And for the good reason. I have yet to see 1 (one) reason how giving infinite cheat machine to students is of any benefit to them.
The secret to mastery remained unchanged for a thousand years: books, lectures and homework. All of this can be done with pen and paper level of technology.
It does scale even better than anything else. Majority of students are just not very good at it. And proving that they have learned sufficiently is hard. Thus schools and universities.
I sense a certain contradiction? If majority of students are just not very good at self-studying methods, then they don't scale well.
Mediocre student that sits through lectures, does the homework and submits assignment on due dates will get a good level of mastery in the subject. The very same student would be totally lost relying only on self-studying - he would not even know what he doesn't know.
It is in sense contradiction. Schools and universities are greatly limited by facilities and teachers. Self studying is not. Thus with modern technology or even with just mass printin you could theoretically educate as many students as you want. In that sense I do say it scales much better.
I fully admit that for efficient time spend and production of sufficiently educated students it is not effective method.
A home can also be built of mud bricks and heated by dung.
The issue you’re describing is a mindset/cultural one, not a technological one.
The same cheating machine can be used as a tutor. Those same cheating students understand, that at the gym, they themselves have to lift the weights to get strong, but that mindset doesn’t carry over to their classes.
LLMs are like anabolic steroids for the brain. Surely you can do more with them, but you are better without them. Access to them should be restricted, especially for untrained minds.
As all the cheating comments are referencing, the students don't necessarily want to get strong (that is a lot of work), but to get the qualification which unlocks the good jobs.
We haven't yet got to the downstream question of what white collar jobs are going to look like in the AI era.
To go with the gym analogy, thus far it would seem that using AI to help study seems to be working out about as well as the Enhanced Games did.
Before we get ahead of ourselves and hype up AI as a tool for making people smarter, let’s first demonstrate that AI really does help them learn better. Otherwise we risk wasting a lot of time and money on silicon snake oil.
I don’t think the comparison is fair, consider the incentives for the Enhanced Games. Once an athletic declared their taking banned substances, they are barred from every other competition, so it’s no surprise that “clean” athletes dominate.
As for proof, I think it’s self evident for anyone who has tried to us it in that way.
If you’re studying math ask it to generate more questions for you to try. If you don’t understand a concept, have it use the Socratic method on you. Learning a language, talk back and forth with it over voice. Writing an essay, have it mark and proofread it and provide feedback based on known good criteria.
This is the most profound comment I’ve ever read on this website. How do we do push the culture in a direction such that productive intellectual pursuits will actually be intrinsically desirable in the AI era?
The definition of mastery in your comment sounds like nonreliance on technology to accomplish [something]. Technologyless Mastery was a natural result from: reliance and overreliance on technology being stuttery, unpredictable, inadventitious and costly.
In other words, it made more sense to 'frontload' hardthought and pondering so one was prepared [for whatever may come].
People also thought aqueducts were the best way to move water for thousands of years.
I can teach symbolic math effectively using clay tablet and reed pen today.
Numerical methods, however, would require at least minimal computing hardware.
I think there should be a guiding principle of using the least amount of tech in teaching any subject, just like principle of least privilege in cybersecurity.
You hit on the head something that I've been trying to put into words for a while now: it seems that the benefit of technology to increase learning appears to be a bell-shaped graph where for each combination of subject matter there is a sweet spot where the level of technological assistance becomes helpful but after which it replaces too much of your own internal neuron-connecting processes and inhibits learning.
That's a strong observation. And I'm wondering if humans will adapt; or if the intensity of "heat and light" required to "weld" connections together is non-negotiable.
> People also thought aqueducts were the best way to move water for thousands of years.
They still do because they still are? I'm hard pressed to think of a better way to move water than creating a gently sloped channel(this includes any pipelines that don't use pumps, by the way) for it to follow from where it is to where you need it. Many places that don't have sufficient local groundwater are still supplied by aqueducts to this day. If you have to pump it, then you're paying for the channel and the energy, which is worse than just using gravity.
It is causing far more pain than benefit in the majority of institutions.
Particularly the cheating. Many students are flagrantly using it + the detection tools either aren't configured to check for it, or, if they do have something like TurnItIn, it doesn't work, and flags legitimately created content sans AI.
I think what we are seeing with all these AI detection tools is a whole lot of shooting the messenger. Technology made the problem of lazy students impossible to ignore, but the solution might be decidedly low-tech: have students spend some time talking about their work with the teacher, face to face.
For people who teach through writing, there are great tools out there like MATCHA: https://matchawriter.com. It's a secure writing environment that verifies authorship by recording the full writing process and, when needed, can restrict copy-paste, automation, and access to other applications.
In my opinion, most of Academia's pains are wholly self-inflicted.
The primary use case of universities are to produce a generation that can contribute to the society in a meaningful and positive manner. Nowadays in-order to do that, AI literacy is a must for graduates just like it was 15 years ago when it was a must to have computer literacy.
AI is a tool like internet was and it's the responsibility of the curriculum to teach proper usage of that tool rather than fighting against it.
We still teach primary school children how to do addition, despite the fact that during their working life they'll have access to a calculator virtually all of the time. A decade ago posting your CS assignment on StackOverflow and handing in the top answer was considered cheating, despite professional software developers doing pretty much that any time they ran into something unusually tricky. Why teach students differentiation when they can just put "derivative of x^2" in WolframAlpha?
Universities are supposed to primarily teach understanding, not basic tool use. Yes, tool use should be part of your education as well, but that's only a small fraction of it. A single "prompting AI to do X" course is probably more than enough. But if you completely skip the fundamentals and make the students fully dependent on the tools, how could they possibly contribute to building the next generation of tools, verify the output of the tools, or transcend the tools?
Yeah, they should indeed be teaching proper AI use as well, but in practice the easy access to AI is making it virtually impossible to teach the fundamentals, and that's the real problem here.
The analogies to prior technologies aren't insightful. The advent of LLMs has some similarities to the appearance of the internet, and very significant differences. Calling it a "tool" is also far too reductive. What kind of tool that existed before just writes everything for you? Is that the same kind of tool as a nailgun or a text editor? The "proper usage" that has to be learned is to learn to think for yourself before you get used to the word-bot doing all the thinking for you.
Yep. That's the premise of this article. The writer used to be the dean of humanities at University of Utah (great school, much of AI exists thanks to their work on CompGraphics, HPC, and Teapots) and is basically arguing that humanites and social science (HSS) programs need to build that kind of literacy.
A generation ago you saw similar resistance to the internet from HSS programs.
Lots of people talk about cheating here. That is not the framing I put in my mind. Or rather, if the students are cheating anyone it's themselves.
LLMs exist and are extremely useful in academia. I am currently riding a productivity wave, where I can spit out formalised proofs – not quite at the pace at which I can think of the theorems to prove, but almost. The thing braking my progress is that I have to build all the existing theory too, not just my own new stuff.
But for students, who do not yet know anything, it is a danger. From observing my students, it seems to take a certain amount of willpower to do the exercises without letting ChatGPT give them (or "explain them") the answer. And reading an explanation is not the same thing as working it out yourself. You just don't learn doing it, by not doing it!
Some students have this will power, some students don't. And it is also situational. Some students have stuff going (work or other factors) which makes it really hard too.
Yes, they can of course keep asking ChatGPT later in life, but they will not have the fundamental understanding which is required to even use LLMs for the advanced stuff. ChatGPT cannot explain you advanced concepts if you don't understand the fundamentals!
So, to test the students' understanding, we are back to having on-site exams, with the old-school supervision. And more students will fail, because there is now a way to solve exercises without learning.
I'm not sure what to make of this? Is any healthy skepticism about AI now treated as motivated luddism?
Academics has its many issues but universities are hardly the only sources of criticism about AI, and for all its benefits AI is poorly understood and often applied in relatively half-assed ways, even when it's well-intended. I would worry if thoughtful individuals weren't skeptical at all about AI.
It also seems reasonable to, say, want to evaluate students on their own merits free of the influence of a "homework machine" (to borrow a phrase from Shel Silverstein).
I'm reluctant to make predictions, and wouldn't describe myself as "anti-AI" but this sort of piece reminds me uncannily of the dot-com bubble.
This author has a ton of axes to grind and a poor grasp of computing history or how industry works. Pretending like it’s strange or even bad that industrial datacenter builders with tens of billions of leveraged capital can outbuild the GPU capacity of chronically underfunded grant dependent universities whose mission is not to create industrial scale compute resources is like complaining local technical school auto-mechanic program isn’t out-producing Ford’s auto manufacturing numbers.
But then as soon as the author rolled out complaints about DEI and ethics, you knew what was going on. Okay, please leave higher ed and work for the industry you love. As you say, they are doing just fine.
> The elite research layer of higher education (Stanford, Berkeley, MIT) is still an AI accelerant. But across the country, as Scott Latham told me, “the vast majority of institutions are in a state of AI paralysis.”
Tbf, the vast majority of institutions also suck.
State flagships and some of their second rung, a couple ivies and ivy tier privates, and military academies provide the bulk of talent that you need. And these universities have dedicated pipelines for people who start off at community colleges.
These universities are also much larger - UC Berkeley's and UCLA's 2025 undergrad population is almost double it's 1960 population [0].
And those who had skill or educational deficiencies can remediate them via non-trad or working professionals masters like GT, UT Austin, and CU Boulder's OMSCS for less than $15k.
The top 20 CS/EE/CE/ECE/EE degree programs (MIT, CMU, Stanford, Cal, UIUC, GT, UT Austin, Purdue, UW, UCLA, UCSD, Ivies, UChicago, NYU, UNC Chapel Hill, Duke, USC, UMinn, UMich) graduate around 15k-20k undergrads a year.
Throw in the next rung of well respected universities (eg. the rest of the state flagships like ASU or IU, the respected regional public programs like CalPoly, NCSU, or SUNY Buffalo, and the handful of regionally respected privates like WPI, Cincy, or both Rochesters) and that number likely doubles.
If you're graduating almost 50k grads a year from target programs, there just isn't enough room for the rest.
Choosing to go to (randomly choosing two universities to pick on) USF or TCU for CS straight out of HS instead of going to your local CC and then transferring to UC Davis or UT Dallas is dumb.
In 1980 you had a little under 2,000 4-year colleges in the US. In 2010 you had almost 3,000 4-year colleges after which the number of institutions started declining. Making more colleges while existing colleges were expanding headcount did nothing to help.
The commercial product is the paper that says "University of Whatever says you learned Whatever". If the university loses credibility (by producing a bunch of people who cannot function without a chatbot) then the product is no longer valuable.
Luckily for us all these good natured AI people will liberate us from the necessity of our own value and places like universities, which give us structured environments to participate in intellectual pursuits, will be in high demand.
This assumes every degree is valuable in the marketplace, which is simply not the case. The product is luxury attendance, the degree is secondary to that.
The issue is more one of current passive, abstract forms of assessment. This is laziness on the part of instructors. They will simply have to move to more complex, project-based assessments and accompanying interviews, which is more time consuming but will be far more accurate than the more traditional system of assessment was anyway. On sum it will be better for everyone.
AI is totally going to replace the university system. This is like the coal and oil guys sabotaging EVs and renewable energy. Good for them, bad for us.
Depending on the type of question being asked and the complexity, human intervention is still needed because LLMs still face an information retrieval problems.
And for the good reason. I have yet to see 1 (one) reason how giving infinite cheat machine to students is of any benefit to them.
The secret to mastery remained unchanged for a thousand years: books, lectures and homework. All of this can be done with pen and paper level of technology.
Zooming out you see that books, pen and paper are all technology. Like even school and language is technology.
Given that AI exist maybe we can roll back some of the school/university and just let people do the work and learn while doing?
Yes not for a 7yo but with 18 you could just work and learn what is required or what you find exciting.
Self-directed studying doesn't scale. To produce thousands of new graduates per year you need industrial-scale methods - schools and colleges.
It does scale even better than anything else. Majority of students are just not very good at it. And proving that they have learned sufficiently is hard. Thus schools and universities.
I sense a certain contradiction? If majority of students are just not very good at self-studying methods, then they don't scale well.
Mediocre student that sits through lectures, does the homework and submits assignment on due dates will get a good level of mastery in the subject. The very same student would be totally lost relying only on self-studying - he would not even know what he doesn't know.
It is in sense contradiction. Schools and universities are greatly limited by facilities and teachers. Self studying is not. Thus with modern technology or even with just mass printin you could theoretically educate as many students as you want. In that sense I do say it scales much better.
I fully admit that for efficient time spend and production of sufficiently educated students it is not effective method.
> with pen and paper level of technology
definitely try out:
galaxy s9 ultra tab (wacom! edit: also, 14.6 inch: bigger than any ipad series)
+ LAMY s-pen
+ paper-feel screen protector
(and optionally felt-tip nibs)
miles better than MPP or AES or other crap: those can't draw diagonal lines without wobbling, thought I was getting parkinsons
I had iPads and Apple Pencils - but nothing beats the pleasure of ink pen on paper.
A home can also be built of mud bricks and heated by dung.
The issue you’re describing is a mindset/cultural one, not a technological one.
The same cheating machine can be used as a tutor. Those same cheating students understand, that at the gym, they themselves have to lift the weights to get strong, but that mindset doesn’t carry over to their classes.
LLMs are like anabolic steroids for the brain. Surely you can do more with them, but you are better without them. Access to them should be restricted, especially for untrained minds.
That’s a bad analogy, you still have to put all the effort yourself at the gym while on steroids.
Perhaps LLMs are a forklift.
Ha! Steroids give you muscle mass without lifting, believe it or not. https://www.nejm.org/doi/full/10.1056/NEJM199607043350101
Of course, side effects are a different story.
Not true actually
As all the cheating comments are referencing, the students don't necessarily want to get strong (that is a lot of work), but to get the qualification which unlocks the good jobs.
We haven't yet got to the downstream question of what white collar jobs are going to look like in the AI era.
To go with the gym analogy, thus far it would seem that using AI to help study seems to be working out about as well as the Enhanced Games did.
Before we get ahead of ourselves and hype up AI as a tool for making people smarter, let’s first demonstrate that AI really does help them learn better. Otherwise we risk wasting a lot of time and money on silicon snake oil.
I don’t think the comparison is fair, consider the incentives for the Enhanced Games. Once an athletic declared their taking banned substances, they are barred from every other competition, so it’s no surprise that “clean” athletes dominate.
As for proof, I think it’s self evident for anyone who has tried to us it in that way.
If you’re studying math ask it to generate more questions for you to try. If you don’t understand a concept, have it use the Socratic method on you. Learning a language, talk back and forth with it over voice. Writing an essay, have it mark and proofread it and provide feedback based on known good criteria.
This is the most profound comment I’ve ever read on this website. How do we do push the culture in a direction such that productive intellectual pursuits will actually be intrinsically desirable in the AI era?
> "It's easier to get and raise a child
> Than to put character in him. No one
> Has ever found a way to make a fool
> Wise or a bad man good. If god had taught
> This to the children of Asclepius—
> How to cure badness or a muddled brain—
> What handsome fees they'd earn!"
That was Theogenis from the 6th century BC. I don’t think we’re any closer today than they were back then.
Alpha School is a decent counterpoint to this.
Of course, it's a highly selected group of students. I very much doubt it's scalable in the way one would hope.
The definition of mastery in your comment sounds like nonreliance on technology to accomplish [something]. Technologyless Mastery was a natural result from: reliance and overreliance on technology being stuttery, unpredictable, inadventitious and costly.
In other words, it made more sense to 'frontload' hardthought and pondering so one was prepared [for whatever may come].
People also thought aqueducts were the best way to move water for thousands of years.
I can teach symbolic math effectively using clay tablet and reed pen today.
Numerical methods, however, would require at least minimal computing hardware.
I think there should be a guiding principle of using the least amount of tech in teaching any subject, just like principle of least privilege in cybersecurity.
You hit on the head something that I've been trying to put into words for a while now: it seems that the benefit of technology to increase learning appears to be a bell-shaped graph where for each combination of subject matter there is a sweet spot where the level of technological assistance becomes helpful but after which it replaces too much of your own internal neuron-connecting processes and inhibits learning.
That's a strong observation. And I'm wondering if humans will adapt; or if the intensity of "heat and light" required to "weld" connections together is non-negotiable.
> People also thought aqueducts were the best way to move water for thousands of years.
They still do because they still are? I'm hard pressed to think of a better way to move water than creating a gently sloped channel(this includes any pipelines that don't use pumps, by the way) for it to follow from where it is to where you need it. Many places that don't have sufficient local groundwater are still supplied by aqueducts to this day. If you have to pump it, then you're paying for the channel and the energy, which is worse than just using gravity.
And how many aqueducts lead to your dwelling today?
The problem can be fixed: adjust or realign the business model and incentives around the needs of the stakeholders.
- Students want a credential because business and society demands a credential
- Universities want income to support their missions of research, education and other "motives"
- Professors want to do research, and they don't really want to teach (in general)
- AI is a resource for learning, assessment and project development
- Society needs educated people
What ideas do you have to make everyone satisfied and maximally functional?
It is causing far more pain than benefit in the majority of institutions.
Particularly the cheating. Many students are flagrantly using it + the detection tools either aren't configured to check for it, or, if they do have something like TurnItIn, it doesn't work, and flags legitimately created content sans AI.
I think what we are seeing with all these AI detection tools is a whole lot of shooting the messenger. Technology made the problem of lazy students impossible to ignore, but the solution might be decidedly low-tech: have students spend some time talking about their work with the teacher, face to face.
For people who teach through writing, there are great tools out there like MATCHA: https://matchawriter.com. It's a secure writing environment that verifies authorship by recording the full writing process and, when needed, can restrict copy-paste, automation, and access to other applications.
In my opinion, most of Academia's pains are wholly self-inflicted.
The primary use case of universities are to produce a generation that can contribute to the society in a meaningful and positive manner. Nowadays in-order to do that, AI literacy is a must for graduates just like it was 15 years ago when it was a must to have computer literacy.
AI is a tool like internet was and it's the responsibility of the curriculum to teach proper usage of that tool rather than fighting against it.
We still teach primary school children how to do addition, despite the fact that during their working life they'll have access to a calculator virtually all of the time. A decade ago posting your CS assignment on StackOverflow and handing in the top answer was considered cheating, despite professional software developers doing pretty much that any time they ran into something unusually tricky. Why teach students differentiation when they can just put "derivative of x^2" in WolframAlpha?
Universities are supposed to primarily teach understanding, not basic tool use. Yes, tool use should be part of your education as well, but that's only a small fraction of it. A single "prompting AI to do X" course is probably more than enough. But if you completely skip the fundamentals and make the students fully dependent on the tools, how could they possibly contribute to building the next generation of tools, verify the output of the tools, or transcend the tools?
Yeah, they should indeed be teaching proper AI use as well, but in practice the easy access to AI is making it virtually impossible to teach the fundamentals, and that's the real problem here.
The analogies to prior technologies aren't insightful. The advent of LLMs has some similarities to the appearance of the internet, and very significant differences. Calling it a "tool" is also far too reductive. What kind of tool that existed before just writes everything for you? Is that the same kind of tool as a nailgun or a text editor? The "proper usage" that has to be learned is to learn to think for yourself before you get used to the word-bot doing all the thinking for you.
> AI is a tool like internet was
until it was overrun by unverified AI slop. Let's just get the slop from the source!
Yep. That's the premise of this article. The writer used to be the dean of humanities at University of Utah (great school, much of AI exists thanks to their work on CompGraphics, HPC, and Teapots) and is basically arguing that humanites and social science (HSS) programs need to build that kind of literacy.
A generation ago you saw similar resistance to the internet from HSS programs.
The $340M for 400 GPUs over two years next to 100,000 in 122 days is the whole argument in one line.
Lots of people talk about cheating here. That is not the framing I put in my mind. Or rather, if the students are cheating anyone it's themselves.
LLMs exist and are extremely useful in academia. I am currently riding a productivity wave, where I can spit out formalised proofs – not quite at the pace at which I can think of the theorems to prove, but almost. The thing braking my progress is that I have to build all the existing theory too, not just my own new stuff.
But for students, who do not yet know anything, it is a danger. From observing my students, it seems to take a certain amount of willpower to do the exercises without letting ChatGPT give them (or "explain them") the answer. And reading an explanation is not the same thing as working it out yourself. You just don't learn doing it, by not doing it!
Some students have this will power, some students don't. And it is also situational. Some students have stuff going (work or other factors) which makes it really hard too.
Yes, they can of course keep asking ChatGPT later in life, but they will not have the fundamental understanding which is required to even use LLMs for the advanced stuff. ChatGPT cannot explain you advanced concepts if you don't understand the fundamentals!
So, to test the students' understanding, we are back to having on-site exams, with the old-school supervision. And more students will fail, because there is now a way to solve exercises without learning.
I'm not sure what to make of this? Is any healthy skepticism about AI now treated as motivated luddism?
Academics has its many issues but universities are hardly the only sources of criticism about AI, and for all its benefits AI is poorly understood and often applied in relatively half-assed ways, even when it's well-intended. I would worry if thoughtful individuals weren't skeptical at all about AI.
It also seems reasonable to, say, want to evaluate students on their own merits free of the influence of a "homework machine" (to borrow a phrase from Shel Silverstein).
I'm reluctant to make predictions, and wouldn't describe myself as "anti-AI" but this sort of piece reminds me uncannily of the dot-com bubble.
This author has a ton of axes to grind and a poor grasp of computing history or how industry works. Pretending like it’s strange or even bad that industrial datacenter builders with tens of billions of leveraged capital can outbuild the GPU capacity of chronically underfunded grant dependent universities whose mission is not to create industrial scale compute resources is like complaining local technical school auto-mechanic program isn’t out-producing Ford’s auto manufacturing numbers.
But then as soon as the author rolled out complaints about DEI and ethics, you knew what was going on. Okay, please leave higher ed and work for the industry you love. As you say, they are doing just fine.
> The elite research layer of higher education (Stanford, Berkeley, MIT) is still an AI accelerant. But across the country, as Scott Latham told me, “the vast majority of institutions are in a state of AI paralysis.”
Tbf, the vast majority of institutions also suck.
State flagships and some of their second rung, a couple ivies and ivy tier privates, and military academies provide the bulk of talent that you need. And these universities have dedicated pipelines for people who start off at community colleges.
These universities are also much larger - UC Berkeley's and UCLA's 2025 undergrad population is almost double it's 1960 population [0].
And those who had skill or educational deficiencies can remediate them via non-trad or working professionals masters like GT, UT Austin, and CU Boulder's OMSCS for less than $15k.
[0] - https://www.universityofcalifornia.edu/about-us/information-...
Ah, sure, they train enough people to serve all needs. What a load of poppy cock.
They do.
The top 20 CS/EE/CE/ECE/EE degree programs (MIT, CMU, Stanford, Cal, UIUC, GT, UT Austin, Purdue, UW, UCLA, UCSD, Ivies, UChicago, NYU, UNC Chapel Hill, Duke, USC, UMinn, UMich) graduate around 15k-20k undergrads a year.
Throw in the next rung of well respected universities (eg. the rest of the state flagships like ASU or IU, the respected regional public programs like CalPoly, NCSU, or SUNY Buffalo, and the handful of regionally respected privates like WPI, Cincy, or both Rochesters) and that number likely doubles.
If you're graduating almost 50k grads a year from target programs, there just isn't enough room for the rest.
Choosing to go to (randomly choosing two universities to pick on) USF or TCU for CS straight out of HS instead of going to your local CC and then transferring to UC Davis or UT Dallas is dumb.
In 1980 you had a little under 2,000 4-year colleges in the US. In 2010 you had almost 3,000 4-year colleges after which the number of institutions started declining. Making more colleges while existing colleges were expanding headcount did nothing to help.
"Institution that assigns homework would prefer you not have a machine do your homework for you"
It’s almost like a place focused on improving the human should operate differently from a place building commercial products.
Couldn't agree more, and yet, attendance itself is a commerical product.
The commercial product is the paper that says "University of Whatever says you learned Whatever". If the university loses credibility (by producing a bunch of people who cannot function without a chatbot) then the product is no longer valuable.
Luckily for us all these good natured AI people will liberate us from the necessity of our own value and places like universities, which give us structured environments to participate in intellectual pursuits, will be in high demand.
This assumes every degree is valuable in the marketplace, which is simply not the case. The product is luxury attendance, the degree is secondary to that.
The issue is more one of current passive, abstract forms of assessment. This is laziness on the part of instructors. They will simply have to move to more complex, project-based assessments and accompanying interviews, which is more time consuming but will be far more accurate than the more traditional system of assessment was anyway. On sum it will be better for everyone.
AI is totally going to replace the university system. This is like the coal and oil guys sabotaging EVs and renewable energy. Good for them, bad for us.
>the two most reliable places to get an answer are the physical library and AI.
Uh in what universe? Obviously physical books are helpful but are not most of the new ones being written with the assistance of an LLM?
Depending on the type of question being asked and the complexity, human intervention is still needed because LLMs still face an information retrieval problems.
Definitely the case that LLMs still face a "not actually rational nor conscious" problem.
Einstein once said physics is like looking at the outer movement of a watch and attempting to understand its internal mechanisms.
LLMs watch the outer movement of our keyboards in an attempt to comprehend our varieties of intelligence (intuitive, rational, symbolic, logical, ...)