I simply have to express my deep gratitude and love for everything you’ve been writing and sharing about GenAI, Ethan. This article is a shining example of what growth mindset plus nonzero sum thinking looks like in action. I’m giddy to experiment with these latest prompts and can’t thank you enough for how much you’re helping me, and so many others, navigate the incredible Cambrian explosion of paradigm-shifting technologies we’re all lucky enough to be alive to experience. Together we go far. Thank you, Professor Mollick, for helping to light the way.
This post is apropos this morning. I teach marketing and entrepreneurship in high school, and your work has inspired me tremendously. We have gone all in with AI this semester and having a great time doing it. I've been creating prompts to use as class activities, but there were a few pain points--students had to copy the prompt exactly, I had no easy way of seeing how the activity was playing out for students, and the not-uncommon access issues. I've been working on a tool to solve these. I literally used it for the first time yesterday, but I'd love it if you and any like-minded educators could check it out: https://edufocus.ai/.
Teachers create prompts that are the AI class activities, these are loaded in a session (different class periods, terms, etc), and then a link is posted for students to complete the activity. There is a log of all the chats in a session and if you require login you can see what each student did and students can access them later.
Just to give a demo of what it does, you can try these two tasks. The first is an activity I made for our 10th grade English teacher in our business academy. He is having students research how we could improve our city using a fictional surplus budget using the United Nations Sustainable Development Goals. https://edufocus.ai/chat/?join=3-24c263
Here is another based on the first activity I did with students with ChatGPT. We used it to discuss if ChatGPT should be banned at the school. This example is also gamified with turns and scoring. https://edufocus.ai/chat/?join=4-5d7142
Ethan, I'm like a dog waiting for its treat...I can't wait for when you post! Your insight, knowledge, and experiments with generative AI are what I as an instructional designer want to know about. I want to bring back to the faculty actionable items they can do NOW. I always send them the link to your posts. Thank you for, as @Maciej Workiewicz said, blazing a new trail for us! I so appreciate you and your efforts in getting to know generative AI in the education field and SHARING with us.
I have a question you may or may not be able to answer (others are welcomed to chime in as well): is there any training, in-depth training, out there on writing extensive prompts to garner the generated content to its full potential? Is it a tiered approach, drilling down, basically scaffolding your prompts or are there other ways to prompt that would work better for certain disciplines? Just curious...writing the 'right' kind of prompt will be key for faculty and I'd like to provide how to prompt so they are successful in their endeavors. :)
Thank you for your time...and keep on blazin' that trail Ethan! Can't wait to take your webinar May 2nd!! Heather - ID at a community college
Hi Ethan, thanks for this great article. I sent you an invite on Linkedin to be a speaker at an elab.ai in New Zealand (online) . Would you be interested? I am expecting over 100 people:)
unfortunately, i am stuck in a meta -crisis - that i mentioned on Conor Grennan's linkedin post -
we still need to solve the hardware, mobile network and electricity problems (assuming mid-day meals; a roof over the head is not a problem and of course, no conflict and of course, no religious indoctrination )
until the next generation everywhere is in a safe/secure (no conflict zone) with basic infra - of food, shelter etc. and mobile connectivity, electricity and hardware ... the burgeoning inequity would undermine many of our goals!
i would like to disagree with your use of the word - "democratizing" - while semantically and functionally correct is not useful in the broader context.
I asked it to explain situational irony. Its explanation was fairly bad, pretty much garden variety not-quite-bright middle school English teacher who doesn't get it and has low expectations. Not impressed. I think your standards may be low.
If you think people are going to learn how to program from dumbed down explanations that gamify things, you are sadly mistaken.
I do learn a lot from you and appreciate the blog, but I think you're getting into a gee whiz this is so great mode that is defeating your purposes. To learn, you have to work and to think. I'm glad you got that little 25% improvement in Ugandan something or other. Nice. But dream on if you think that people are going to get smart from gamifying physics and programming.
Still, a little improvement is better than none. However, you seem unaware of Hattie's (sp?) research showing that almost any intervention works well for a while. The kind of changes you're seeing are what's typical for any intervention. Nothing special except the AI label.
I think a few things are getting muddled here, and it is worth clarifying. The three studies I cited were all randomized controlled trials, and 25% improvements in income directly from an educational intervention are huge. That is a major impact on life outcomes, and shouldn’t be dismissed cavalierly. Plus, the experimental approach lowers the chances that they are a Hawthorne Effect (an intervention having an effect just because it is an intervention).
Second, none of the work here is about gamification. Simulation and game-based learning (which I have worked on for a decade )are quite different from gamification, and they can really impact outcomes. But this post focused less on that and more on the way AI could provide multiple ways of trying to understand a topic in an age appropriate way. (The Python game/program was just part of it, but one I found very neat.) We learn best from multiple methods of teaching and working combined, which makes it useful that the system could create a reading and a quiz and an activity and a toy model, etc.
Third, I totally get that the AI screws up and gives weak answers sometimes. Thank goodness, or there would be no room for us. The idea is that the explanations & material that the AI comes up with can assist teachers in doing their job, it isn’t a one-shot replacement for teachers. Got a mediocre explanation? Tell the AI how to improve it. It is much faster (and cheaper) then most other ways of prepping.
I do think Professor Mollick's claim that 8% to 25% improvements are so important is rather naive. John Hattie's book on education claimed that an effect size of .25 was pretty much the minimum for any kind of significance. Basically, intervening tends to produce gains, for a while. So I don't think his experiments are pointing to anything much, at least not those that are around 15% or below.
Thank you for your thoughtful reply. Have those experiments been replicated?
I think gamification and anything else that teaches basic skills is well worth doing. E.g., a game-based app that actually taught fluent reading would be a marvelous thing since the skill builds on itself and has millions of applications. However, I do not think adding "methods" helps. It often hurts. I will give you an example from Quizlet. Quizlet uses flashcards as the base method. It also has functions that let you take "quizzes" in various formats such as clozes, multiple-choice, and matching. These methods are slow and pointless if you want to (for example) learn a foreign language or memorize facts. They are a good match for the silly stuff done in school, but they are counterproductive. A student would be much better off actually working with flashcards for ten minutes to memorize (hear, say, translate, etc.) a set of words or build fluency with them than he would be doing multiple-choice questions or filling in blanks.
So, to generalize from this, I do not think that it is true that "We learn best from multiple methods of teaching and working combined." "Multiple methods" are often quite inefficient, esp. since they encourage students to try this and that and never get serious about anything. One effective teacher using a few efficient methods is quite a bit better than five teachers using five methods each. So no, that is an error.
Perhaps you didn't intend it to be as general as it came across. What I would agree with is that the method that a student will actually stick with is superior to the supposedly more effective method he gives up on. But that's not so much a matter of "multiple methods" as it a matter of having options that let a student find out what he works with best, as an individual.
I do agree that the AI can help mediocre teachers improve their awful teaching, and the fact is that most teachers are mediocre. Some of its explanations are quite a bit better than what you get from most textbooks, especially since it can simplify things.
Thank you again for all you do. I'm learning a lot from you and have so many leads that I don't even have time to pursue them. My own students have made huge gains in writing in two months of working with ChatGPT.
I'm interested in what you have been doing with writing with your students. I started this semester with ChatGPT in my classes (here is something I did back in January–https://blog.gameful.me/robo-redraft-rumble/). I've found that I've replaced most reading in my marketing class with ChatGPT prompts. They are actually reading more because it is in context of a problem they are trying to solve. It's takes PBL to a whole new level.
Honestly, the main thing I have done is to have them dialogue with the AI and push it to criticize their writing and give them suggestions, then have them revise accordingly.
One thing that is becoming clearer than ever is that people who work hard can make extraordinarily fast progress with the aid of ChatGPT, so that the gap is widening, because the fact is, you don't learn anything at all unless you push yourself, and asking for criticism -- be it from an AI, a tutor, or a teacher or a friend -- is an invitation to push yourself. I have a few students who won't even ask for criticism. They just say "please fix this so that I get an A" or some such (they actually give me logs of such dialogues).
Maybe that is better than nothing, actually, but they will be outstripped at mind-boggling speeds by more dedicated students.
I don't know how I missed your reply! Better late than never. I'm right there with you with you with the students who are still looking for a shortcut. I've found that I can prompt ChatGPT to provide limited help. I think of the Blooms 2 Sigma problem that has been discussed in terms of AI lately. If AI can be a personal tutor that gives tremendous gains, it will only widen the achievement between students who choose to interact with the tutor and those who don't.
I do think it is better than nothing. I have students using ChatGPT to help write "Shark Tank" investor pitches. Some students have engaged with the learning, others just wanted to get to the end. Either way 100% of them have a presentation to give. The question and answer session will be the determining factor.
I've been working on a interface that uses the ChatGPT API to allow sending students prompts and logging their chats. The writing tutor might interest you. https://grobots.ai/
Well, I tried it out, but to be blunt, I don't think it was helpful. I wrote this snarky response, and its responses were generic. I think it was giving good general advice for essay writers, but not advice specific to what I wrote.
I wrote:
ChatGPT should not be banned in high schools. Of course not. Why not? Well, let's go back to something I learned from a high school principal: "Don't pass a rule you can't or won't enforce." This falls into both categories. Can't enforce it because students can submit their work to the bot and get useful feedback or even whole essays that are undetectable. Yes, yes: you can tell that they "cheated" in some way if the work is far beyond their capacities. So what? They'll learn to make it worse to cover their tracks. So you can't enforce this rule. But also, you won't enforce it. You won't enforce it because the truth is that the bot is a great help to students, you will find that out soon, and the help it gives is going to make your life easier. No one makes his own life harder. Oh, but what if "you" are the principal? Well, the teachers will just ignore your crazy rule, just as they usually do.
I do think Professor Mollick's claim that 8% to 25% improvements are so important is rather naive. John Hattie's book on education claimed that an effect size of .25 was pretty much the minimum for any kind of significance. Basically, intervening tends to produce gains, for a while. So I don't think his experiments are pointing to anything much, at least not those that are around 15% or below.
I think I could produce a 15% improvement in most teachers' performance by telling them I'd visit five minutes a week, without notice, on two separate occasions; or by giving them some simple instruction (if they followed it) such as "have a one-minute quiz at the end of every class with an answer that shows knowledge of the content"). Really, those kinds of gains are not that hard to get.
Ethan, this is a very good post. You are blazing a new trail here. I posted your prompt to Bing Creative and played a little bit with it. One specific use I see for my own approach is to make programming and modeling accessible to more of my students. I followed your template and asked GPT-4 to explain the Central Limit Theorem to university students and then followed up with this prompt:
write a code in Python that helps to visualize how central limit theorem works. Do this by giving the user a possibility to choose among multiple distributions, for example binomial, Poisson, Beta, and normal distributions. Allow the user then to choose the number of samples and the size of each sample. The code then generates draws from a given distribution and calculates sample means. The code should also visualize the result by showing a histogram of the sample means.
It worked, but the code in Bing, for some reason, can't be copied. I did the same with the basic version of GPT-4 via OpenAI and pasted the code to Google Colab. It works as intended. I remember coding this myself. It would take several hours. Now it is just a minute away.
I also tried more complex code. The first principle I found to work well is to write in normal language (rather than to try mathematical notation). Thus, instead of 2^n just write raise 2 to the power of n. The second one, and I think you mention it yourself somewhere, is to break the code into smaller chunks. But then some knowledge of coding may be required for successful assembly.
Love this! Your examples of practical applications of AI are more thoughtful/useful than 90% of the writeups from other writers on AI that I come across these days.
It's fantastic to hear that you've launched Wharton Interactive! I'd like to share a platform I've developed for learning Financial Mathematics called 21ifm (https://21ifm.com/). This platform enables educators to design interactive lessons tailored to students' interests, motivating them to engage with advanced concepts and take ownership of their learning.
By using personalized explainers, teachers can assign homework prior to the actual class lesson, as well as post-lesson assignments. Students can then incorporate their findings, their classmates' insights, and the teacher's guidance to complete a comprehensive project. This approach fosters a more immersive and collaborative learning experience.
Pavan you have done a good job. It's true that the design of the site is difficult for me. But I think this is a matter of taste. Good luck in your next works!
I think Bing's explanation of entropy using the analogy of a deck of cards is misleading. Bing states that the deck of cards is a closed system, and being such entropy can only remain the same or increase. However, there is an external agent shuffling the cards, and conceivably the deck can return to the same order which contradicts its own statement. This is obvious when you start with a smaller ordered set of objects.
That is all well and good, but it doesn't address the real problem. The lack of standards in tracking and finding the best edtech. The lack of interoperability. Have millions or billions of people able to create educationally material on the fly is fine but you still have a data overload issue. What we need recommender and performance tracking systems so that learners can navigate this sea of materials. Because the real democratization is not going to be anyone being able to produce materials, it is going to be anyone being able to educate themselves with only tools and no educators. It is chat GPT as the teacher.
Thank you for the enlightening article on the transformative role of AI and EdTech in democratizing education, specifically the powerful capabilities of GPT-4. At Eureka Simulations, we share your enthusiasm for the impact these technologies can have on learning.
In our work, we've seen first-hand how simulations play an instrumental role in creating engaging, interactive learning environments, bringing theoretical concepts to life in an experiential way. Similar to the revolutionary leap you describe with AI, simulations enable learners to engage with complex ideas and scenarios, enhancing their understanding, and allowing them to tailor their learning journey to their needs and pace. They also provide educators with a powerful tool to deliver personalized, impactful education at scale.
Your emphasis on democratizing access to quality education echoes our mission. The power of AI, as demonstrated by GPT-4, combined with immersive simulations can significantly enhance the way we learn and teach, making quality education accessible and tailored to all, regardless of location, language, or background. It's an exciting time in education, and we are looking forward to contributing to this groundbreaking journey.
Great read, Ethan. I just started a masters degree program in learning and technology, and your piece is very timely for some discussions I am having in the course about learning design and AI.
Hmmm...well, my Chinese students come to high school from middle school, and by the time they arrive, they already know a very large part of AP Physics "in Chinese," although they need help with challenge problems and English. But you are going to have the "liquid group" stand in a blob and then have the kids run around. Entropy Musical Chairs, so to speak.
This is fine as a game, but it is not for middle schoolers, it is for children of around seven. The US education system doesn't teach squat in middle school. That's where the US falls behind the rest of the world. This is a good example of why.
Hey Ethan, I'm trying to run my own version using Google's Bard. What notebook did you use to run visual code within it? Bard exports the code straight into Google Colab, but there's no visual graphics
Hi Ethan, I’m doing a talk on this topic soon here in The Netherlands. Your input was more than helpful! Thanks a lot for sharing this! Jeroen Krouwels.
I simply have to express my deep gratitude and love for everything you’ve been writing and sharing about GenAI, Ethan. This article is a shining example of what growth mindset plus nonzero sum thinking looks like in action. I’m giddy to experiment with these latest prompts and can’t thank you enough for how much you’re helping me, and so many others, navigate the incredible Cambrian explosion of paradigm-shifting technologies we’re all lucky enough to be alive to experience. Together we go far. Thank you, Professor Mollick, for helping to light the way.
Ethan, thank you for this. I'm the Dean of Students at the NYU Stern MBA program, and I literally posted about this two days ago: https://www.linkedin.com/feed/update/urn:li:activity:7054424505502347264/
That is a cool article. Thanks for sharing
This post is apropos this morning. I teach marketing and entrepreneurship in high school, and your work has inspired me tremendously. We have gone all in with AI this semester and having a great time doing it. I've been creating prompts to use as class activities, but there were a few pain points--students had to copy the prompt exactly, I had no easy way of seeing how the activity was playing out for students, and the not-uncommon access issues. I've been working on a tool to solve these. I literally used it for the first time yesterday, but I'd love it if you and any like-minded educators could check it out: https://edufocus.ai/.
Teachers create prompts that are the AI class activities, these are loaded in a session (different class periods, terms, etc), and then a link is posted for students to complete the activity. There is a log of all the chats in a session and if you require login you can see what each student did and students can access them later.
Just to give a demo of what it does, you can try these two tasks. The first is an activity I made for our 10th grade English teacher in our business academy. He is having students research how we could improve our city using a fictional surplus budget using the United Nations Sustainable Development Goals. https://edufocus.ai/chat/?join=3-24c263
Here is another based on the first activity I did with students with ChatGPT. We used it to discuss if ChatGPT should be banned at the school. This example is also gamified with turns and scoring. https://edufocus.ai/chat/?join=4-5d7142
Thank you for the continued inspiration!
Ethan, I'm like a dog waiting for its treat...I can't wait for when you post! Your insight, knowledge, and experiments with generative AI are what I as an instructional designer want to know about. I want to bring back to the faculty actionable items they can do NOW. I always send them the link to your posts. Thank you for, as @Maciej Workiewicz said, blazing a new trail for us! I so appreciate you and your efforts in getting to know generative AI in the education field and SHARING with us.
I have a question you may or may not be able to answer (others are welcomed to chime in as well): is there any training, in-depth training, out there on writing extensive prompts to garner the generated content to its full potential? Is it a tiered approach, drilling down, basically scaffolding your prompts or are there other ways to prompt that would work better for certain disciplines? Just curious...writing the 'right' kind of prompt will be key for faculty and I'd like to provide how to prompt so they are successful in their endeavors. :)
Thank you for your time...and keep on blazin' that trail Ethan! Can't wait to take your webinar May 2nd!! Heather - ID at a community college
No training yet. I’ll make sure to share as we develop some materials.
Hi Ethan, thanks for this great article. I sent you an invite on Linkedin to be a speaker at an elab.ai in New Zealand (online) . Would you be interested? I am expecting over 100 people:)
TY :)
Thanks Ethan,
i love your post for using Chat-GPT well.
unfortunately, i am stuck in a meta -crisis - that i mentioned on Conor Grennan's linkedin post -
we still need to solve the hardware, mobile network and electricity problems (assuming mid-day meals; a roof over the head is not a problem and of course, no conflict and of course, no religious indoctrination )
until the next generation everywhere is in a safe/secure (no conflict zone) with basic infra - of food, shelter etc. and mobile connectivity, electricity and hardware ... the burgeoning inequity would undermine many of our goals!
i would like to disagree with your use of the word - "democratizing" - while semantically and functionally correct is not useful in the broader context.
I asked it to explain situational irony. Its explanation was fairly bad, pretty much garden variety not-quite-bright middle school English teacher who doesn't get it and has low expectations. Not impressed. I think your standards may be low.
If you think people are going to learn how to program from dumbed down explanations that gamify things, you are sadly mistaken.
I do learn a lot from you and appreciate the blog, but I think you're getting into a gee whiz this is so great mode that is defeating your purposes. To learn, you have to work and to think. I'm glad you got that little 25% improvement in Ugandan something or other. Nice. But dream on if you think that people are going to get smart from gamifying physics and programming.
Still, a little improvement is better than none. However, you seem unaware of Hattie's (sp?) research showing that almost any intervention works well for a while. The kind of changes you're seeing are what's typical for any intervention. Nothing special except the AI label.
I think a few things are getting muddled here, and it is worth clarifying. The three studies I cited were all randomized controlled trials, and 25% improvements in income directly from an educational intervention are huge. That is a major impact on life outcomes, and shouldn’t be dismissed cavalierly. Plus, the experimental approach lowers the chances that they are a Hawthorne Effect (an intervention having an effect just because it is an intervention).
Second, none of the work here is about gamification. Simulation and game-based learning (which I have worked on for a decade )are quite different from gamification, and they can really impact outcomes. But this post focused less on that and more on the way AI could provide multiple ways of trying to understand a topic in an age appropriate way. (The Python game/program was just part of it, but one I found very neat.) We learn best from multiple methods of teaching and working combined, which makes it useful that the system could create a reading and a quiz and an activity and a toy model, etc.
Third, I totally get that the AI screws up and gives weak answers sometimes. Thank goodness, or there would be no room for us. The idea is that the explanations & material that the AI comes up with can assist teachers in doing their job, it isn’t a one-shot replacement for teachers. Got a mediocre explanation? Tell the AI how to improve it. It is much faster (and cheaper) then most other ways of prepping.
I do think Professor Mollick's claim that 8% to 25% improvements are so important is rather naive. John Hattie's book on education claimed that an effect size of .25 was pretty much the minimum for any kind of significance. Basically, intervening tends to produce gains, for a while. So I don't think his experiments are pointing to anything much, at least not those that are around 15% or below.
Thank you for your thoughtful reply. Have those experiments been replicated?
I think gamification and anything else that teaches basic skills is well worth doing. E.g., a game-based app that actually taught fluent reading would be a marvelous thing since the skill builds on itself and has millions of applications. However, I do not think adding "methods" helps. It often hurts. I will give you an example from Quizlet. Quizlet uses flashcards as the base method. It also has functions that let you take "quizzes" in various formats such as clozes, multiple-choice, and matching. These methods are slow and pointless if you want to (for example) learn a foreign language or memorize facts. They are a good match for the silly stuff done in school, but they are counterproductive. A student would be much better off actually working with flashcards for ten minutes to memorize (hear, say, translate, etc.) a set of words or build fluency with them than he would be doing multiple-choice questions or filling in blanks.
So, to generalize from this, I do not think that it is true that "We learn best from multiple methods of teaching and working combined." "Multiple methods" are often quite inefficient, esp. since they encourage students to try this and that and never get serious about anything. One effective teacher using a few efficient methods is quite a bit better than five teachers using five methods each. So no, that is an error.
Perhaps you didn't intend it to be as general as it came across. What I would agree with is that the method that a student will actually stick with is superior to the supposedly more effective method he gives up on. But that's not so much a matter of "multiple methods" as it a matter of having options that let a student find out what he works with best, as an individual.
I do agree that the AI can help mediocre teachers improve their awful teaching, and the fact is that most teachers are mediocre. Some of its explanations are quite a bit better than what you get from most textbooks, especially since it can simplify things.
Thank you again for all you do. I'm learning a lot from you and have so many leads that I don't even have time to pursue them. My own students have made huge gains in writing in two months of working with ChatGPT.
I'm interested in what you have been doing with writing with your students. I started this semester with ChatGPT in my classes (here is something I did back in January–https://blog.gameful.me/robo-redraft-rumble/). I've found that I've replaced most reading in my marketing class with ChatGPT prompts. They are actually reading more because it is in context of a problem they are trying to solve. It's takes PBL to a whole new level.
Honestly, the main thing I have done is to have them dialogue with the AI and push it to criticize their writing and give them suggestions, then have them revise accordingly.
One thing that is becoming clearer than ever is that people who work hard can make extraordinarily fast progress with the aid of ChatGPT, so that the gap is widening, because the fact is, you don't learn anything at all unless you push yourself, and asking for criticism -- be it from an AI, a tutor, or a teacher or a friend -- is an invitation to push yourself. I have a few students who won't even ask for criticism. They just say "please fix this so that I get an A" or some such (they actually give me logs of such dialogues).
Maybe that is better than nothing, actually, but they will be outstripped at mind-boggling speeds by more dedicated students.
I don't know how I missed your reply! Better late than never. I'm right there with you with you with the students who are still looking for a shortcut. I've found that I can prompt ChatGPT to provide limited help. I think of the Blooms 2 Sigma problem that has been discussed in terms of AI lately. If AI can be a personal tutor that gives tremendous gains, it will only widen the achievement between students who choose to interact with the tutor and those who don't.
I do think it is better than nothing. I have students using ChatGPT to help write "Shark Tank" investor pitches. Some students have engaged with the learning, others just wanted to get to the end. Either way 100% of them have a presentation to give. The question and answer session will be the determining factor.
I've been working on a interface that uses the ChatGPT API to allow sending students prompts and logging their chats. The writing tutor might interest you. https://grobots.ai/
Well, I tried it out, but to be blunt, I don't think it was helpful. I wrote this snarky response, and its responses were generic. I think it was giving good general advice for essay writers, but not advice specific to what I wrote.
I wrote:
ChatGPT should not be banned in high schools. Of course not. Why not? Well, let's go back to something I learned from a high school principal: "Don't pass a rule you can't or won't enforce." This falls into both categories. Can't enforce it because students can submit their work to the bot and get useful feedback or even whole essays that are undetectable. Yes, yes: you can tell that they "cheated" in some way if the work is far beyond their capacities. So what? They'll learn to make it worse to cover their tracks. So you can't enforce this rule. But also, you won't enforce it. You won't enforce it because the truth is that the bot is a great help to students, you will find that out soon, and the help it gives is going to make your life easier. No one makes his own life harder. Oh, but what if "you" are the principal? Well, the teachers will just ignore your crazy rule, just as they usually do.
I'll check it out, thanks.
I do think Professor Mollick's claim that 8% to 25% improvements are so important is rather naive. John Hattie's book on education claimed that an effect size of .25 was pretty much the minimum for any kind of significance. Basically, intervening tends to produce gains, for a while. So I don't think his experiments are pointing to anything much, at least not those that are around 15% or below.
I think I could produce a 15% improvement in most teachers' performance by telling them I'd visit five minutes a week, without notice, on two separate occasions; or by giving them some simple instruction (if they followed it) such as "have a one-minute quiz at the end of every class with an answer that shows knowledge of the content"). Really, those kinds of gains are not that hard to get.
Ethan, this is a very good post. You are blazing a new trail here. I posted your prompt to Bing Creative and played a little bit with it. One specific use I see for my own approach is to make programming and modeling accessible to more of my students. I followed your template and asked GPT-4 to explain the Central Limit Theorem to university students and then followed up with this prompt:
write a code in Python that helps to visualize how central limit theorem works. Do this by giving the user a possibility to choose among multiple distributions, for example binomial, Poisson, Beta, and normal distributions. Allow the user then to choose the number of samples and the size of each sample. The code then generates draws from a given distribution and calculates sample means. The code should also visualize the result by showing a histogram of the sample means.
It worked, but the code in Bing, for some reason, can't be copied. I did the same with the basic version of GPT-4 via OpenAI and pasted the code to Google Colab. It works as intended. I remember coding this myself. It would take several hours. Now it is just a minute away.
I also tried more complex code. The first principle I found to work well is to write in normal language (rather than to try mathematical notation). Thus, instead of 2^n just write raise 2 to the power of n. The second one, and I think you mention it yourself somewhere, is to break the code into smaller chunks. But then some knowledge of coding may be required for successful assembly.
Just followed Mark's suggestion and indeed, the sidebar is limited. Copying works when Bing chat is accessed as a website, not the sidebar.
Love this! Your examples of practical applications of AI are more thoughtful/useful than 90% of the writeups from other writers on AI that I come across these days.
It's fantastic to hear that you've launched Wharton Interactive! I'd like to share a platform I've developed for learning Financial Mathematics called 21ifm (https://21ifm.com/). This platform enables educators to design interactive lessons tailored to students' interests, motivating them to engage with advanced concepts and take ownership of their learning.
By using personalized explainers, teachers can assign homework prior to the actual class lesson, as well as post-lesson assignments. Students can then incorporate their findings, their classmates' insights, and the teacher's guidance to complete a comprehensive project. This approach fosters a more immersive and collaborative learning experience.
Pavan you have done a good job. It's true that the design of the site is difficult for me. But I think this is a matter of taste. Good luck in your next works!
I would be grateful if you could share your opinions. Could you please let me know what aspects of the site design made it difficult for you?
I think Bing's explanation of entropy using the analogy of a deck of cards is misleading. Bing states that the deck of cards is a closed system, and being such entropy can only remain the same or increase. However, there is an external agent shuffling the cards, and conceivably the deck can return to the same order which contradicts its own statement. This is obvious when you start with a smaller ordered set of objects.
Yea, that is one reason of many reasons we need teachers in the loop. To both check answers and qualify them to students.
That is all well and good, but it doesn't address the real problem. The lack of standards in tracking and finding the best edtech. The lack of interoperability. Have millions or billions of people able to create educationally material on the fly is fine but you still have a data overload issue. What we need recommender and performance tracking systems so that learners can navigate this sea of materials. Because the real democratization is not going to be anyone being able to produce materials, it is going to be anyone being able to educate themselves with only tools and no educators. It is chat GPT as the teacher.
Dear Ethan,
Thank you for the enlightening article on the transformative role of AI and EdTech in democratizing education, specifically the powerful capabilities of GPT-4. At Eureka Simulations, we share your enthusiasm for the impact these technologies can have on learning.
In our work, we've seen first-hand how simulations play an instrumental role in creating engaging, interactive learning environments, bringing theoretical concepts to life in an experiential way. Similar to the revolutionary leap you describe with AI, simulations enable learners to engage with complex ideas and scenarios, enhancing their understanding, and allowing them to tailor their learning journey to their needs and pace. They also provide educators with a powerful tool to deliver personalized, impactful education at scale.
Your emphasis on democratizing access to quality education echoes our mission. The power of AI, as demonstrated by GPT-4, combined with immersive simulations can significantly enhance the way we learn and teach, making quality education accessible and tailored to all, regardless of location, language, or background. It's an exciting time in education, and we are looking forward to contributing to this groundbreaking journey.
Best,
Joaquim Virgili
Founder, Eureka Simulations
Great read, Ethan. I just started a masters degree program in learning and technology, and your piece is very timely for some discussions I am having in the course about learning design and AI.
Hmmm...well, my Chinese students come to high school from middle school, and by the time they arrive, they already know a very large part of AP Physics "in Chinese," although they need help with challenge problems and English. But you are going to have the "liquid group" stand in a blob and then have the kids run around. Entropy Musical Chairs, so to speak.
This is fine as a game, but it is not for middle schoolers, it is for children of around seven. The US education system doesn't teach squat in middle school. That's where the US falls behind the rest of the world. This is a good example of why.
Hey Ethan, I'm trying to run my own version using Google's Bard. What notebook did you use to run visual code within it? Bard exports the code straight into Google Colab, but there's no visual graphics
Hi Ethan, I’m doing a talk on this topic soon here in The Netherlands. Your input was more than helpful! Thanks a lot for sharing this! Jeroen Krouwels.