Flip the classroom, make students learn the material on their own (Using AI or whatever resources they want to use) and then in-classroom time is divided on working on problems (without AI assistance which can be controlled in this environment) and quizzes/exams (again without AI). We don't need lectures anymore, they are an incredibly ineffective way to learn.
Lectures have been an incredibly ineffective way to learn forever. Faculty continue to lecture, and we continue to build lecture-style classrooms, further enshrining this poor approach. Active learning works, and yet both faculty and students dislike it. Faculty like to talk and pretend they're teaching, and students like to listen and pretend they're learning.
All to say—I wish it was this easy to change the academy. But it's not.
Good lectures are phenomenally useful, far better than individual unguided review of the material. They're often IME very interactive too.
STEM subjects are particularly hard to create good lectures for. And STEM expertise and speaking/communication skills don't always overlap either.
The non-STEM classes I learned the most in are the ones I learned the most in lecture in. The STEM classes, on the other hand, were pretty scattershot without as much correlation.
An LLM-based toolset could likely be much better than that and at least as good as bad lectures, but the guardrails are gonna need to be really really really good.
> Lectures have been an incredibly ineffective way to learn forever.
Mainly due to shortage of very good lecturers, no? I can not see a better way to cultivate the professional pride than to attend lectures of truly remarkable professors. The style, the manner, the attitude go much beyond the dry proofs. I'm an applied math major.
You're describing your response to a performance, not to a learning opportunity. Is it fun to watch a great performer? Sure. Is that an effective way to learn? No.
Does lecturing have a place in disseminating ideas? Sure. I love that scene from Oppenheimer when he attends Heisenberg's lecture, being exposed to cutting edge ideas directly from the mouth of a truly remarkable professor. Watching that gave me a better appreciation of lecturing's original purpose and historical importance. But that's very different from teaching well-understood concepts and skills.
I'm not sure about your experience but most of my professors would teach classes tangentially related to their field of expertise. So we were almost always exposed to the "well-understood concepts and skills" alongside what was cutting edge in their niche at the time.
It was subtle, but easy enough to pick up on if you were being attentive in class..
Sorry, but I must disagree. There is much more to learning process than just the material itself. We are social animals, so the emotional aspect matters to the majority of us. Highly technical fields are not an exception. The attitude of the lecturer and his reaction to the questions from the audience, sidetrack discussions -- it all counts. At least to me and the people I have known.
At the same time, lectures of those with no charisma is a real torture, no doubt about that.
15 years later I can still remember parts of my lectures that were interesting. The only thing I remember about studying is those now banned little energy drink shots.
I still remember one course where the lectures were basically just 1:1 summaries of the textbook, so I said "This is a waste of time. See you at the exam", and many of the more traditional showing-up-is-half-the-battle crowd were like "GASP, you can't do that!" because from very little we're taught that attendance is more important than outcome.
Citation needed, the last time a colleague engaged with the education college and supposed educators they looked at an induction proof with sheer bewilderment as if teaching this at all was impossible
I had some "reverse classroom" classes back in college, and it was the best kind of class for me. Read the papers on your own time before class, and spend class discussing and in tests.
It did, however, absolutely require everyone to prepare for every class. Some people complained a lot about this, which might be why this was not as popular as more common lectures.
I worry the sort of core classes that are in massive lecture halls (and everyone remembers hating) would be more of a "blind leading the blind" situation if run like this. Maybe not with lavish funding and small classes. I had a couple classes like you describe, but they were on specific topics I was very interested in near the end of my undergrad program.
You are right and this is the way, it's just a lot easier said than done, especially the quizzes/exams part.
There is a tension between:
- authentic assessments (not multiple choice/
- grading resources
- computer lab availability
- students using AI in very discreet ways on their own laptop when taking an in-person exam
So overall you are on point, it's just really hard to do honest authentic assessment at scale right now (in person or otherwise).
I see a potential for it to get much better, or much worse... Hard to tell which way it will go right now.
Yep. I've been using AI to teach myself system design and it's been a god send. Was struggling with other courses because I couldn't have it tailored to what I wanted.
The grant application process is wild now, half the questions I'm to respond to feel AI generated as well. I feel certain that the reviewers are just taking my responses and feeding them.into another AI to evaluate too.
I'm not as pessimistic about its impact on scientific publishing. Yes, you can churn articles faster, but if people catch you gaming this system to extreme lengths your reputation will take a huge hit. And the system is very transparent and visible, so it'll follow you forever.
And I welcome the change. In my long experience in academia, I've only found two types of practitioners:
1% are the absolutely brilliant minds that academia was originally created for. People that, without a doubt, leave their mark in the vast corpus of human knowledge. I consider myself fortunate for meeting and learning from them, and I thank the academia ecosystem for that.
But the remaining 99% are the maximalists, as described in the article. More papers/students/grants, then repeat. Worse enough, they're absolutely useless outside of academia, as they never did anything at all outside that bubble.
An embarrassing lot of CS professors would stumble around your average production codebase.
I think AI is just the final nail in the coffin for the latter bunch, as they have been dogs eating their own tail for decades already.
I think it's very field dependant. In maths and biology I've seen very few professors that could reasonably be described as maximialists here; computer science really seems to be the outlier. My impression is that impactful CS professors tend to be more strongly associated with either maths, or the field that their cs research is being used in.
Arts faculty on the other hand seem to basically just be a popularity competition.
Matches my experience. Hard science may not have that many maximalists. Applied science and technology (electrical, electronics, computer science, even mechanical), I've found plenty.
What got me out of academia (yup, I was a professor) was:
Do you have vast field experience and want to get into the classroom to teach how it's really done? Tough luck, you should've spent your time writing papers.
No matter how much you know or how good you are, everything is about feeding the maximalist machine, if you're an outlier, worse people with better "scores", more papers and never leaving faculty will forever beat you until they retire.
I took a good look at the publishing process. Absolutely everything about it was back-channeling to carefully select the topic, scope and reviewers of a paper to get it through the process. Goodhart's Law at its finest.
Advice given to me: "aim for a lower rank and be happy with teaching the whole thing while the old professor takes a nap in the corner".
AI or not AI, anything destroying that self-perpetuating bureaucracy is welcome.
I don’t believe that at all. Maybe it’s killed large lecture classrooms. There’s a lot of other ways to engage students and ensure they’re learning but it involves being more active and actually communicating with them instead of yapping lectures at them and making them write about it. I find most college classes that involved lectures to be a waste, why sit and listen to a teacher summarize a book for an hour? My best experiences were classes with active engagement and producing things in class or bringing them back to class to share discuss and learn.
"Yes, going back to paper and pencil strains our current resources, but is a likely necessity" -- When I reached this sad, arthritic point in the article it was clear that the author has no constructive idea about what happens next in education. There is no critical dialogue regarding assessment and its necessity. There is no critical understanding and projection for what intelligence as a service represents for society. Just a simple monastic shrug and a familiar scent of old world pencil lead.
It frustrates me that your comment is being down voted, because I agree with your sentiment.
Too many educators don't actually know how to educate, and only focus on what to educate.
LLMs are offering some ways to dramatically improve how people learn (and therefore how well they learn, what they learn,etc - to rapidly accelerate and improve outcomes). However most educators, who are ignorant to the principles of how people learn, have no idea how to harness that potential. The result is in most cases students are just using AI to sabotage their own learning, because no better alternative is being offered.
It's a hard problem... But it's a shame that so few people are working constructively and pragmatically on it.
> LLMs are offering some ways to dramatically improve how people learn
But in practice, they are overwhelmingly having the opposite effect, and if we're realistic about the structural incentives that have no practical path to being changed in higher ed, this will continue to be the case indefinitely.
I hope some have already thought of this addition to the AI-inundated academic's bag of tricks -- in college and graduate school I was fortunate to have some important one-on-one conversations with professors, guests and peers where I had dialogues which meaningfully stimulated and advanced my thinking in a variety of subjects. It is clear that frontier models can provide similar opportunities on demand in nearly any subject. One assessment could be student sharing of raw chat logs on a course relevant topic where particular questions were engaged and discussed. The focus is not on the prose product, but the development, the grappling and the questions. There is intellectual value here in the depths of the nested questions, corrections, and unique additions made by the student during the exchange. If not a substitute for an essay, it could be required pre-planning for it.
I'm working on my PhD nonetheless. Here was a meditation of mine on the Literature Review that is relevant here:
---
Academic literature (AL) predates the social media (SM) of recent decades.
AL identifies and preserves human knowledge across time, space, language, andexperience level.
These implicit "goals" contrast with SM's emphasis on the present moment, colloquial focus, idiomatic word choice, and general disregard for larger context beyond the.
Therefore, when engaged in producing AL, a detached mindset seems helpful.
One enters a discourse with people across the ages who were soberly adding to the sum of human knowledge, not pursuing SM "likes", and not regressing to the mean of some Artificial Intelligence corpus.
AL seeks to cover the prior art, to honor those who came before, to minimize duplication of effort, and to filter the "novelty" out of ideas by detailing their pedigree.
The writing style of AL targets a reader half a century in the future of unknown gender, nationality, and depth within either the topic or with the English language itself.
Therefore, the writing style needs to favor:
- Simplicity in word choice. Archaic words or definitions, however valid, are not preferred. Two or three shorter words is often easier on that unknown reader.
- Linearity in development. No dramatic tension. Set the ideas in front of the reader and move through them in order.
- Connectedness. The chain of the ideas is obvious as we move from one to the next, so that reiterating to remember the current topic is less necessary.
- Cohesiveness. A paragraph should almost stand on its own, because it contains enough information to make its point even if quoted within another article.
Solving the paper submission is easy. Just hold frontal interview where the submitter defends their paper. They can't create papers every day and still be knowledgeable about them in depth.
We are hurling to a reality where the only noteworthy metric is human to human validation.
You could do a two stage process, a blinded text review and an unblinded review that can only be failed if you have a vigorous lack of knowledge about your work.
I highly doubt anything like that will be implemented though.
With the volume of outputs in today's academia, this is simply not possible. There are conferences with tens of thousands of submitted papers, grants have hundreds of pages, etc.
If a journal finds that it's getting more papers than peer reviewers are willing to go through, how does a more heavyweight, synchronous review process solve the problem? Many researchers already find peer review requests annoying, they're not going to agree to hold a bunch of video calls.
What superhuman? This system will never have a 99.99% accuracy based on its current prediction models and data input, neither is it a targeted to make us superhumans in the first place.
It still will need human supervision for corrections and if it doesn't it won't require humans to process further. Humans are not in the central picture of the future of AI.
Flip the classroom, make students learn the material on their own (Using AI or whatever resources they want to use) and then in-classroom time is divided on working on problems (without AI assistance which can be controlled in this environment) and quizzes/exams (again without AI). We don't need lectures anymore, they are an incredibly ineffective way to learn.
Lectures have been an incredibly ineffective way to learn forever. Faculty continue to lecture, and we continue to build lecture-style classrooms, further enshrining this poor approach. Active learning works, and yet both faculty and students dislike it. Faculty like to talk and pretend they're teaching, and students like to listen and pretend they're learning.
All to say—I wish it was this easy to change the academy. But it's not.
Good lectures are phenomenally useful, far better than individual unguided review of the material. They're often IME very interactive too.
STEM subjects are particularly hard to create good lectures for. And STEM expertise and speaking/communication skills don't always overlap either.
The non-STEM classes I learned the most in are the ones I learned the most in lecture in. The STEM classes, on the other hand, were pretty scattershot without as much correlation.
An LLM-based toolset could likely be much better than that and at least as good as bad lectures, but the guardrails are gonna need to be really really really good.
> Lectures have been an incredibly ineffective way to learn forever.
Mainly due to shortage of very good lecturers, no? I can not see a better way to cultivate the professional pride than to attend lectures of truly remarkable professors. The style, the manner, the attitude go much beyond the dry proofs. I'm an applied math major.
You're describing your response to a performance, not to a learning opportunity. Is it fun to watch a great performer? Sure. Is that an effective way to learn? No.
Does lecturing have a place in disseminating ideas? Sure. I love that scene from Oppenheimer when he attends Heisenberg's lecture, being exposed to cutting edge ideas directly from the mouth of a truly remarkable professor. Watching that gave me a better appreciation of lecturing's original purpose and historical importance. But that's very different from teaching well-understood concepts and skills.
I'm not sure about your experience but most of my professors would teach classes tangentially related to their field of expertise. So we were almost always exposed to the "well-understood concepts and skills" alongside what was cutting edge in their niche at the time.
It was subtle, but easy enough to pick up on if you were being attentive in class..
> Is that an effective way to learn? No.
Sorry, but I must disagree. There is much more to learning process than just the material itself. We are social animals, so the emotional aspect matters to the majority of us. Highly technical fields are not an exception. The attitude of the lecturer and his reaction to the questions from the audience, sidetrack discussions -- it all counts. At least to me and the people I have known.
At the same time, lectures of those with no charisma is a real torture, no doubt about that.
15 years later I can still remember parts of my lectures that were interesting. The only thing I remember about studying is those now banned little energy drink shots.
I still remember one course where the lectures were basically just 1:1 summaries of the textbook, so I said "This is a waste of time. See you at the exam", and many of the more traditional showing-up-is-half-the-battle crowd were like "GASP, you can't do that!" because from very little we're taught that attendance is more important than outcome.
Citation needed, the last time a colleague engaged with the education college and supposed educators they looked at an induction proof with sheer bewilderment as if teaching this at all was impossible
I had some "reverse classroom" classes back in college, and it was the best kind of class for me. Read the papers on your own time before class, and spend class discussing and in tests.
It did, however, absolutely require everyone to prepare for every class. Some people complained a lot about this, which might be why this was not as popular as more common lectures.
I worry the sort of core classes that are in massive lecture halls (and everyone remembers hating) would be more of a "blind leading the blind" situation if run like this. Maybe not with lavish funding and small classes. I had a couple classes like you describe, but they were on specific topics I was very interested in near the end of my undergrad program.
You are right and this is the way, it's just a lot easier said than done, especially the quizzes/exams part.
There is a tension between: - authentic assessments (not multiple choice/ - grading resources - computer lab availability - students using AI in very discreet ways on their own laptop when taking an in-person exam
So overall you are on point, it's just really hard to do honest authentic assessment at scale right now (in person or otherwise).
I see a potential for it to get much better, or much worse... Hard to tell which way it will go right now.
Yep. I've been using AI to teach myself system design and it's been a god send. Was struggling with other courses because I couldn't have it tailored to what I wanted.
I had a chat with my state legislators to streamline education; shift all public school and university funding to libraries staffed by SMEs
Mandate N hours year of and guided group work for under 18s
Mandate N hours for becoming an SME for roles that require such
Break the pipeline from the factory era of linearly pumping out kids who are just smart enough to run the machines
Guided group work? Have you ever learned anything in a group project other than how lazy half of your group is?
Just because an idea is different doesn't mean it's good
I wrote an essay about how I am handling AI in my CS courses:
https://htmx.org/essays/universities-and-ai/
The grant application process is wild now, half the questions I'm to respond to feel AI generated as well. I feel certain that the reviewers are just taking my responses and feeding them.into another AI to evaluate too.
I'm not as pessimistic about its impact on scientific publishing. Yes, you can churn articles faster, but if people catch you gaming this system to extreme lengths your reputation will take a huge hit. And the system is very transparent and visible, so it'll follow you forever.
> And the system is very transparent and visible
Is that really the case though? It seems like quite a lot of cases get hushed up and swept under the rug.
On the flip side, the value of new, clever and repeatable experimental results goes up when compared to regurgitated publications.
Very much this. Very funny to see we're back to hardware is still hard.
And I welcome the change. In my long experience in academia, I've only found two types of practitioners:
1% are the absolutely brilliant minds that academia was originally created for. People that, without a doubt, leave their mark in the vast corpus of human knowledge. I consider myself fortunate for meeting and learning from them, and I thank the academia ecosystem for that.
But the remaining 99% are the maximalists, as described in the article. More papers/students/grants, then repeat. Worse enough, they're absolutely useless outside of academia, as they never did anything at all outside that bubble.
An embarrassing lot of CS professors would stumble around your average production codebase.
I think AI is just the final nail in the coffin for the latter bunch, as they have been dogs eating their own tail for decades already.
I think it's very field dependant. In maths and biology I've seen very few professors that could reasonably be described as maximialists here; computer science really seems to be the outlier. My impression is that impactful CS professors tend to be more strongly associated with either maths, or the field that their cs research is being used in.
Arts faculty on the other hand seem to basically just be a popularity competition.
Matches my experience. Hard science may not have that many maximalists. Applied science and technology (electrical, electronics, computer science, even mechanical), I've found plenty.
What got me out of academia (yup, I was a professor) was:
Do you have vast field experience and want to get into the classroom to teach how it's really done? Tough luck, you should've spent your time writing papers.
No matter how much you know or how good you are, everything is about feeding the maximalist machine, if you're an outlier, worse people with better "scores", more papers and never leaving faculty will forever beat you until they retire.
I took a good look at the publishing process. Absolutely everything about it was back-channeling to carefully select the topic, scope and reviewers of a paper to get it through the process. Goodhart's Law at its finest.
Advice given to me: "aim for a lower rank and be happy with teaching the whole thing while the old professor takes a nap in the corner".
AI or not AI, anything destroying that self-perpetuating bureaucracy is welcome.
> An embarrassing lot of CS professors would stumble around your average production codebase.
As would be expected. The value of computer science has very little overlap with navigating a typical production code base.
Well, as would be expected of the 1% of brilliant-mind CS professors.
As for the remaining 99%, maybe they should stop pretending they know (and giving advice) about navigating a typical production code base.
I don’t believe that at all. Maybe it’s killed large lecture classrooms. There’s a lot of other ways to engage students and ensure they’re learning but it involves being more active and actually communicating with them instead of yapping lectures at them and making them write about it. I find most college classes that involved lectures to be a waste, why sit and listen to a teacher summarize a book for an hour? My best experiences were classes with active engagement and producing things in class or bringing them back to class to share discuss and learn.
"Yes, going back to paper and pencil strains our current resources, but is a likely necessity" -- When I reached this sad, arthritic point in the article it was clear that the author has no constructive idea about what happens next in education. There is no critical dialogue regarding assessment and its necessity. There is no critical understanding and projection for what intelligence as a service represents for society. Just a simple monastic shrug and a familiar scent of old world pencil lead.
It frustrates me that your comment is being down voted, because I agree with your sentiment.
Too many educators don't actually know how to educate, and only focus on what to educate.
LLMs are offering some ways to dramatically improve how people learn (and therefore how well they learn, what they learn,etc - to rapidly accelerate and improve outcomes). However most educators, who are ignorant to the principles of how people learn, have no idea how to harness that potential. The result is in most cases students are just using AI to sabotage their own learning, because no better alternative is being offered.
It's a hard problem... But it's a shame that so few people are working constructively and pragmatically on it.
> LLMs are offering some ways to dramatically improve how people learn
But in practice, they are overwhelmingly having the opposite effect, and if we're realistic about the structural incentives that have no practical path to being changed in higher ed, this will continue to be the case indefinitely.
Yes, it's tragic.
It's as if electricity had the potential to save lives but everyone was just frantically electrocuting themselves.
I hope some have already thought of this addition to the AI-inundated academic's bag of tricks -- in college and graduate school I was fortunate to have some important one-on-one conversations with professors, guests and peers where I had dialogues which meaningfully stimulated and advanced my thinking in a variety of subjects. It is clear that frontier models can provide similar opportunities on demand in nearly any subject. One assessment could be student sharing of raw chat logs on a course relevant topic where particular questions were engaged and discussed. The focus is not on the prose product, but the development, the grappling and the questions. There is intellectual value here in the depths of the nested questions, corrections, and unique additions made by the student during the exchange. If not a substitute for an essay, it could be required pre-planning for it.
I'm working on my PhD nonetheless. Here was a meditation of mine on the Literature Review that is relevant here:
---
Academic literature (AL) predates the social media (SM) of recent decades. AL identifies and preserves human knowledge across time, space, language, andexperience level.
These implicit "goals" contrast with SM's emphasis on the present moment, colloquial focus, idiomatic word choice, and general disregard for larger context beyond the. Therefore, when engaged in producing AL, a detached mindset seems helpful.
One enters a discourse with people across the ages who were soberly adding to the sum of human knowledge, not pursuing SM "likes", and not regressing to the mean of some Artificial Intelligence corpus. AL seeks to cover the prior art, to honor those who came before, to minimize duplication of effort, and to filter the "novelty" out of ideas by detailing their pedigree.
The writing style of AL targets a reader half a century in the future of unknown gender, nationality, and depth within either the topic or with the English language itself.
Therefore, the writing style needs to favor:
- Simplicity in word choice. Archaic words or definitions, however valid, are not preferred. Two or three shorter words is often easier on that unknown reader.
- Linearity in development. No dramatic tension. Set the ideas in front of the reader and move through them in order.
- Connectedness. The chain of the ideas is obvious as we move from one to the next, so that reiterating to remember the current topic is less necessary.
- Cohesiveness. A paragraph should almost stand on its own, because it contains enough information to make its point even if quoted within another article.
Solving the paper submission is easy. Just hold frontal interview where the submitter defends their paper. They can't create papers every day and still be knowledgeable about them in depth.
We are hurling to a reality where the only noteworthy metric is human to human validation.
Paper reviews are traditionally blinded, so the reviewer doesn't know the authorship of the paper they're reading.
You could do a two stage process, a blinded text review and an unblinded review that can only be failed if you have a vigorous lack of knowledge about your work.
I highly doubt anything like that will be implemented though.
With the volume of outputs in today's academia, this is simply not possible. There are conferences with tens of thousands of submitted papers, grants have hundreds of pages, etc.
Grants typically have at most 10 pages of actual content. The rest is mostly compliance with regulations.
If a journal finds that it's getting more papers than peer reviewers are willing to go through, how does a more heavyweight, synchronous review process solve the problem? Many researchers already find peer review requests annoying, they're not going to agree to hold a bunch of video calls.
Big part of the annoyance is that journals demand basically free labor - while costing massive amount of money if you want to read them.
A review call might just end up being less work then reading a lot of slop papers.
How about we embrace the era of the superhuman?
It's not the era of the superman coming. It's the era of the sewers man.
What superhuman? This system will never have a 99.99% accuracy based on its current prediction models and data input, neither is it a targeted to make us superhumans in the first place.
It still will need human supervision for corrections and if it doesn't it won't require humans to process further. Humans are not in the central picture of the future of AI.
> This system will never have a 99.99% accuracy
Do humans?
Superman is here. Just missing physical form. That’s coming. 2028.
year of the Linux desktop?