OpenAI Forum
+00:00 GMT
Sign in or Join the community to continue

Harvard’s AI-Enhanced Classroom: Revolutionizing Learning with Custom GPTs

Posted Nov 21, 2024 | Views 2.5K
# Higher Education
# Future of Work
# AI Adoption
Share
speakers
avatar
Siya Raj Purohit
Education GTM @ OpenAI
avatar
Jake Cook
Faculty @ Harvard Business School

Jake Cook teaches the popular "Digital Marketing Workshop" course for second-year MBA students at Harvard Business School, covering key areas such as SEO, paid and social media, Amazon strategy, influencer marketing, analytics, and basic machine learning applications with Python and GPT-4. His course also explores emerging trends like generative AI, privacy challenges, and personalization, often enriched by insights from leading industry experts. Learn more about how ChatGPT can be used by students to prove mastering digital marketing concepts and how AI can reduce plagiarism here.

In addition to teaching, Jake actively supports academic research and case study development, contributing to Harvard’s Executive MBA program and collaborating with Harvard Business Review on cases focused on data strategies in AI-powered e-commerce.

A lifelong entrepreneur and educator, Jake co-founded Tadpull, where he works with data scientists, software engineers, and digital marketers to help brands achieve sustainable growth through data and AI. With extensive experience in e-commerce, analytics, and digital marketing since 2007, Jake also pursues personal projects in machine learning, direct-to-consumer ventures, and the digital nomad lifestyle.

+ Read More
avatar
Natalie Cone
Forum Community @ OpenAI

Natalie Cone launched and now manages OpenAI’s interdisciplinary community, the Forum. The OpenAI Forum is a community designed to unite thoughtful contributors from a diverse array of backgrounds, skill sets, and domain expertise to enable discourse related to the intersection of AI and an array of academic, professional, and societal domains. Before joining OpenAI, Natalie managed and stewarded Scale’s ML/AI community of practice, the AI Exchange. She has a background in the Arts, with a degree in History of Art from UC, Berkeley, and has served as Director of Operations and Programs, as well as on the board of directors for the radical performing arts center, CounterPulse, and led visitor experience at Yerba Buena Center for the Arts.

+ Read More
SUMMARY

The OpenAI Forum hosted an engaging session titled "Harvard’s AI-Enhanced Classroom: Revolutionizing Learning with Custom GPTs", featuring Jake Cook from Harvard Business School and Siya Raj Purohit from OpenAI. Siya provided insights into AI-native universities, showcasing how ChatGPT transforms education at individual, team, and institutional levels. Jake presented real-world applications of AI in his classroom, emphasizing creative experimentation, personalized learning, and the power of play to unlock students' potential. The event highlighted innovative uses of AI, including custom GPTs for dynamic teaching and adaptive learning, fostering both efficiency and critical thinking. The discussion underscored AI's role in reshaping education and left participants with actionable insights and strategies for adoption.

+ Read More
TRANSCRIPT

So we do have a packed agenda tonight, as usual, folks, and we're going to start with my colleague Siya, but before that, I want to remind everybody of what I always want to remind us, and that is we record these events, so if during the Q&A you want to ask your question live by raising a hand, please remember that you'll be on camera.

One of the other benefits of us recording these events is that we will publish them for on-demand viewing in the wake of the event so that you can share them with your peers and with your colleagues and with your friends, invite them into the forum afterwards or re-watch the event and remind yourself about some of the content or perhaps who we learned from tonight, so I hope that none of you mind that. If you do mind and you don't want to be on film, then just type your question in the chat versus raising your hand and I won't call on you live.

So tonight we're experimenting with a new format. We're excited to introduce the webinar. It allows us to host a much larger audience, but still allow our members to ask questions and meet the speakers live. Your camera and mic will be off by default, but you can still engage by typing in the chat, submitting questions in the Q&A, reacting with emojis, and at the end during the Q&A, if you'd like me to call on you personally and spotlight you, then just raise your hand, please. So that is a little bit different than traditionally.

Traditionally, we either move into a meeting format and we can see everybody's faces or you have to type all of your questions in the Q&A, so this is a little bit more of a hybrid.

All right, let's get this party started.

So I'm Natalie Cohn, your OpenAI Forum Community Architect, and I like to start all of our talks by reminding us of OpenAI's mission. OpenAI's mission is to ensure that artificial general intelligence, AGI, by which we mean highly autonomous systems that outperform humans at most economically valuable work benefits all of humanity.

Welcome to our forum on Harvard's AI-enhanced classroom, revolutionizing learning with custom GPTs. I'm thrilled to host this session and explore how AI, specifically custom GPTs, can reshape the educational landscape and empower both educators and students.

Today's session is about how an educator at Harvard is using AI to bring learning to life in the classroom. It's part of a speaker series that we hope is making our higher ed faculty, lecturers, and graduate student teachers in the forum become more comfortable and better equipped to adopt AI in their pedagogy and even their administrative purposes. AI is for all of us, and we believe it's very helpful to learn from one another. So I hope that you're finding this series useful.

Tonight I'd like to introduce Jake Cook. Our special guest teaches the digital marketing workshop course at Harvard Business School focusing on SEO, social media analytics, and applications of AI like GPT-4 and marketing. He co-founded Tadpole, where he helps brands achieve sustainable growth through data and AI and collaborates on research and case studies with Harvard Business Review. A seasoned entrepreneur and educator, Jake brings over 15 years of experience in e-commerce, analytics, and digital marketing. We'll hear from Jake in just a little while.

Siya Raj Paroit is an education leader at OpenAI. Siya helps grow ChatGPT EDU across higher education, is deeply passionate about leveraging AI to foster equitable learning environments. Please join me in welcoming Siya to the forum.

Thanks, Natalie. Hi everyone. Great to see so many amazing people here in the forum tonight. It's so good to have you, Siya. I'm going to go ahead and just pass the mic to you.

Sounds good. Thank you, Natalie. So hi everyone. I work on the education team at OpenAI, and I spend a lot of time working with universities to think about what the future of education looks like. So a big part of my job is helping develop the vision for what AI-native universities can look like and helping support universities in that transition to becoming AI-native.

So I'm going to provide our current thinking around what an AI-native university looks like and how we're enabling transformation at different types of higher ed institutions. ChatGPT EDU, which is our enterprise product designed for education, launched in June of this year. So it's about five and a half months old. And the reason it came into being is because we realized that the number one use case for ChatGPT was around teaching and learning.

So that's no surprise, like a lot of us are knowledge workers and basically use that every day to just perform better at our current jobs. And students are using that to learn more effectively as well. This product launched in June, and we have been working with a number of higher ed institutions that are adopting this for students, faculty, and staff.

When we think about AI transformation in the education space, we see it happening at three steps. The very first step is individuals that are using this, using ChatGPT to improve their own work. So one professor told me that he writes so many letters of recommendation that he created a custom GPT that lets him create new letters of recommendation using his voice and past examples in a faster way.

So he uploaded his previous letters of recommendation, and he drops in a paragraph about the student that he wants to write about, and it creates a new letter of recommendation in his voice. And it saves him so many hours every month. So these kind of individual challenges and tackling them with AI is the very first step of transformation. The second step is teams and departments that work together to solve their common goals.

So one university told me that it takes 40 people several weeks to assign which course goes into which classroom on a campus. It's a very tedious activity of mapping out how many students in the class, what time is the class, which rooms are filled, and now ChatGPT does it in a few minutes. And this is how teams work together to solve at the team level.

And the final level of transformation is at the organizational level. This is when organizations like universities and enterprise companies create a lot of different AI touch points across campus to make it much easier for their community to engage with the knowledge of their campus in a very deep way. So what we envision AI-native universities to look like is to have multiple AI touch points from when a student comes to campus for the first time and goes through orientation, to have an orientation GPT.

So they can ask questions like, how do I change my roommate? Or where's the best pizza place in town? And basically just understand that knowledge of orientation more deeply. Then students will go into classes that have custom GPTs, and they can talk to the knowledge of their two-year MBA program or whichever program they're in much more deeply. They can ask questions like, which CEO handled layoffs well? And get the exact examples from what their professors teach.

They'll go into student clubs that will have the history in GPT form, and they can basically talk to it in very dynamic ways. And finally, they'll have career services that's much more equipped to help support them by providing information such as practicing interviewing with a recruiter, practicing interviewing with a partner, and just help them get better at back and forth exercises.

Right now, ChatGPT is really good at three things. One is around information retrieval. So being able to pull from this repository of information that exists and get the exact insights a student is looking for. It's good at back and forth exercises. So when you think about negotiation or any type of thing that needs a simulation, which Jake will talk a lot more about today too, it's really good at that. And then finally, thought partnerships. So being able to go back and forth and think more deeply.

So these are the types of things we're thinking about and how we're working with professors and university leaders to enable transformation in education. If any of you are interested in learning more about ChatGPT or how we're working with universities, feel free to send me a message on the forum or over LinkedIn. And now I'll be passing it over to Jake in a minute. Just for context, Jake and I met through his student Isabel, who I met on the Harvard campus a couple months ago. Isabel told me that Jake has some of the best applications about how to use AI in the classroom.

He's rethinking both curriculum, but also AI applications in the classroom. So I'm very excited to pass it over to Jake and have him take over now.

Wonderful. Well, thank you, Siya, Natalie, and Caitlin. It's been so much fun putting this together. My hope this evening is there's some good ideas here. We can have some good conversations around. I'm going to go ahead and share my screen. I don't have the RAM to process and present at the same time. So I won't necessarily see chat.

but we will get to the Q&A as we get through it. But if there's something really burning, Natalie or Caitlin, feel free to interrupt me and I'll do my best if we're in the moment there to answer it. So here we go.

All right, everybody seen that on your end? Hopefully, perfect. Okay, well, really quick a bit about me. My name is Jake. I'm the co-founder here at Tadpole and I've been an adjunct professor since 2007. I had a mentor from Stanford who helped start the d.school who told me to drop out of a PhD program, become an adjunct if you really wanna teach. And so I followed his advice and here I am. And so a lot of what we're gonna talk about today is kind of the hands-on stuff that I've been fortunate to kind of develop and work with as we go through everything.

So we have kind of three key takeaways this evening I wanna just cover briefly. One is this kind of mindset. So kind of re-imagining your talents, which I think is a really fun thing that AI is starting to unlock and I'm seeing that in students firsthand. Number two is habits. So kind of this idea of play as a way to kind of trick students or colleagues maybe that don't wanna play with this stuff or see what's possible and how you can unlock those aha moments with prompting. And the third thing is kind of how you can unleash your creativity with custom GPTs. And I've got a couple of different examples we'll go through that as well. And I always come up with a little rhyme for students, but I try to remind them each day in some small way, make time to play. And that's, I think a real fun way to kind of unlock these frontier models on what's happening.

Let's start with a little bit of data. Like what are attitudes on AI? This was something I saw on LinkedIn over the weekend from their research group. And when you look at this kind of, gaining AI skills and helping my career progress, surprisingly, when you look at Gen Z, they're lagging behind the Gen Xers and the millennials by quite a bit. And they're not too far off from boomers. So this idea of younger folks adopting AI, what I find in the classroom is kind of like, well, I've got this, I don't need to really worry about it. I'm kind of crushing AI. And so, and it won't have a big impact on my career perhaps. But this gets to kind of our first key takeaway, I think with students, when we look at these things is kind of re-imagining your talents. And I talk about, I'm not a blank until now. And this came out of office hours this fall where students would show up and say, well, I'm taking this digital marketing class, but I'm just not technical. I'm not really good at creative things. I'm not good at writing or copywriting or doing images and things like that. And I'd always kind of come back to them and say, well, with AI, that's until now. So I always, this is one of my favorite quotes from Steve Jobs, what a computer is to me, it's the most remarkable tool we've ever come up with. And it's the equivalent of a bicycle for our minds. I think if Mr. Jobs was still with us, we would call this kind of the self, AI is kind of the self-driving car, perhaps even a flying car that we've all been wanting. So the biggest thing when I noticed with generation AI as I talked and work with them is this kind of rethinking their competencies. So again, this, I'm not analytical, I'm not an entrepreneur, I'm not a coder, I'm not until now is a big one. And so when we get into working with the students, the best way is to kind of get them playing with something. So here's a quick example where students might say, I'm not technical or I'm not a coder, and we'd hand them a data set. And this is a date of just a website, traffic to the website, maybe, or revenue and all these types of things. And they would say, oh, I go put this in Excel and try and maybe build a line chart or something like that, but I'm not technical and I'm not a coder. And then I would say, well, it turns out with chat GPT, you can actually use, you can write Python. We can run this in Google Colab in the cloud. And lo and behold, we can unlock a machine learning algorithm to predict revenue to a website. And you don't need to know the API, you don't necessarily even know Python, you can debug it and work it live. And so this becomes a superpower where we take tools like coding and Python from the high priests of the internet, and we kind of disperse it among those of us that maybe didn't feel comfortable with these tools as well.

So here's a fun fact. This is not an academic study, but I have a strong belief that 99% of all students love field trips. From kindergarten till graduate school, we all love field trips. I think a lot of education is missing a little bit of an element of play. And so we came up with these ideas of kind of field trips in the class. And I think this is one of my favorite ones was this kind of idea. It's Friday night, go home, take a photo of your beverage rack of choice at Whole Foods or Trader Joe's or your favorite grocery store, and then prompt it. And so I wanted them to do something in this example where they had to put a little bit of skin in the game. There's a little bit of risk involved with this where they had to kind of really, you know, they're gonna spend not a lot of money, you know, anywhere from five to 20 bucks, maybe or whatever, which isn't insignificant as a student, but enough to like, okay, I wanna put some risk, I'm gonna trust the AI here with a little bit of a, you know, risky choice. And so this is one in typical Boston fashion, we put seafood on everything. So there's, you know, pizza night with clams, act as a wine expert, you know, what do you get? And what we see is it kind of comes back with recommendations based on this photo. And then when you ask it to justify it, and now we get into kind of like, why did you pick this wine? Why was it that? And then the next week when they do it again, ideally is they find out, well, I didn't like your picks. They go to that same chat, they boot it up with a different image and say that pick was really good or really bad. And through all this, what they're learning is basically reinforcement training of the model off a data set they care about with a little bit of risk involved too. But we start with something really low risk to get people playing with it on that side of it. And what we did with that, this is another example for playing is, you know, go home tonight and take a picture of your photo. This is a picture of my photo. I love Korean food and kimchi. And I'm saying, okay, and I love David Chang and Momofuku. So I was like, all right, act like this based on all this stuff. You know, this is important. I'm gonna get fired as a chef in my house. And there's some Easter eggs in here I'll unpack in a second. And it comes back with a kimchi and spicy chili crisp stew. So we're gonna make a recipe. You know, I'm gonna spend some time. So there's investment of time in this, but it's not the end of the world if it bombs out, but they're starting to see what's possible. And in these little exercises, we're starting to learn actually multimodal prompting, where we're taking an image and text and we're kind of shoving it in there to see what's possible in a very low risk way. So that's how I can try and warm students up or clients that aren't really familiar with AI, maybe they're scared of it. You know, we tend to fear what we don't know or understand or experience. So building out these little micro moments with little investments, but a little bit of skin in the game, I think is a great way to get people kind of over their fear of that with playing as well.

Okay, so back to the field trips. We've had field trips to the grocery store. This was a live field trip we did with Harvard Art Museums. And the business challenges, you know, it's a free art museum. So how do you get people 20 to 40 to sign up and join the art museum? I mean, it's free, why would I pay money? This is kind of a revenue strapped age range as is. So that was the first initial challenge, but then we added a twist to it. What's the AI challenge? So how can we generate creative to advertise to this or images and things like that for Instagram that we might advertise with this? The students had their homework. So part of their prep for class was to go to this incredible art museum and find art that matters to them, stuff that resonated with them that they thought was cool and interesting.

So they would go to it, find the art they liked. They took a photo, they made an ad with Gen AI, and they had to record their prompts. And then they would submit those back. And some students were kind enough to actually let us show these live in class. So it's pretty fun. So this is Mimi Sachs, who is one of my absolute star students I've had the pleasure of working with. And she's generously donated a lot of her work from class tonight to kind of put some flesh on the bones here. So this is a really great example, I thought, with Mimi where she was using multi agents here. So we're using Adobe Firefly and you'll see what she did. So she basically went to Firefly, said, hey, here's a bunch of prompts, classy and polished art campaign, encouraging people to become members, it's free, be playful and all that. And then what was interesting is then she went through and had chat GPT critique it. What are playful expressions and things like that. And so basically give me some ideas on that, on what I can do using two AI models to play with, which I thought was really brilliant.

And look at that, isn't that cool? I mean, we kind of, I thought this was fantastic. You could certainly run maybe an Instagram campaign against this and see, it's got a fun quirky thing, looking this good, doesn't come cheap, which has a great iterative piece on that. And then it could call the action and the logo of the brand there as well. So this was, I thought, one of the best examples from the field trip. So thank you, Mimi.

Okay, key takeaway number two. So we get people to think about playing as kind of a hook to get them to, messing around with what's possible. And then we're talking about this power of play when we combine it with this prompting framework, which I call ratty. And this ain't Google. So this is how we start class. I give them a very kind of interesting common digital marketing challenge.

of how we might buy something kind of complicated on the internet. So we might start with a Google search and we're on our desktop. Then we look at reviews and then we went to a tech review site, and we're basically looking at Google shopping ads all the way through over like say nine days.

And that's basically what the class I teach is about. Where would you put your time and money? Should you spend more on search engine optimization to rank better? What about something like creating YouTube videos? How does this all work out, right?

And so we kind of run, we wrestle with it, and then we kind of start to ask the AI, what would you do? And this is what was really fascinating to watch with my students.

And a quick, I guess background about Harvard Business School, we're very blessed that the school is very proactive on adopting the technology and the curriculum. We have kind of our own private GPT and there's sort of sandboxes we can play with all sorts of different models without anything leaking.

But this class, Class of 2025 that's gonna be graduating this spring was the first class to kind of come through starting last fall where they had access to this kind of out of the gate.

And one thing I was a little, I guess, surprised about was maybe just going through some fundamentals and letting them play with some of the stuff and showing them live in class how to do some of this stuff was really fascinating.

So when we went back to this and started playing with, okay, now we can use AI to answer these marketing questions, they would kind of treat it like Google. So how do I market Nest thermostats? You know, so that's why I sell thermostats online, digital marketing for Nest, very basic short phrases we would all type into Google as well.

And that was when the big moment went off in my mind, like, oh, we should actually pause and walk through the fundamentals of prompting to unlock the magic of these large language models. So hopefully this is something of interest for those attending tonight, but the way the R stands for define the role you want the GPT to play, the assignment states specifically what you want done and ignored.

Deliverable, ask for the format you want this back in. And sometimes with citations, if you're doing things that are kind of, you know, more specific within the academic or research realm.

Iterate, and this is something I think a lot of students and people kind of struggle with is just pushing it multiple times, or even just hitting on that first prompt, just ping it again, see what it does back to that plane.

And there's exclamation point, tell it to focus and this is important. Whereas with some of these large language models, we don't really know why, but there's some funny examples I've seen online where, you know, evaluate my code, I'm gonna get fired if this is wrong. All of a sudden it pays attention.

So the kind of Mad Libs or way to kind of fill this out is actually a blank with blank credentials and who is blank.

Assignment, think about blank, blank, but not blank.

Deliverables, blank, blank, blank.

Iterate, I like this because of that, but not that because of this.

This is important, pay attention, I know you can do it to kind of tee it up.

So that basic prompt of what's the best way to market a Nest thermostat turns into something much, much more in depth.

So we're going through, and this is what you're kind of showing them in class, like, oh, this is actually a very robust prompt that we would type out.

And what's fascinating is you start seeing device changes. So when people are on their phones, like all of us, we don't like to type a lot. So I think Mimi and I are potentially exploring some opportunities to research ways are people prompting by device and time of day.

So when you're on your phone and maybe it's late, you don't wanna spend a lot of time on that. But when you're on desktop and you can type all this or use an audio, I use that a lot too, to translate these prompts, you can go a lot deeper on that.

But that kind of ratty framework gives you something, like a checklist as you go through it as a habit to think through that when you prompt it as well.

This was really fascinating. And then they were able to kind of compare in class, like their little, what's the best way to do digital marketing for Nest with this really robust thing as well, which was really fascinating.

Now we're gonna get into some fun stuff here going through custom GPTs.

And my hope is we'll have plenty of time for Q&A and we go a little bit deeper on what's possible on that as we get into it here.

So kind of the big idea that dawned on me was data exists everywhere. And now with AI, we just simply have to collect it and bring it to life.

So this conversation we're having right now, data that might be online in various forms on YouTube, it just kind of exists everywhere.

And then when you can kind of collect it and focus the models on that, you can really bring it to life in some kind of fascinating ways.

I'm gonna split this section into kind of two parts.

One's gonna be like more for those of us that maybe you're teaching or educating folks around AI.

And then the other one will be more about like actual uses, like very specific uses from class on that side of it.

Okay, so the professor one, real-time feedback, what you can do, student will be more like custom focus groups and a personal tutor and intern as well on that side of it.

Real quick. Oops, hold on one second. I'm gonna do, excellent. Make sure that's loading real quick. Yeah, perfect.

Okay, and the last one was gonna be teaching cases with data for those that are interested here real quick.

Data's everywhere, right? And so one thing that I've noticed is building a custom GPT for class. In full disclosure, I'm very lucky to work with some talented folks. And I, at Harvard has this thing called the Christensen Center. And Willis and his team put on these workshops for professors that are interested in kind of the art of teaching, which is really, really fun.

And Jeff Busching has done a lot of really interesting work in his class, launching technology ventures. And based on that and some other stuff, I built out this kind of custom GPT and I'll show some ways I use it in a second.

But the biggest thing is grabbing case studies that I might write for the class. And the way Harvard works is we give them a story or a scenario of a business situation. And then the students are supposed to debate it.

So I think to see this point where we have simulations and students can debate something back and forth, this is a great use of that as well.

I also have teaching notes, a lot of stuff I remind, just in various formats. I have that in text format. I have a couple of books I use and have kind of written myself that I feel like are kind of more up to date and I update frequently.

So I can kind of build that into the GPT. Glossary of terms, any topic tends to have a lot of jargon and it's easy to forget, especially when you're lecturing or you have guest speakers, just the barrier on the languages and the acronyms and how students get overwhelmed, right?

And I picked this one, for example, like what is ROAS and why does it matter?

But this is a good way to kind of break down that. And the students can have this live in class. So they hear something in class, they can quickly query the custom GPT and they can kind of keep things moving as well. Or if they forget something and it's built, I don't remember something from two, three weeks ago, it's a great way for them to review live as well.

What it includes is a live QR code in class. Students can scan it on their phone. So I have a Bitly link and they can just boot it up and get right into it, which is helpful.

And this is the big hack I've found is I have a Calendly invite where students can go and book certain times in my calendar. And in part of that, asking them with the link, hey, make sure you go to the GPT, the custom GPT, ask at least three questions prior to office hours. And anything you feel that it didn't answer or you were a little bit lost on, then let's come to the office hours.

The class has over 70 students. So to try and do office hours with that many is a little daunting. And so what I find is this lets us really elevate the office hours conversation.

Aside from the existential career questions that we all have at different times in our lives. But when we get into the digital marketing stuff, it's fun because we're not going through basics or fundamentals.

We're like, okay, I understand how paid ads work and why SEO matters and these analytics. Now, how would I combine those to launch a startup to get seed funding? And that's been really, really fun to watch that time get much more valuable for both of us.

So how does this work? I try to grab some screenshots and very specific things and working with some other colleagues and friends, I think these custom GPTs can seem a little bit daunting.

And then when you realize what it is, it's pretty amazing.

When I watched the launch last November, I remember exactly where I was. I was a big fan of Sam's, how to start a startup class from Stanford. And there's so many great lectures and just really good advice in there for those that are interested in entrepreneurship.

And then he kind of finished the demo with like his office hours using some of that content from the course. So that really was kind of a clouding, a parting of the clouds for me.

And so what I've done here is gone through and like basically built some prebuilt prompts that you might ask, how do you measure customer lifetime value? How do companies fail at commerce? What data is needed for AI and marketing to kind of tee it up under the configure?

And then I organize every one of these by class for easy version control. About 70% of my class, I have to kind of throw out and rebuild from scratch because the space moves so quick in digital marketing and analytics and AI.

So having like each section mapped to the core syllabus and each text file kind of maps over to that, makes it real easy to go through and visually look it over.

Even using a.

to like, you know, edit with it as well. The other thing is doing teaching notes or maybe on a verbal, like maybe I'll have the deck and I'll go through it.

I'll take that audio file, put it into text, push and pull with it, and then I'll update it into here as well. So that's been really helpful versus a giant text file. It's getting better, but I've noticed like the more it's kind of chunked up, it seems like it gets better answers and returns quicker as well. So I think there's also something on that side as well for performance.

And then you can fine-tune these. For those who don't know, you can fine-tune these whether you want to like create images or if you want to go out and search the web, code interpreter and data, if you want the students to be able to like work with some coding off some data sets. And I'll show you an example here in a second as well. So it's kind of the fine abilities.

For very specific stuff, it's very technical. I don't want them maybe getting going out and getting lost on Reddit or Google and getting bad advice. So I might turn off web search for a very specific one about a very specific topic, just based on what I'm seeing in industry and what needs to be done there.

Training data. Okay, so the fun thing with machine learning is you need data to train on, right? And one idea that kind of dawned on me this fall was like, wow, what if you treat the class like an algorithm and you got to train it? And how would you get data to train the algorithm from the students? And so this was a big unlock for me as a professor was getting feedback in real time. So basically in class, I have them scan a QR code at the end. I try and stop at five minutes before class is over and I have them scan this and I ask them three simple questions. What did you find interesting? What did you find confusing? And what would you like me to do different? If you say better, by the way, students always have something that you should do better. And Harvard students aren't shy on being very blunt. I'll talk about that in a second. But what would you do different or better is a tension there, right? Ask for different. And then basically you have your own data sets. They scan it, it goes to a Google form. I download that CSV after when it's all done. I can now analyze each class specifically and the overall arc of the class as we go through the semester. And then I can adapt off the data that I'm getting in real time. And now when the class is done, I can also compare year over year off my course evals. And for those of you that teach, sometimes the course evals are a little more broad and there's not really good nuggets on did this topic resonate or what was really good about this one? It's more like, is a professor available? Do they care about the students? I'm more interested in like what worked really well in SEO or data science. What do we need to go deeper on? So having that very granular is really helpful.

So this is what it looks like coming out of Google Forms, just the raw CSV file. And then this is an example, basically, hey, how did the Gen AI and data science class go? And this is telling me students appreciated it. They need deeper preparation. So maybe I failed them a little bit on getting good examples in the prep. And then it tells me, and I think this is most important, is it tells me where to kind of start class. So for those of you that teach, sometimes you have this curse of knowledge and you're just marching through the semester in the course. You might've lost them on class three and you're on class six and they built upon all these topics and you don't know it. And also I find students don't always want to raise their hand and say, I'm so lost. Or in a room, there may be other peers are nodding along and then you get people feeling like, oh man, I don't get it, but I'm not gonna say anything. And the only way you have that feedback loop is in office hours. And if someone comes to office hours and you probe them, then you can kind of get some clues. But this was a huge unlock for me as well.

Like, oh, okay, wow. Maybe spend more time in prep and I can ask how many students felt this way. Give me a number, like 30% of the students felt this way, 80%, what does that look like? You know, all that. So that was really helpful.

Another, I guess like thing, this was also a big aha moment for me as a professor. When you read student feedback and it's anonymous, it can really sting at times. And when you spent months preparing for a class and writing a case and all this work and the students kind of skewer you, out of 70 students, you remember the three or four that maybe were a little bit more blunt or raw in their feedback. And that kind of can stick in your head. And AI, having AI go through all this and summarize it for you is really helpful because you don't tend to like, oh man, everybody feels that way. It's like, oh no, this was only a few that felt this way. The other fun part with this is I teach class from 11.50 to 1.10.

And so we had a great example as we were going through this where students were saying class sucked, was super boring.

And it was a really good teachable moment to say, you know, maybe you're all our hangry because you haven't eaten and you're going big days. But I said, this is a good example of how AI works.

If I don't have any good training data on this, I don't know how to like adapt. So if you say class sucks, I need to know why it sucks. Why was it boring? And it was kind of, you know, everyone got a chuckle out of it, but it was kind of a good example of like, hey, you know, I need to know the why behind this.

And then I can, we can, I can learn from the AI and we can identify the themes on that as well beyond just the basic sentiment. So that was a fun way to kind of turn something into a teachable moment just basically how AI works.

This is another quick example too. We have these cases at Harvard. We begin these little business cases and this is one I wrote. It's a three-part series and it's like, hey, should a company invest in building out data? Should they just sell their stuff on Amazon and they make money, but they lose data in the process? What do you do? And this is kind of all hypothetical. Where we could say, oh, they should, they shouldn't. And we have this fun debate. But what's really fun is with the custom GPT, now you can turn this to light and say, well, let's see this on the other side if they did it. And so what I'm able to do is basically build a synthetic data set with the custom GPT from all these various sources. So it's a company, it's a company, it's a company. So I can, so it's a fictitious company, but I can go online and find basic user reviews on Amazon or different B2B sites or whatever, use that as kind of seed content and have it over time build out this really big, basically CSV file of a data set. And it's kind of the magic wand of what you wish you would be able to show students. And I'm checking in on what's realistic from industry, but we have everything from first order date, customer's net promoter score, how much time is spent on a website. I mean, all this amazing stuff in one place as well. And then we're able to kind of push this a little bit more in the actual live GPT. So we're having it visualize this. And this was what was really, really fun in class because it produced a hallucination.

And I remind the students, you got to treat AI like a very eager intern that stayed up all night and drank Red Bull. It's gonna just try and give you answers. It's gonna try and please you. The tone in which it comes back can make you feel if you're not prompting and pushing it that it's very accurate, it's overly confident. And it was fascinating because live in class, we had a hallucination where the students were like, "Wait, that distribution is way off. There's no way that's right." And we say, "Okay, cool. How do we validate that?" "Oh, well, here's the Excel file. Now let's go back and..." "Oh, it turns out we do have a discrepancy. Okay, this team go prompt with a little bit different nuance." "Does that match reality from what Excel's telling us?" Which was really, really fun to see them kind of go through that and have that aha moment in class.

And then the other part too, with some of these custom GPTs is you can kind of contain it around your topic. Then you can also find the edges of the boundary conditions of where it may break down and have that hallucination for the students live, which is really, really, I think, important.

Okay, that's some examples from what I use in class. Let's flip now and talk about some students. We start with the midterm project. It's a tough economy. And so I was thinking like, what could we do to create something for the students where they would learn some digital marketing and fundamentals, but also have an asset they could use after they graduate. So what we had everybody do is spin up their own personal website. 50% of my students want to start their own company. So they can either build a site, whatever they care about, a personal brochure website for themselves, help them get a job or create their startup. And that was a really fun way to kind of walk them through how to do that. So really quick, the ones that are interested in startups, we had them go through and basically do customer development conversations. So if they're talking to a potential customer, they're doing prototypes and testing, with permission, have them record that audio and turn that into text. If their competitors have stuff on YouTube or they've seen stuff on YouTube, grabbing that, turning it into text, comparing competitors' websites, and then have it write website copy and things like that. And so this is an example called MarketMinder. Custom GPT. And it's like, what are the top three most important things to make a purchase decision? And then they're seeing very specific things come back, mentioning competitors and things like that, and what they have and what they don't like, and stuff like that as well. So that becomes this ever-growing custom GPT as they're building out their startup. For those that were building their own website, this is Mimi. We gave this example of kind of using the AI in the midterm as a critic. So they would build their own personal website. Some students, one student, Juan, built his entire website purely with AI.

So the images the copy he spun it up the design look and feel using prompting within I think it was Wix So he was like literally the whole thing had been done with AI to the last bit of it What I had him do And I did this last year as well as I have them use the ratty framework to go through to evaluate their website So a lot of them were saying like I'm a recruiter. I work at uber You know act, you know deliver me, you know about my about page what's weak in my positioning? How should I stress my internship experiences? You know give me specific recommendations page by page as well, which is really interesting What I do is record a loom video and this is what's really fascinating as you start to see them thinking critically For two reasons one is it's something they really care about they've got skin in the game again So that that prompting they want to get it, right? So the loom video and then you can identify them as they go through the iterations for the screenshots You're seeing the critical thinking come in the kind of human in the loop as they work interact with the AI and then yes What do you agree with what I disagree with and they're kind of going through and say no this isn't right Actually, they have a good point on this which was fascinating to see as well Another student use real quick is This is a Mimi as well We're basically can the question was can you 3x in three years and the AI challenge was, you know, can you elevate your work? So the question is because company grow three times in three years and I give them a ton of data That's way beyond what they could probably analyze in a normal You know sprint of a week for the final and so it's got add text data add performance data Customer cohort data just lots and lots and lots of data on this as well And it's supposed to kind of come up with an analysis So this is Mimi's stuff, which I thought was great and a huge shout-out to Mimi for allowing me to share some of this live It's better when you see the actual students. They're the heroes here But this is a really good one So she grabbed some which is a tool use semrush And she type of has spent so much time as a consultant making these and that's so cool to see like that's where the aha Moments are happening. I was like, oh it it's an efficiency Gainer here. And so this is something she created you see the prompt acting like a strategy consultant Maybe competitor grid including all these things what the road should be and all that kind of stuff as well, which was really really cool

This is another one I thought was really fascinating They add it looked in the topic by ad With all the ad copy and her own custom one and asked it to turn an Excel file Which represents ad copy and column corresponded. So we're now we're moving some analytical stuff here using some stuff with Excel And then the last piece is they took all that and then created an actual image for an ad So we had truly data driven creative off of this ad set. So what were the best ads that perform? What do they have in them? What were those images looking like? What was it speaking to and now taking all that and prompting an AI to build an ad so we've gone from I'm not a coder I'm not technical. I am a coder and I am creative and which was really really cool as well So I open up for some Q&A really quick three key takeaways mindsets I think we can all reimagine our talents with this which is so fun Our identities of I in this or I wasn't that or you know, I'm not technical. I'm not a coder now you can be This habit of kind of finding fun things to play with AI and then using the prompting to kind of unlock the magic and Then these methods you can have kind of unleashing creativity with custom GPT's as well And as I say in each day in some small way make time to play That is my contact info if you have any questions, and I think we can probably I'll flip over We can open it up to some hopefully some Q&A if you may have some some interest. Thank you Jake that was awesome. I'd also like to bring Siya back into the spotlight Oh Perfect so I see so I've got some Q&A up here Do we just kind of run through these really top-to-bottom Siya? Does that make sense? We'll start with Siya because Siya wants to kick it off with a question and then I'll call on the folks with their hands raised For Q&A if that works for you. Perfect. That's great

So Jake a lot of faculty are thinking through how they should update their curriculum For classes to just prepare students for the professions that are enabled by AI. How should they think about this? I think you know Students whether you call it cheating or not, I mean once they get their heads around this They're gonna they're gonna be using it And so the research is pretty clear It's pretty hard to tell when they when they return something back like a written artifact, right? So I've tried to kind of just lean into it and think about as a kind of that labor Multiplier it does require you to kind of rethink how you might do exams Which is can be a little bit tricky. One of the best things I found is Based on how good they're prompting if they produce an artifact as an image or a data visualization, that'll tell you a lot right there So that's been a big a big hack to try and figure out how to doing that as well I like the idea to have offloading some of the the 101 stuff to to your you know from the curriculum and having a live Like hey, you can go look this stuff up now come in class and now ask better questions So I think that's another way that's kind of nice to have a TA if you will I can handle some of that stuff I think one of the most exceptional parts of your presentation was basically how you've thought about the digital marketing journey as a knowledge worker and Broken it down into how knowledge workers are using AI like as digital marketers in the field and how should students learn how to use AI? With like respect to those kind of projects

So do you have any recommendations on like kind of like the thought process behind that like when you were designing this? How did you think about mapping it to future jobs? I think I'm very lucky to kind of work in industry and you start to see where where the world's gonna go where it's going right now and so Knowing things that don't require a lot of Quite frankly human create, you know, nothing's not important But I mean there's you know It's just like I think it's just like when Google was invented right like instead going, you know Looking information up all over the place. It just was such a force multiplier for efficiency and So coming up with ways where what's like normal grunt work or boring analysis work for business students that could just be automated away But if you don't teach them to prompt they totally miss it And I think that's something that I learned a big lesson on this fall They don't have the prompting fundamentals. They you know, it's kind of garbage in for sure one really interesting example about prompting is that a professor was telling me that now he requires the students to use chat GP team and Building an essay and instead of just measuring the output He measures how many prompts it takes for students to get to that output Some students are so good at prompt engineering that it takes two or three prompts never really good essay And others go back 19 or 20 times and he uses that as an indicator of articulation of what you're looking for. I think There's some data existing that we're looking into it at HBS perhaps some research studies perhaps to look at that kind of prompting

I'm super curious prompting by device and time of day So initial indicates like a lot of students like I was you know, you must your work from 10 to 2 a.m And so what is that prompting? What's what's in that as well? But I think the iterations and how many times are going through it and what's their first swing at something? That's like a goldmine. So you start thinking about could you start to evaluate students? mastery of subjects based on how sophisticated prompts are and then that comes back to you as a professor I think a little bit on wow, I Did a poor job explaining this because they asked some pretty bad prompts initially I need to boot them up better on the prep or something like that Natalie over to you for questions.

Yeah. Thanks for kicking us off I just want to remind the crowd that if you raise your hand I'm gonna call on you and spotlight you so you can actually ask Siya or Jacob your question in Person, so William Gehring you're gonna be our guinea pig tonight. I hope that's what you were intending William is a professor emeritus of psychology William's coming and Jake that's the beauty of this community is that we try and give our community members access to the really awesome Speakers and lecturers that join us. Cool. I think it's great. Hi, William. Thanks for joining us.

Oh Thanks for the presentation. It's really interesting I Yeah, I've had some pretty good success with custom GPT's in my class But one thing I struggle with is the cost to students and I I'm hesitant to require them to pay for the premium subscription but the free version has some limits and our university is apparently not in there is see you might have some thoughts about this our University isn't not interested. I guess in enterprise licenses. So how have you handled that Jacob? And you know, what are the possibilities there? kind of like I mean, it's like HBS kind of has us baked in But we have some students that you know, you know cost is very much important So they get GPT now for free as part of their tuition costs I Think there's other ways with like being maybe you know where they can maybe play with being within a free way to look at it It's not say a custom GPT They can't build their own but I've used being sometimes when people are a bit more budget sensitive and they don't want to use the the full GPT on that side of it. So that's been kind of my one hack to get around it. I wish I had a better answer William to add to what Jake said of Harvard University

uses ChatGPTEDU, and that's how they have the collaboration element of custom GPTs. That is for a paid version versus, but it's also available in Teams. So if that's something you want to pursue for a good class, like a Teams product could also make sense. But it's not available in the free product that provides other features.

Yeah. I mean, I think Teams still would cost students money probably even more perhaps. It's still a challenge because there are students who are really prohibited from the cost. We're trying to make the free product as useful as we can for them. But yeah, the custom GPTs are limited to workspaces that onboard on the ChatGPTEDU or enterprise.

One thing I just want to add is that we have learned a lot, William, that in a lot of circumstances, there's so much work that you can do without the custom GPTs. If you have some specific students that you'd like me to invite to the forum, we actually have resources here for learning, 101 and 102, and technical office hours that can support them, that would translate to even the free version. So I'm very happy to invite some of your students to the forum and help them learn and figure out some useful adoption even of the free version.

Okay. All right. Really good to meet you and thanks for joining us.

Nice meeting you too. Thank you so much.

Okay. Now I'm going to tend to some of the questions in the chat team. So our highest upvote questions so far.

Okay. I think this one's for you, Jake. How does the pep talk help the LLM? What differences have you noticed when you include it versus not?

Oh, for reasons we're still figuring out, it seems like when you tell it, pay attention or you can do this, I know you can't. I'm going to get fired. Whatever reason on the training data, it seems like you get a little bit more results out of it. So a little better focus. So I think that's really good. I think also saying, I want you to do this, but not that. That I've noticed is a huge one as well. Then this is where I think domain space is really helpful. If you know the movers and shakers in a knowledge space and say act like this, this, and this, they can be really fascinating what it comes back with as well.

So for example, that customer data I was showing that case study, one of my dear friends and heroes is a professor at Wharton named Peter Fader. A lot of work on customer lifetime value modeling, just a wonderful human being working on some cases right now. But I'll say act as Pete Fader and evaluate this data set. It's like he's got so much published research out there in the world. It's fascinating to see what it comes back with versus a general analyst. So that role thing I found to be really fascinating as well.

Awesome. Thank you so much, Jake. As I mentioned to everybody here, this is actually the first time we're using this format of webinar. So I think what I'm going to do since it takes a few moments to promote somebody to a panelist, I'm going to start with a question from the chat while I'm promoting somebody from the panel. So bear with me guys.

Yan and Ding, I'm going to promote you. So you'll have the second question. Now I'm going to move to the chat and ask a question from someone that actually didn't want to be spotlight. So Philip Berneville, he is asking, how can AI personalize learning while ensuring that all students meet standardized benchmarks? Does AI-driven education mark the end of standardization or can both approaches coexist effectively? Maybe Jake can start with that one and then we can move to Siya.

So yeah, I can take a first swipe. I think I've taught some grad courses on data science and Python, which can be a little intimidating for folks. I love this idea of the personalization where you can ask the questions you didn't want to ask, like what is this code doing or what is this statistical concept? So I think there's the idea like that personalization that's just amazing where students can do that in the background. I think standardized test is interesting. A lot of what I've tried to develop my course around is like, we don't time it, two hours it takes. Nothing in the real world like you got two hours to do this thing. So I try to mimic as much as possible in the real that's going to happen at least in the industry and I have that luxury coming from the business school on that. Does it mark the end of standardization? This is just my own opinion. I do not represent Harvard in any capacity as I say this. But I think this is a fascinating way where maybe you had a poor math teacher in middle school and you got turned off to math, and now you just think I'm not good at math. AI represents a chance to get a redo or a mulligan in a way that I think is fascinating where a standardized test would say that, well, you weren't good in the SAT, so therefore you couldn't be an engineer or data scientist. I feel like that's what's so cool. If you have curiosity now around a topic, you're totally those chains of standardization and failing can be broken now in my opinion. So how about that for Viva La Revolution?

Do you have any thoughts on that, Siya?

So my opinion is that AI is going to enable personalization at a scale that we just have never had before. I've worked in education for 12 years and personalized tutoring was always the holy grail. That's what we're striving for. Now we have achieved it. I have a personalized tutor that I talk to every day. It knows my goals, it knows my skills, it knows the projects I'm working on and helps me become a better knowledge worker. I think that skill and that approach is going to enable students to be able to study hyper-personalized degrees. My opinion is that in the future, MBAs will become very specific. If you want to become an MBA student who studies sports management or wants to become the CEO of a sports company, you'll get the exact case studies and the exact models, and you'll learn supply chain related to that. So I think over time, people will be able to express their aspirations and get exactly the knowledge that they want to achieve that. I think that means that standardization will change in terms of the rigor is what standardization will measure, and the actual content will be catered to a student's desires. Thank you so much, Siya. I love that answer. I will also share that we actually have executive leadership from the college board, which is not university standardized testing, but it's the nonprofit in the United States that creates the SATs and all the AP tests. So I think we should get them in the room to help us answer this question because I am sure that they're thinking about this.

Now, we have a few panelists lined up. Yannan, would you like to ask your question? She's a PhD candidate at Stanford University. Yannan, can you unmute yourself so we can hear you?

Okay, Hendrik, maybe we'll move to Yufan.

Hi, Yufan.

Hi.

Thanks for joining us. Will you introduce yourself?

Yeah, sure. I'm Yufan Lin. I'm Assistant Professor of Marketing at Cal Poly Pomona, and I'm also the Director for AI for Business and Leadership certificate here.

Wow.

Yeah, so I definitely infused a lot of AI in my teaching and also in my research. So I have some specific question that I want to get your feedback. Should I proceed now?

Yeah, please.

Yeah.

So my question is, so I've been working on... So I'm glad to see that you have published your health and business cases. And so I'm working on that for publish my cases. And I want to incorporate a customized GPT as a case companion. So, but I don't want the GPT to just reveal the answer. So I want it to be more like a tutor so that can guide students with the case analysis questions. So what would be some of your suggestions on how to implement that?

So key idea I have is one is as a tutor, second is they have a kind of a plan, want to kind of guide the student by asking student questions step by step, but without asking all at the same time, because that can overwhelm students, right? And finally, allow the student to kind of explore the scenario while still on track. So those are the challenges I'm contemplating to see your feedback.

Yeah, I think there's some fun ways you can hack it with maybe a Google form where they have to kind of go through prompt by prompt and check it off as they go through for the chain logic. Because I think you're exactly right. When you give me this and it just pukes back all this stuff and the students get overwhelmed, like this and then this and then teaching them and maybe you give them like, ask it about this and then this and then this. And what's the nuance in the case here? What's...

of a checklist and then you can go in and grade and say, well, you know, you only did three of the five that you're supposed to, you know, and then having them, you know, create an artifact. And I love the idea of like an artifact they create out of that custom GPT, whether it's a data viz or an image, or maybe like a strategic plan or a checklist or something from that, and then have that cold call in class. So I'm gonna pick three randomly. I have a random number generator I built in Python. Here we go. Okay, lucky number 25. Where are you? Okay, what do you got for us?

And I think that's kind of a way that keeps everybody honest from that as well. And then also we're working on some stuff at Harvard Business Publishing on teaching notes. So almost like a GPT for the students and then GPT for the instructor, where a lot of instructors, especially with digital marketing, they're not, you know, it's kind of a pain. It changes so quick. So a GPT that they could go in and just have questions and boot them up on as well. So. Yeah, that's a good idea to have a two version.

And just one follow up question I have is, so I've been testing both several different logic engine models, and I noticed that actually they behave quite differently. Yes. And they give very different answer and oftentimes complement each other. So for example, so the currency, the kind of the gold standard is open AI's, Charger BT and Clouds, Sunet, which also have the ProJet and Artifact. So when I compare the answer side by side, they highlight some of the complimentary good points, right, but each of them are actually not complete. So what's your suggestion on allowing students to use multiple logic engine models together? Because right now I can only think of a way using API call, like through like open routers, but I feel like that might be too technical for the students. So I just wondering what's your take on that?

I think the example Mimi did where she's using Firefly from Adobe and then having chat GPT, like critique it. I thought it was like really brilliant where they're getting back and forth. That was something that we did live in class where you're like, okay, let's use Cloud and then we're gonna have Cloud evaluate this with GPT and we'll bounce them around and see. They're a little bit surprised because I think they're used to Google where if you just, you know, the Google results kind of come out the same based on the keyword query. And then when you throw some of the stuff in there, they're always like, oh, why is it so different? And oh, there's some similarities here. And, but that was a really fun one we did where they get them, they kind of go back and forth to see. And then, yeah.

Yeah, thank you. Yeah, great question. So let's bring Mohamed onto the stage. He's been waiting backstage for a while. Mohamed, would you please introduce yourself? Oh, hi, how are you? So I'm Mohamed Tajbapour. I'm assistant professor of marketing at SUNY Oswego. And I'm actually teaching a course on chat GPT. So yeah, thank you. Jacob, I really like the idea of multi-platform use of, you know, generating prompt from one platform and using it in the other one. That's very interesting idea. So I have a broad question for you. And then I have another question, which is for Siya. So Jacob, I was using chat GPT for my course, and then I figured that I can use it in other courses. I created a GPT for my marketing research class, and I gave it as an assignment to my students. So the GPT would give a student a sample of survey questions. And the student's job was to find the level of measurement. Is it nominal? Is it ordinal? Is it interval? Is it ratio? Yeah. And then, yeah, the interesting part was it helped a lot with the student learning because the GPT was telling not the answer to the student. It wasn't giving away the answer. It was helping them figure the answer. And the process was giving them questions. I'm sure you had like the same experience with the assignments that you give to your students. So I figured that it's doing a great job of teaching the students. Each of them will receive individual training. So what would be our job as faculty in future? How do you see that with the improvements in different AI platforms that we have? What do you think our main role would be in future? And then after that, I'm gonna ask a follow-up question from Siya. Thank you.

We had orientation on this in August of 2023 and Ethan Mollick from Wharton came to campus and he basically uploaded this super famous case on Apple and then had it return it. And you could literally just hear, oh, like just the room, just like you just felt like cataclysmic moments, like everything's different now, right? And so I feel like the job as a professor now is part of it's interesting. I think a lot of it's like soft skills. Teaching how to work in teams and groups and stuff is still very important in designing exercises around that. And then number two is like getting them understanding where the frontiers are of their knowledge. So especially like with data science, right? Like if you're working and you don't really understand the statistical concepts and it started to do visualizations or, you know, if you don't know how to like validate that or you don't understand what it's doing, that gets to be a little scary. So I talk a lot about like teaching the students like there's butter knives and then there's really sharp knives and then there's scissors. And like, we're not gonna run with scissors, we're gonna play with the butter knife and then you can play with a really sharp kitchen knife. But we gotta make sure you know like how dangerous these are. I'll turn it over to Siya. I have one more question for you, Jacob, based on what you said.

So students are learning faster and better and our job would be, I agree more like supervising that and helping with the soft skills, especially team working and all of that. So I have one more question.

So I figured that students are learning much faster and doing their assignments, their coursework much better than before. So in the future, like five years from now, do you see bachelor programs to be still four years or should we have like a two year bachelor program and one year master's program?

Oh, I think that's really fascinating. I think in my world where I've e-commerce and digital marketing, it's all apprenticeship model. I mean, you can go, you gotta get in the trenches and I mean, to change is so quick. So the idea of building simulations where people can get hands-on stuff really, really quick and boot up, I think is important. That said, I went to a liberal arts school which you had to take the broad based, and I still think there's something for having a good broad based foundational knowledge of some areas and topics, which isn't really helpful as well. And going forward, like doing how to write, right? Well, having an AI and learning your voice and tone is powerful. So I still think like knowing how to write in your own tone is helpful. I can tighten up the logic and things like that will be interesting, but yeah, you start to see a world where I think Siya pointed out earlier, I wanna do this, this, and this in my career. I've got these kinds of experiences. Design a personal learning track over 12 months. I've done that for my own self. Like I learned Bayesian statistical methods. I've got 12 weeks. Walk me through what I should do each week. There's your learning plan. I'm not going on Amazon, buying five books and trying to piece together something, which is great. Thank you so much. And I really appreciate your presentation. I found a lot of stuff that I can use in my courses from your presentation. Thank you. It's my day to hear that. Thank you so much, Mohamed. Yeah, I have a, okay. I had a question for Siya if we have time. I'm sorry, Mohamed, we don't have time. But drop your question in the Q and A, and I'm gonna see if Jacob and Siya have time to respond to those questions that are in the Q and A that we don't get to tonight. Yeah, that'd be great. Two more guys. Giorgio has been waiting for a while, and then I have Mimi waiting in the sidelines. And I wanna give a few moments to spotlight her, introduce her to the community, and ask her about her experience in your class, Jake. So Giorgio, will you please introduce yourself?

Yes, good evening. My name is Giorgio Lagna, and I'm an assistant professor at UCSF in San Francisco. And recently I've been teaching also, by the way, my field is molecular biology or general biology education. So a little different, Jacob. But so lately I've been teaching more often in community colleges and local colleges, for example, physiology. And as you probably have noticed in your field as well, many students, especially the neediest ones, have trouble coming to office hours. So I love your idea of these office hours, but, and I've actually made one too, with the intent, as I wrote in my Q&A questions, to not try to give away the answer if possible. Chatbots are really good at answering questions, but this is tricky because I don't want them to answer the question in a very straightforward way, but rather perhaps elicit the answers in a Socratic way by asking simpler questions and breaking it down. And that's easier said than done, I've noticed, although there is preliminary success. And I was wondering if you have experimented with that. Thank you.

Yeah, and I think figuring out what you can leave in and out, I think turning off the web search functionality with custom GPT can help a little bit. I'm a little jealous because I think teaching hard sciences is, you know, that's, you know, Newton's laws still apply 400 years later, right? So there's very statistic things that there's not a lot of hallucinations or some of that kind of stuff. So maybe you can kind of tee that up a little bit, but maybe that's where part of you, you extract out about your own GPT and don't let.

that'll go to the web and then have it basically so they're not giving away all the answers as well. And then I think as we were talking earlier, kind of a checklist where they have kind of chained logic together. They go through and ask, you know, ask the prompt, go in, blah, blah, blah.

I think for hard sciences too, the ability to, you know, what's this concept? What's this acronym? What's this Latin word mean? I think that's amazing for them to kind of have that fingertippy as well. That's someone who suffered through flashcards. Yes.

All right. Great. Thank you so much. Thanks for coming. Great question. Oh, Natalie, you might be on mute, I think.

Giorgio, I was just saying, thank you so much for joining us in community. We can't wait to learn from you. It sounds like you're doing some really awesome work too.

Mimi, welcome. Hey, Mimi. It sounds like you're the star student. Thanks so much for having me.

Hi, Jake. Hey, good to see you again. Definitely. So Mimi, definitely I'm going to reach out so that you can facilitate a round table for us. Sorry, I think there's just a little, a little bit of lag in the wifi. I didn't mean to cut you off. But, so I will reach out so the rest of the community can learn from you, but tell us a little bit about your experience being in Jake's class and how did you end up like being the prime adopter? Be honest, be honest.

That is much too kind. Being in Jake's class is such a joy. I think that what Jake has done has really taught us all the muscle of how to use AI to really think about doing almost everything better. And better can mean a lot of different things. It can mean thinking about how to do work more efficiently, for example. Something really neat about our class is that it's an inverted classroom model. So we watch lectures at home. And sometimes I have a question about one of those lectures and I think, oh, it would take me a really long time to watch the entire lecture all over again. But with AI, I can upload the video transcript and really quickly say, okay, I remember there was something about SEO at this lecture. What was that thing again? It can also mean how can I understand the foundations a lot better? So I have a lot of silly questions that I would not feel comfortable asking in front of my 70 classmates, but I can use AI to ask those silly questions before I come in. And then of course, we talk a lot about using AI creatively, but we constantly think about how we can use AI as a thought partner, like Siya said, how to make creative assets, how to do data analyses that we previously thought were beyond our capacities. So I think we've really thought through what are all the different use cases we can use these tools for?

That's awesome, Mimi. And would you say in your background, like, do you have experience as an engineer or did you come to this class as a less technical student in the classroom?

I did not have an experience as an engineer. I did some basic coding while I was an undergrad and we learned a little bit of coding at HBS, but I by no means was an engineer. I by no means was a computer scientist myself, but I experimented with writing Python while I was in this class. And I just experimented with building a lot of things that I otherwise would not have been able to build.

If I can just really quick interject, what was so fun about having Mimi in class was you give these students these tools and you watch what they can create with it. And that was what was really fun was watching someone like Mimi who from the get-go kind of perked up and we had some great office hours conversations and it was like, just seeing what they can do when you just give them like the nudge, like, hey, go for it, see what you can do. Let's see what it's gotten there. Fail, let's make stuff. The hardest things I have to do at Harvard is get students to fail and play. So a lot of trying to get them to play and make stuff and it didn't be perfect. And Mimi embodied that spirit of the course probably better than almost anybody. So yeah, it was a joy having her.

Awesome, we're so happy that we were able to spotlight you, Siya, I mean, Mimi. And Jake, I was really blown away by your presentation. It was so specific and clear and concise, lots of actionable items. And just so the community knows, we are also working on building a community GPT. So we're already rendering these transcripts as one-pager outlines, like key takeaways. And now we're training the custom GPT so that community members can ask questions and then find the exact spot in the event where we learned it. And I'm sure we've learned so much from you, Jake, and yours is gonna be definitely one of the top hits. So thank you so much for joining us. And I hope this is just the beginning of a long collaboration together.

Of course, I was honored to participate and excited to learn with everybody else in the forums as well as we go forward.

Beautiful, yay. And thank you, Siya, so much for being here tonight. Thank you for being the person that actually kicked us off with this AI Native Faculty Fellows Series. It's been a really beautiful opportunity to connect with educators and learn from them. So thank you so much, Siya.

Thanks so much. So we had one person we didn't get to in the background, Dante. Please just drop your question in the Q&A and we'll follow up later and hopefully get everybody's questions answered in the wake of this event.

Before I leave you this evening, we've already gone a little over time, but I have just a couple of announcements. I wanna remind everybody that everybody that showed up here, we're all presented with our LinkedIn profiles on the right-hand side. So this is your community. Please connect with people. You can DM them in community. You can ask them for a coffee date. You can find collaborators here. In the wake of this event, if you go to your messages, you'll find this event title with every single person that showed up and you can network that way if you didn't wanna do it live. So if today's session inspired new ideas or raised further questions, we encourage you to explore more about chat GPTEDU and connect with our forum community. Let's keep this conversation going as we work together to make AI a very powerful ally in education. So if you have interesting use cases you'd like to share with us, please DM Siya, who we met tonight, or if you're outside of the United States, if you're in Europe or if you're in Asia, if you're in South America, if you're in Canada, please DM Jaina Devani, who's running EDU in the go-to-market team outside of the United States. She'd love to hear from you as well. And as always, you can always reach out to Caitlin or myself and we'll respond as quickly as possible.

Upcoming very exciting event, December 3rd from 6 to 8, 15 Pacific time. We've invited world-renowned mathematician and forum member, Terence Tao, along with OpenAI's VP of Research, Mark Chen, to share their thoughts on OpenAI's latest model release and AI's role in advancing scientific discovery. Folks, this is really a once-in-a-lifetime chance to have a seat at the table with Mark and Terry. These are not people that typically present, let alone show up to an intimate community round table. We'll all be face-to-face for that Q&A, so I really encourage you to show up. And also, we're gonna share the OpenAI forum member referral in the chat. If you have network, like folks in your network that you think would also find that community talk really amazing, we encourage you to share their member referral application with them, and we would really love to host your network as well. On the very last note of the evening, we're forming interest groups in the community to connect members with one another. So if you have ideas on how we should shape these interest groups, please take this survey that Caitlin's gonna drop in the chat now. This is your community once again, and we absolutely take all of your feedback, your comments, questions, concerns to heart, and that is really what's shaping all the features in this community.

All right, folks, happy Tuesday. We've gone so far over, but thank you for bearing with us as we experiment with this new format. I think it's kind of what, extended the timing a little bit. Jake, thank you for sticking around with us a little bit over time. We're so grateful. Siya, thank you for being here. Mimi, thank you for being here. Thank you to our community manager, Caitlin Maltbie, who really is like the tactical magician that pulls all of this together, and I hope you have a wonderful evening. We'll see you soon. Good night, everybody.

+ Read More
Sign in or Join the community

Create an account

Change email
e.g. https://www.linkedin.com/in/xxx or https://xx.linkedin.com/in/xxx
I agree to OpenAI Forum’s Terms of Service, Code of Conduct and Privacy Policy.

Watch More

1:04:24
Learning to Reason with LLMs
Posted Oct 04, 2024 | Views 5.1K
# AI Research
# GPT-4
Exploring the Future of Math & AI with Terence Tao and OpenAI
Posted Oct 09, 2023 | Views 19.9K
# STEM
# Higher Education
# Innovation
OpenAI's AI Trainer Community Mixer with Special Appearances from Research Leadership
Posted Jul 12, 2024 | Views 11K
# AI Research
# Expert AI Training
# AI Safety