learn to socialize, which is a much more valluable skill than knowing pre-university curriculum.
My schooling was alright but I can imagine a much better version given different circumstances.
(although, to be fair, it was a different time. AI would have surely made it much better)
If you’re the sort of kid who gets everything right first go then you join the “rocket ship” and spend 2hrs a day with AI then play for the rest of the day. You would do well in any school.
If you don’t get it right quickly then you join the “Pirate Ship” and have to spend all day grinding and obeying AI while your friends play. It seems pretty terrible.
https://buildcognitiveresonance.substack.com/p/fourth-report...
socioeconomic status (SES) is very often a huge confounder when it comes to anything from health issues, educational outcomes, and many other things. rich kids do better - the SES of the parents of Nobel Prize winners [0], young athletes [1], etc are all indicators of this
[0] https://www.chemistryworld.com/news/nobel-prize-winners-far-...
[1] https://www.sciencedirect.com/science/article/abs/pii/S14668...
I think the most likely outcome is AI makes American education worse, not because if the technology, but because of the “leaders” building and implementimg it.
It doesn't mandate curriculums, choose technology, anything like that? Only recommendations.
They might have been able to monitor its effects but they wouldn't have been able to direct its use or non-use; that's not what they did.
Essentially it was/is an anti-discrimination organisation, making sure that states gave an education regardless of wealth/race/status.
If you don't understand, here's Trump's secretary of education explaining how it'll make our children excel: https://www.youtube.com/shorts/xsRUk0dMJu4
You have to tell it what you want out of the session. A kid could just type in the question and get the answer, but that’s not tutoring.
When I use them for learning I tell it my goal, summarize my current status, and then tell it to teach me incrementally with a quiz after each step. You can get fancy and have it reach back into history to throw in extra refresher questions.
It uses very few tokens compared to a coding session so you can use the highest model for this and the usage is negligible. Using the top model available to you is important to keep things accurate. I wouldn’t have trusted this back in the GPT 4 or Sonnet days, but with Opus, Astra, or Fable it’s great.
For longer sessions set up a directory and have it keep notes about your progress, problem points, and what was covered in the past.
The sad reality is that many of the most senior professors and lecturers in top universities are their for their research rather than teaching ability, and that the success of the elite institutions is somewhat self-selecting as regards their cachet and filtering done by their admissions process (i.e. Ivy League institutions do not offer athletic scholarships).
So that's about 0.75%.
The rest were either competent, good or amazing.
(Then again, one of the remainder was internationally famous for being a (self-acknowledged) media whore and while I know he added value it wasn't always to the subject at hand)
School is now distant enough for me that I have reached the brain fog age where I cannot be commenting on internet things intermittently while writing documents…
Learning new topics or skills is a very good example, I had the same experience as you with Claude when I learned a new language.
I'm not a huge fan of prompt engineering or elaborate skills that are longer than a couple sentences. But you have to give the AI context. The more context the better
If a random person comes up and asks you a malformed question how do you answer? What's worse if what if the question is not actually malformed in some contexts and because you're a wizard that has massive amount of contextual information there are plenty of right answers for the 'wrong' question you just asked.
In schools classrooms tend to avoid unlimited exploration as it would take unlimited time. On your AI prompt the next thing you ask could be anything, anything at all. Just exactly how is the AI supposed to know if you're an idiot?
"Computers are useless. They can only give you answers." (he was not joking by the way)
Of course, if computers DO threaten to take over literally everything, then people are even more scared and react aggressively, as demonstrated in this 1939 article describing IBM's vocoders:
"The Pedro Peril" (the first AI that actually talked back was nicknamed Pedro, and described as "the latest and most fearful gadget from IBM", the journalist's conclusion was to suggest Pedro be deported as he would spread propaganda and maybe even become a dictator) (unsigned Washington Post editorial, 1939)
Here's Pedro in action: https://www.youtube.com/watch?v=0rAyrmm7vv0
I find it very easy to engage in a Socratic dialog with the AI. If I ask it, it will actually quiz me and carry on a tutoring session. And it doesn't mind if I ask dumb questions.
I've walked away from colleagues trying to explain things similarly because if you're not getting it after a few attempts you're going to hit a wall of their ability to explain, their patience, time, etc
I used to be a tutor, and the kids who wanted to get ahead were people who more needed someone to give them hints as they learned. The ones who needed to get the grade were a lot more reticent and needed a lot more attention and prodding.
I don't think AI as it currently exists is a good choice for those folks.
If there’s proper oversight (tutor teaches the student and reviews knowledge while the student uses the AI to study or test themselves) that’s fine.
If it’s just the AI then a student isn’t going to know what they don’t even know to ask.
This can also work with books or adequate material someone is basing their learning on.
When I first started calculus, I was furious at being taught Newtonian Physics and some bits of chemistry without calculus. It's a stupid bunch of equations you have to learn, which Newton worked out from calculus. I'm glad people are getting it more often together now, but also just a bit jealous at the same time.
Similarly Youtube didn't replace teachers - it afforded remote learning and correspondence courses a new medium for transmission, recording, and playback of teaching.
More to the point none of those were trained on a corpus containing the history of human pedagogy, were capable of inference, or of generating dynamic learning resources based on best-practice models with inherent self-correction and pedagogical reinforcement mechanisms.
So usually the claims about replacing tutors like in this article come from a place where they usually give a caveat that maybe for highly motivated and ambitious students, they can figure out how to accelerate their learning with AI. But then that undermines the whole argument of transforming education if it wouldn't work with students who are average or below-average in their academic achievement.
Not really. I learned far more from YouTube in subjects and topics that I was deeply interested in. School or college teachers didn't go that deep.
Whether it's pc or not, there are there are bad teachers. I skipped those classes in college and learned from MIT OCW and random tutorials.
So, YT did replace some teachers for me, and it continues to do so for many people.
All failed and all useful. This time ...
Tutoring is a process, and the end result is that the student gains a demonstrable new capability. The how isn't as important as the end result. If we end up with a bunch of people with increased personal capabilities, it's a win. If we're lazy and we end up with a bunch of people leaning on crutches, we failed.
Time will tell, I guess. Could be a surprise answer to Bloom's 2 sigma problem or a very stunted generation, or a mix of both.
I disagree. If, after you have been tutored, you perform well, but still need tutoring for the next year's exam in that subject, then you haven't been "tutored", you've been given a similar enough copy of the exam questions to train on.
As I keep telling my kids, good results is a side-effect of good habits. If you're half-assing things, it shows up in your results. If you studious and methodical, it also shows up in your results, just in a different direction.
AI is used primarily to half-ass things, do things without checking, getting good vibes by that sycophantic tone.
Get your kids into a rut of good habits, and they'll run in that groove forever, regardless of tutors or AI. Half-ass your way through life, then sure, they'll get fucked over by AI tutors too.
One huge issue IMO is that right now, around the world, most people are not taught at all how to learn for themselves nor how to critically think on their own. And that will indeed result in using it as a crutch and not a good tutoring tool.
I don't think so. In fact, I'll go one further: I don't think we're in disagreement, just that I've been communicating poorly.
Allow me retry: My point is NOT "AI makes tutoring worse". My point is that AI tutoring is irrelevant to sustained academic performance.
1. Academic success is a side-effect of good habits (high correlation, not all the way to `1`).
2. A good tutor will instill the good habits for a specific subject (IME, different subjects require different habits, but all require the student to be diligent and consistent at least)
3. Half-assing things, as a habit, will not materially improve outcomes.
4. Most AI use is for half-assing things.
I did not mean to imply that AI in tutoring cannot be successful, only that AI-use tends to result in half-assing things.
Your anecdote is almost the opposite of half-assing things.
> One huge issue IMO is that right now, around the world, most people are not taught at all how to learn for themselves nor how to critically think on their own.
That's a different problem, I think. As I am learning with my two boys, you don't need to teach critical thinking specifically. Critical thinking naturally arises out of being diligent and thorough.
I have experimented a bit with AI and have found that AI isn't always great at catering precisely to the syllabus/marking criteria of a specific exam boards even when you give them all the docs publicly available (although I haven't tried frontier models). Somebody who has worked as an examiner for the exam board knows the tricks for how to structure longer answers to both get the marks and to not waste time in the exam. Of course, it may be that I haven't prompted well enough, but if I can't manage to do that, then I probably can't expect my children to!
It has to be a second step, a critic. If it's the first step you don't end up with any way to know when it's being crazy.
The tutor from Cameroon is much better.
She tells me when I’m getting the basics wrong and designs lessons to go back and fix the basics.
AI loves to tell me how great I am and do lessons at the frontier of what I understand to make me feel good about how much progress I’m making.
:-)
(to make it clear: this was a joke, even though I'd actually be interested whether this works)
AI tutors will absolutely be worse, not better, but they may also be only a fraction of the price.
Because what is required to be able to know when an LLM is hallucinating? Expertise.
Meaning: exactly the one thing students don't have. By definition.
Remind me - why did ChatGPT get rid of study mode?
The people who need tutors are those who are not self motivated. AI doesn’t solve that problem.
It's great at giving you the illusion that you're learning though.
If you know an outline of the material and want to structure it your own way (ex, explain attention to me using Haskell notation) it's really great. But otherwise, human notes win.
learning is not a passive activity. likewise an llm can be used in a passive or active fashion.
Humans have LOOONG contexts
- LLMs cannot motivate a person to learn because they aren't people
- Learning requires rewiring your brain; you need to try and read content without explanations of every question you might have. Struggling is required for learning
- LLMs cannot generate structured content on long scales. I have tried this many times. Your average TA can easily write a calculus textbook.
- LLMs are RLHFd to just agree with you. Learning requires disagreement and tutoring requires assuredness
-the thing that motivates you doesn't have to be the same thing that teaches you (e.g. a book can be a useful learning tool even if it isn't human)
-you can absolutely choose to interact with an LLM in ways that generate productive struggle. If you naively ask it for answers it won't be much use in this regard. But that's just a poor choice of how to use the tool. Just like hitting yourself on the head with a book doesn't mean the problem is with the book as learning tool.
-You don't need structured content 'on long scales' to learn. You learn a little bit at a time. Long textbooks are a side effect of the economics of moving paper around. Not a prerequisite for learning. An LLM will happily mimic the table of contents of a textbook or an outline for course. And you can use future prompts to generate the bits you need as you need them. One-shoting a textbook is a silly use of the tool. Plus, we're not short on textbooks. We're short on ways students can get customized help: answers to the specific ill-posed questions they have in their heads. LLM's are quite good at this. You can have textbooks AND LLM's (and professors)
-LLM's are happy to disagree with you. Just ask them to and they oblige. Ask them to respond as if they were a tutor and they will mimic the assuredness if appropriate.
- Students (Humans) aren't going to struggle on average. They're always going to take the path of least resistance. I know of some successful algebraic geometers who did indeed hit their head on heartshorne in grad school.
- a) Structured content generated on long scales is way more effective because you can tie together many details. It would be weird to learn from chapters, each unrelated to each other in style and content.
- b) We're actually not short on students getting help. Office hours were abound at my university. I actually found that customized help was detrimental; if a TA answered every question a student had, the student wouldn't put the effort into trying to learn the material. Imo you want to see students struggle. I've seen students go from F to A- by just forcing them to struggle on the content.
- Not really. I discuss complicated math and programming with models regularly and when I say something like "actually you're wrong" they say "yes, you're right". If I ask them to be confident, they will not change their wrong answers. Humans have a much better conception of self and opinions than LLMs
If you're giving it prompts like that, you've shot yourself in the foot.
Ah, now that makes sense.
It is shocking how quickly many people are to dismiss the benefits of human interaction and connection since they aren't quantifiable.
If you have a regular tutor and replace it with an AI, how many times do you think you'd accidentally meet your AI in town getting coffee and have a nice chat?
You will never get the chance to sit and talk with your now-elderly tutor, in the future, and tell them how much they helped you, or pay their deeper contribution forward by tutoring someone else.
Your AI tutor also doesn't know a job you should apply for, or a local group you could join for your other hobby they now know a lot about, or… or…
I mean the list of intangibles goes on and on, doesn't it? But here at HN people like to disrupt.
If you just hand off learning the lower level knowledge to AI, once people get to needing higher levels they will get lost.
For how long, though.
>Mr Buckland also claimed the tutoring industry doesn't 'actually do anything unique' but instead provides comfort to students - something AI can replicate.
Jesus. That's really not a good thing.
It's not every day you see someone so incredibly dismissive of the product they're selling themselves.
I believe that considering that AI is much cheaper than their service offering, and thus customers of Dymocks Tutoring and Talent 100 often ask them how much more they are actually offering for the much more expensive price has put this company under a lot of price pressure from customers.
So, I bet that Dymocks Tutoring and Talent 100 wants to change their main service offering to target parents from rich families instead who aren't so stingy.