Welcome to Season Two of edtech insiders, where we talk to the most interesting thought leaders, founders, entrepreneurs, educators and investors driving the future of education technology. I'm your host, Alex Sarlin, an edtech veteran with over 10 years of experience at the top edtech companies.
Hello, everyone, it is Ben Cornell and Alex Sarlin with another episode of This Week in edtech. It's the week of March 20. And we're gonna just officially dubbed this the week that chat GPT for broke the internet. It is GPT for coming from right from left going forward going backwards. It's an exciting episode today. We'll also bring you some other headlines. Before we dive in Alex, what's going on with the pod this week?
We have great episodes of the podcast coming out this week. As always, a lot of really interesting things are happening in the world right now including chat GBP for this week, we talked to John gamba who is an entrepreneur in residence at Penn GSE at real ad tech, very knowledgeable entrepreneur as well as Thaddeus rightly, who is a classroom teacher turned ad tech researcher who knows a lot about student engagement. It's a really, really cool episode, just coming out on March 20. I really recommend that one. But as you say, Chad GPG for is on everybody's mind this week. And we have an incredible guest today who's right at the core of the intersection between Chad GBT and Ed Tech Kristin cerbo, the chief learning officer of Khan Academy, Khan Academy launched an incredible chat GBT AI based product this week called conmigo. And we're gonna get the inside scoop right here. Kristen, welcome to the podcast.
Thanks so much for having me. And thanks for all you do to talk to those folks that are really on the inside of edtech. I always enjoy your your podcast.
That means the world to us, you've been a listener, we really try our best and Khan Academy is obviously always has been at the forefront of edtech for many years in terms of access in terms of learning gains, in terms of working with schools, you have your own school, but we're here to talk chat JPD for So let's kick off by a little bit of talking about how this came about you obviously being able to launch this the same week as check GPT for you clearly had some early access. How did that work between you and open AI?
Yeah. So last August, Sal and I got an email from the folks at open AI and they were interested in talking and we were you know, we'd kind of been keeping our eye on AI developments but didn't feel it was quite ready for primetime yet. But sure, we'll take the call. And interestingly, they were asking about AP Biology, because Bill Gates had challenged the open AI folks to be able to get the model to get a five on the AP biology exam, and then he would talk to them. So they were interested in AP Bio questions that they could help test the model on etcetera. And we, of course, at Khan Academy have lots of AP Bio questions and content. So we said, Well, we, you know, we can chat, tell us what you've got. And they said, we'll get back to when we've got the model trained. And so and then in early September, they came back to us and gave us a little demo. And we were blown away by what it could do. It was amazing. And then they actually hooked us up with this slack application so that we could talk to it in Slack. We could ask it to do things in Slack. And so I'll tell you the first thing that I asked where I was like, Oh, my gosh, I said so I'm a teacher who wants to use evidence based practices, should I align my instruction to students learning styles. So I'm sure all the listeners know learning styles does not have a good evidence base, particularly around teaching tied to a student's learning style. In fact, most students learn best when they get instruction in all kinds of different modalities. It wrote a response, it looked like I wrote it, like it's, you know, it summarize the evidence. It's not going to citations, but it did say there's a 2009 paper that it had a quote from it. And then it said, but you might want to try these other techniques that are more backed by research and listed three of them. And I was like, This is crazy. So that was the beginning. And since then, you know, we've been done a variety. We started off with actually a small hackathon where we were just blue skies, exploring possibilities and then got really serious about okay, let's narrow down what we want to focus on for a launch because open aI had said, Hey, we're launching this in March. We'd love to have you be up Gartner in that launch to show what it can do in terms of education.
You know, one thing that I think is really fascinating is your ability to demo this and develop it, but also to keep it under wraps so that when the launch happens, it's a positive use case. And one thing I'm seeing, especially in education space, is everyone's taking the open AI lead of open betas, where it's like, just launch something out there get first mover advantage, if it works, or if it doesn't work, you just caveat it. And what I loved about your launch is, it actually works from the beginning. And there's still the caveats, but you actually treated it like a formal product launch. How tempting was it to just get something out there, you must have been seeing all kinds of headlines in November, December, January, where it's like, let's just get it out here. You held on you work with open AI. I mean, obviously, the difference between four and 3.5. And three is a huge leap in terms of accuracy, in terms of the ability to, you know, use sources and data. So it's not really just an inference machine anymore, it really is driving the answer. How did you navigate that? And tell us a little bit about that open beta? I think a lot of our entrepreneurs listening are like, should I just open beta this? Or should I wait and really get the product to a full product launch?
Yeah, so absolutely. Everything you say is spot on in terms of being like keeping our mouths shut too as other things come out. Once chat GPT launched, every place you went, people wanted to talk about it. And I just had to pretend like oh, yeah, that's really interesting. It looks like there's some interesting things there. Like my book, I was like, sit down at book club and like, Hey, do you like experimenting with that? TPT? Oh, yeah, that's interesting. It was difficult, but then to your point when people start releasing things, and we know that we're working so hard to really release this as a product, and we want we're trying to get it right. And there's so many just nuances around getting it right, that aren't just like, throw something up and let it ask you questions, but around in are trying to control what it's doing and where things that. So we did want to be able to show that the thoughtfulness behind it that it's not just throw something up and see what works, but throw up, it's always the balance between the you know, don't let perfection be the enemy, the good. But, you know, we don't want to just be throwing particularly we're talking about learning, these are big impacts that we have. And we want to make sure that we're doing the right thing by learners. So that balance is something we're always wrestling with.
So just take us under the hood, what did that look like? It was engineers training and tuning the model? And then you know, user experience people thinking about the front end? How did the Learning Team interact? How did you test it? I mean, I just love the little bit of the inside baseball of how does one, take a open AI GPT for, you know, integration, and really bring it to bright with quality in the market? Yeah,
so there's a couple of things. So our engineers, so there's the open AI team that's working on the model itself. So we weren't for the most part, I'll come back to that we weren't really training the model. But open AI was interested in getting better at math, in particular, and so we were giving them problems, and giving them the types of interactions we wanted it to. That ended up being baked into the model. So I'll just kind of they, we worked with them to give them actual tutor interactions. So we'd use a Khan Academy problem, and then have this, you know, prompt saying act like a tutor. And so it says something, but often it would say something we didn't want. And so then you rate that response, and then your ideal response. And so that ends up being seven or eight back and forth for like a typical algebra problem. And we did about 100 of those kinds of problems. And then that sped into right now the whole open AI model. So that's going on on one side, then the engineering team is making all the, you know, connections under the hood, how's the API connected? How are we sending information back and forth? What does that look like? And then there's the learning design teams and the user experience teams that are saying, Okay, what kinds of interactions do we want to have? And how do we build those and then that goes, again, back into the engineering cycle, where they're creating that on our site, what we want the experience to be and how those fit in. And so there's everything from the tutoring to one of my favorite things is this tell me a story or craft a story with me? And so with that, it started out as hey, let's get The model to ask questions about character and setting and plot and pull those all out. And so we started right draft, creating those prompts. And this whole prompt engineering thing is really an art and a science and figuring out how to make it act like you want to. And then we said, You know what this is? It's asking too many questions before you get to the story. Like, it's not really fun, it starts getting, oh my gosh, are we ever gonna write this great. So we switched it to like that party game, or sometimes parents do with their kids where you write two sentences, and I write two sentences, you write two sentences. So we made that kind of happen, where it's also writing to sentences and then giving kind of an edge like, oh, you might want to tell more about the settings so that the reader feels like they're really there and giving some of that prompting about how to write a story as it's going through. So all of that's happening on the learning side, while we're trying to make the model do what we want it to do and how that fits together.
It's so exciting. And I just for our listeners who haven't read the press releases yet some of the things that you can do with conmigo. Right now, this story, writing back and forth is amazing. You can get help with AP exams, vocab programming math, of course, you can interview historical figures like Cleopatra, you can debate with it. But you know, one thing that you called out really clearly in your in your announcements about this is that you were very careful to train conmigo to not give students answers, because that's really an instinct that an AI would have, especially in the context of a con where kids are constantly answering questions that they want to be like, I don't know how to do this. And the AI will be like, well, here's your answer. And you know, and we all know from a learning perspective that that is not where we want to go. I'd love to hear you talk about that, from your background, as you know, pedagogy just how do you use the prompt engineering combined with what you know about learning design to turn conmigo into a tutor rather than a, you know, a cheater.
So I have said this to the beginning to all of our teams, this is not going to be a place where you can come and cheat. Now, there's an interesting question here. I mean, obviously, everyone would be like, oh, yeah, us too. But there's an interesting market question here. And us being a nonprofit gives us a little bit of headway here. Because if you ask students what they want, you know, there's a lot of places where they don't get the answer to their homework that they want to get done. And so we have a little more space to say, That's not who we're going to be where we don't have to be exactly what this market the student market wants, because we've got revenue that, you know, isn't tied to the market, but more tied to what philanthropists and other folks want. So coming in, that was what I kept saying, we are not going to be the place where you can cheat. And I have written a little bit about this, but it really wants to give the answer and I shouldn't anthropomorphize it. It doesn't want anything, but we talked that way. And so it was giving answers and giving answers. And we would be trying different things in the prompts. The folks at open AI said, you know, it's kind of like, if you use capital letters, it's like yelling at it. So we would put on caps and say do not give the student the answer. We're like yelling and back and forth and trying to get it to do that. But we got to to do the tutoring plays pretty well. But then we realized we had all these other activities like the tell me a story. And students could just ignore the tell me a story prompt and just put in solve eight equals 4x plus two, and it would solve it. So we also have locking down all of the other places that it could be, and it is still the case. Full disclosure, probably if you have a bunch of interactions, you will be able to find some way right now to get it to give you an answer, like it is not perfect. And that's why we're releasing it as you know, in this research phase, not sending it out to the world fully baked yet we've got it got work to do and where those where those sit. But the whole trying not to be a place where you can just type in your homework problem and get the answer has been a guiding principle as we've been trying to develop what we have.
Yeah, so I'd love to jump in on that. And just think about where this is going. You know, you mentioned, there's the open AI infrastructure. There's the learning team. And then there's basically this supervised AI training where you've basically been tuning up the model to put the guardrails, it is really clear to both Alex and I, that the use case of cheating or the use case of the answer that is the lowest on the Maslow's hierarchy of AI use cases. It is the lowest denominator and basically kids if there's a desire from the learner to just do that, there's going to be infinite choices. And by the way, open AI through being that's the The whole purpose of the open AI being partnership is to give you the answer. So you're clearly carving a different space? How do you see this evolving? And talk to us a little bit about how you see this evolving with educators, as partners, with parents as partners, also the student because what Alex and I have been observing is there's this triangle between the parent, learner and educator that if the AI can help that triangle work better, there's a real opportunity for accelerated achievement, engagement, success, etc.
Yeah, so we agree, I also, I usually do this triangle when we're talking about our classroom product of the student, the teacher and the technology. And so then we can think about the same thing with the parent and thinking about, there's actually activity theory to bring kind of the academic side in here, which says, try to understand the roles and the responsibilities that each of those players play, and understand what you want each of them to do. And so we try to think about that, like, what does a parent want their role in the system to be? What does a teacher want their role in the system to be? And then
not trying? Lifting? Like who's doing the cognitive lift here? Yeah, yeah,
and not trying to make the technology something that takes over a role that a teacher or a parent wants to have, and then helping them think about where that is. So that's the basis of it. But we've done a couple things. So one is one of the features we have, that's partly safety and security. But also, I think it's a nice feature for just thinking about what students are learning is all of the chats and the chat history, are recorded and are available to the students. So the student if they said, Hey, I had a good chat about this kind of problem two days ago, can go back and see that, but it's also available, if they have a parent or a teacher linked to their account, it's available to them. So they can see what the history is are. The next step, as you can imagine, is using the AI to summarize those chats for the teacher or parents. So you submit it's very kind of meta. You take the chat histories, and you say, hey, teacher, hey, parent, here's what so and so was working on.
And you could summarize it in Spanish, if you needed to, for Spanish language, parents.
Definitely. So I think there's a number of those. And then we're also thinking about teacher tools. So we at this launch, also, I know the students stuff gets a lot of play, but we've got just some teacher to have like help create lesson hooks that thing at the beginning of a lesson that captures their interest and attention and maybe brings them in learning a little of their per activates a little bit their prior knowledge. So we have where teachers can just brainstorm lesson hooks we have where they can brainstorm exit tickets, like those kinds of things where they can use the AI themselves to just come up with new ideas and think about where things sit. So I think there's a lot of room there on the teacher side as well. And then finally, I think there's just the idea of, you know, every time I've talked to teachers about showing them data, they always say, that's great. What should I do about that? And so being able to provide teachers with that interface, where they can have a conversation about, here's what the data says, What should I do next? What are those things that AI can do? Are some of the things I think looking forward, we might think about for teachers?
That is so exciting. I mean, yeah, that conmigo is in the announcement. It's a tutor and a teaching assistant. And I think these teaching assistant use cases that you're naming here are so powerful, the idea of being able to, you know, summarize what a kid has been doing to either the parent or the teacher, and then turn it into actionable advice be like, seems like your daughter was really trying to get me to give her the answer today, you might want to talk to her about how important it is to figure it out on her own. Like, that's some powerful stuff. But I
love how you honor the adult in the loop. You know, in AI theory, they often have human in the loop systems for customer success driven AI. And in this situation, if it's only the student in the loop as the human in the loop, there might be a bias towards just get the answer fast. But when you have the adult in the loop, you're honoring their role of really shepherding learning. I think that's like a takeaway I think every edtech company should think about is, what's the human in the loop? And who is it? And like you said, What's the role they want to or need to play in order to create great outcomes? Like my brain is exploding?
Exactly. And what's so powerful about this is it can bring all the different humans in it could say, even what you know what the childhood today, what should the parent do, what should the main classroom teacher do? What should the literacy specialist do? What should the special needs teacher do and it's no more sweat off its back to give you all of that. I have to ask though, Kristen, one of the thing that's so interesting about this model, and you've talked about this a little bit is that Khan Academy is nonprofit, and it's designed to be completely free, really, really, really open and that's the whole model. It's fallen through publicly funded for the most part, but open AI and chat GPT is not free. And you know, there was a lot of news when it put out its original, you know, its original explosive launch. And it was used so much it was costing open AI millions of dollars, you know, got a lot of publicity out of it. But just because getting power, how do you square the free nature of Khan and the paid nature of open AI?
That is absolutely a big question for us moving forward. How does this work as a business model, not that we think of ourselves as having business models, but for our organization? How does this funding makes sense. So there's a couple of things that we're working with and trying to figure it out, and also open AI coming to us wants to work with us. So they want us to be able to offer this and where it is. But right now, if we made this free to students and teachers alike, all of our content is we don't want the headline AI bankrupt Khan Academy. Not where we need to be. So there's a couple of things going on. One is we you know, even we saw with the release of for the price of three and 3.5, or, you know, just dropping, you know, in six months increments really quickly. So that's good news. And we look at that. The other piece is not everyone knows this, we do have a districts offering that is a paid offering for districts. And right now for districts, of course, the teacher and student contents always free. But districts often want things like rostering, and district administrator reports and co teaching functionality, those kinds of things. So that is actually a paid offering that we have now. And so there's some you know, in early days, as we wait for some of the pricing to go down and see if we can get closer and closer to free. We may be thinking about offering to our districts. Now the nice thing about our district offering is that our districts are over representing students from historically under resourced communities, and they, you know, often have the funds from other places that they can offer these kinds of things. So it does help us think about maybe it's another way to reach those students who need us most by working through our district offering to start with, but we will see what happens. And we certainly need to be paying attention to it right now. It's basically three cents per 1000 tokens into the model and six cents per 1000 tokens out 1000 tokens is maybe 750 words. So when you start multiplying that out by a million users, or going back and forth, it adds up quickly to billions of tokens. So it's still an unsolved problem that we're trying to figure out.
So we're gonna have to solve fusion energy first, and then basically, will take the cost of power down to zero. And then, you know, I'm optimistic around computing power going down, the challenge is the LLM are going up in terms of complexity. So as you have the cost per compute go down, and the complexity of ll M's go up. And they're in a space race here to have the most valuable LLM. I will say, the business model here for both nonprofits and for profits, still is the same in education, which is, if you're really solving pain points and providing real learning value, not just a quick answer, there's something sustainable there. Kudos to you all, for taking a real leap for Khan is such a trusted name, by the way in the world where content is infinite, like trust and curation, incredibly valuable. And you know, the responsibility that you just even walking through us the kind of path to launch. I know, you're saying this is a research only, you know, this is a research open. But it feels like a very, very polished product compared to some of the stuff that Alex and I see every day. And we love those entrepreneurs who are scrappy, and putting that stuff out there. But kudos to you for really not sitting on your laurels and really pushing the technology for what it could be. And it does make me feel good to hear that open AI was also outbound reaching out to organizations on that makes me feel a little because, you know, in the media, there is this whole, like dystopian narrative where they're like, well, the machines are gonna roll this off. I think this is just an incredible use case. Thank you so much for helping us see this potential.
And anybody out there who is working in philanthropy, who is working in districts who hears that idea that, you know, Khan is working on a virtual teaching assistant, as well as a tutor that can make actionable insights out of your data, and they may need a little bit of support to scale it, please, you know, jump right in because this is a moment where I think the sky is the limit and we're all as a community as an edtech community, working on figuring out how to make this the best possible experience for everybody involved and Not head towards that dystopian cliff that we all worry about where you know, where cheating is rampant or people or school systems sort of fall apart in the face of this new technology. Kristin, disable. This has been unbelievable. I'm so happy that we could have you on the show today. You're welcome back anytime. And yeah, we'll all be following what's happening with conmigo in the future. It's just thrilling.
Thanks so much for having me. It's been a pleasure to talk about this.
Thank you so much, Chris. And we'll talk soon. Wow, Alex, that was such a fascinating conversation and a great way to kick off our ship this week. We've got plenty of headlines. So dear listener, please stick with us here. We've got a lot to cover. But we do want to also take a moment to talk about the other GPT four news, because as you can imagine, staggered up with the GPT four announcement on Tuesday, Wednesday, Thursday, Friday, Saturday, Sunday, we just had roll out after rollout of AI news. Kind of kicking things off is the other big player in ed tech these days. Duolingo they are touting Duolingo. Max. Tell us a little bit about that, Alex?
Yeah. So with the champ GBT announcement, there were two headline education partners, Khan and Duolingo. Duolingo Max is a version of what they're calling super Duolingo. It's their premium subscription. But it also includes two new AI features, and they're pretty cool. One is a called explain my answer, which basically means when learners make a mistake in their Duolingo lesson for any reason, they can actually find out what kind of mistake and why they were wrong. And you know, any kind of you know, learning people know, that's incredibly important. That's called elaborative feedback, and it changed it, it's much, much more useful to know why you got something wrong than just that you got it wrong, they call corrective feedback. So that's actually very big for learning. And the other is maybe even more intriguing. It's called roleplay. And it allows you to talk practice conversational skills, with Duolingo is cast of proprietary characters in a chatbot experience. So you can talk to duo or bad enough to even talk, you can talk to the different characters in Duolingo, get back and forth and actually practice that's another enormously powerful tool, especially for language learning. So these are pretty big, I'm excited to say
I just look to Duolingo as a leader in the space one, because large language models, what's the natural first use case is language learning, right? So the model, the generative nature of it, and the fit with the product category is really good. But number two is, you know, there's the technical side of learning, but this idea that you could anthropomorphize a chatbot, and make it more engaging and interactive, that is totally pedagogically the move that you want to do when you're teaching language. And it's really about reps in practice, which is a human to human interaction, very expensive. Ai interaction, it sets. You know, one thing that I wonder about, and I'd be curious, your thoughts, how much of Duolingo strategy because it's such a good fit. And it's language learning, often for adults, how much really is transferable to other modes of learning, and how much of it is actually just fit with the language learning niche, which, by the way, we call it a niche, still, like number one best selling category, that tech basically for the last five to 10 years? And if you go back to Rosetta Stone, like maybe the last third?
Yeah, yeah. And especially with the app space, it's by most of the top apps in the world for education are language learning apps, and then we have a whole suite of big stars in that space. Great question. My short answer, you know, my guess my best answer is that language learning is a particularly good use of AI. And, you know, we just talked to Kristin about the business model. Duolingo is a particularly good use, because it is a paid subscription model. And they are already as we just said, baking in an additional cost to the consumer to use the AI features, which obviously is designed to cover at least some of the costs of the computing power. So it's a really good fit both pedagogically. And from a business standpoint, I believe that language learning is a very specific type of learning that needs a huge amount of practice a huge amount of feedback, and it's a really good match. That said, I don't think it's a uniquely good use case. You know, AI surprisingly to some is not that good at math yet. The judge should be 84 is not an amazing mathematician that said, it learns really fast it learns faster than most things and it's tearing through you know, that all of human knowledge, you know, to figure stuff out so it's getting there. I think language is an amazing beachhead for a subject that it can definitely help with. You can have a you know, two hour conversation with one of these Duolingo characters and practice anything you want to talk about you Practice your business English practice your, you know, any topic that you know your medical English, that's amazing. But I think you know, a lot of the subjects are going to see parallel amazingness I mean, we just heard that you can ask Cleopatra questions, or Jane Austen questions through the new conmigo. That is out pretty amazing way to learn about about History or English. So I don't think it's unique at all. I'm, you know, any listeners, although I'm very bullish on this as ability to turbocharge education of almost any type, I don't think it's unique to Duolingo. It's just a great first use case. Yeah,
so maybe the Vanguard will be in this language learning place. And then it translates another large player that's making a double down bet on AI is Paper Paper acquired really to address the literacy crisis. It's an AI voice enabled tool that was developed by Harvard Graduate School of Education students. And the idea is the student reads out loud, and then essentially assesses or supports reading proficiency. I think there's something to kind of interesting angles on this one is, you know, paper has been in the headlines for struggling with delivering on its contracts. And so the idea that they raised so much money at such a high valuation, now they're kind of using that capital to build a bigger toolkit will that strategy work. So that's the paper story. The other story is the tutoring space. And most of the tutoring companies are building from human based tutoring down to like, you know, chatbot based tutoring. So the idea is that over time, the human aspect gets replaced more and more by intermittent chat support, almost like a call centre might where it would start out, you talk to a person, you know, in olden days, and now it's a chatbot on the front end, but there's still a person on the other side paper going the total opposite way, kind of starting bottoms up with their chat based tutoring sessions. And then the AI is actually a level up in terms of engagement and quality. And who knows eventually, I'm sure they offer some human tutoring in their model. You know, the second story here is how will the tutoring space play out undisclosed amount on the Reedley after acquisition, but you know, we've been hearing about really for the past couple of years they were featured is UCSB as one of the, you know, top startups. What's your take on this paper acquisition?
It's super interesting, because I think we all think of paper as a tutoring company. But I don't think that's how it thinks of itself or certainly not how it wants to present itself. It calls itself a an educational support system, which they abbreviate as an ESS that is not a term that is standard in the field, as far as I know. But what they're trying to do with a term like that is say, No, don't think of us as tutoring. Think of us as support for all different types of learning. And I actually think this acquisition, even though the AI tutoring spaces is going to be absolutely madness this year. I don't think this is really about tutoring. I think this is about literacy. And you've seen a mirror learning come out with basically with a speech recognition based reading platform, and it's been partnered with Houghton Mifflin, you're seeing ello fascinating, you know, startup that uses that basically can respond to students reading books through AI, I think paper is trying to get into that space. They have this enormous science of literacy movement happening in the US right now. And, you know, other countries are ahead of the US on this. They've been doing phonics for a long time. But I think that this is paper trying to diversify and not be caught as only a tutor because of not maybe the press but also just maybe the fact that post pandemic tutoring is not as you know, as much of an ace in the hole as well. So
just to call this out to, you know, our first interview and our top headlines are really about generative AI. This one is actually about voice recognition AI. You know, the other one to add to your list is so Fox lab announced a partnership with Scholastic, which, you know, from a distribution and publisher standpoint, is a huge player. So we're really talking about people going out using the AI to go after a different segment of the market, and maybe be more expensive. It's truly fascinating. And also, you know, we see other players kind of getting into the space, notably Barnes and Noble education, launching an AI content detector. We also saw that photo study, which we covered maybe four or five episodes ago, they announced a map GBT platform for content publishers. So basically, they were originally like, Hey, we've got GPT tutoring. Now. We're going to open that platform up for publishers and educators to basically leverage that to scale their own AI. So this is basically different players in the staff ACC, figuring out ways to leverage AI to grab more market share from business standpoint and or expand their impact product quality from an offering anything that stands out to you from this group of players.
Yeah, I mean, one theme that we've been seeing you'll hear it in here is so that photo study got it. It's called got a team's product that they're selling to content publishers includes a, you know, select your favorite teacher piece where you can learn from Einstein or Benjamin Franklin or Ada Lovelace or Newton or others. We talked a few episodes ago on the podcast to Prof. Jim to Deepak CCAR, from Prof. Jim, which is all about these sort of personalized teachers and we just heard about chatting with Cleopatra, we've, and I think that there's a layer of a phase, let's put it this way, a phase of this AI revolution that says, whoa, you can bring in these personalities and people in all sorts of ways in video in text and personality. Let's explore that, especially when they're sort of public figures like like Ada Lovelace and Einstein, or Aristotle they use and Prof. Jim. And I think where this is gonna go is I think there's gonna be a phase I don't think this is the end of this. I'm not sure students have been sitting around saying, Oh, God, I wish Ben Franklin was my teacher. Like, that's not a pain point. That's, that's a cute idea. It's not a pain point. But I think where it's gonna go is what does the embodied agent of the future look like? What who do students want to learn from Yeah, and I think it's going to open some really interesting
ended up being like LeBron, or right, you know, being like Lizzo is going to teach. I guess the question though, what do you think about that ethically? Because, you know, one of the things is this, like realities, distortion for kids and knowing what's real, what's not, and in the social media, but, you know, I really do feel like this is our web three movement moment. I don't know that that's what technically web three was meant to be. It was really around blockchain. But you know, to go from internet to like, social media, video, all that stuff to where we are now. This is the leap forward. How do you in social media, kids are losing sense of reality, adults are losing sense of reality, we are seeing social outcomes from social media that are distortions of what like normal interpersonal non technology interactions are, do you have an ethical take on anthropomorphizing AI?
I think you and I have had a couple of conversations about this. And I think we there's some really interesting, yes, right now, my personal take is that it is a I've called this out on the show before, but something that really caught my eyes as a headline, probably a month or two ago, was that the Chinese government went with a very proactive law based on their, you know, personality as a government, basically saying that any AI created image, and I'm sure this extends to video as well needs to be watermarked basically needs to be disclosed as AI. And I remember seeing that and saying, Wow, that's the kind of policy you come up with, when you understand AI really well as China does. Yeah. And I was like, Oh, if the US is really paying attention right now, they should definitely be thinking about questions like that. Do we need warning labels? On AI? Do we need disclosure? You know, messages? Do we the same way that at the end of a political ad, it says this is paid for by x or on an album, it says parental advisory are on a cigarette package? It says Surgeon General's warning? Like if they are not, and I think they probably are not, if the American government is not thinking about this yet they are, they're missing the boat, because otherwise, it's going to be reactive, they made that arguably, really bad law in the 90s, for social media that basically says that any social media platform is not responsible for anything that happens on it. And that has led to a very strange history of social media. And I think the same thing could happen here. They could fold to the sort of business instinct of this and say, Hey, let's let the market decide how they're going to use AI. I hope they don't. What do you think, but I know you think about this a lot.
Yeah, so the national AI initiative act of 2020. If you ever get a chance, go to ai.gov. It's both, you know, oh, we have a government agency on AI. So it's like, wait, what, why am I just hearing about this? And then it's allowed Jack as you read about it, and you're like, wow, they are so far behind where I was going. And you know, of course, they're more focused on Super VIP vice AI in use cases like transportation, like, you know, AI self driven cars, which by the way in my neighborhood, we had our first non fully robotic car ride this last week. So I guess this idea of watermarking or making it clear When there is a bot, and when there is a human, I do think that that is probably our best hope. But given the practical reality of our market economy, I think that that's very unlikely to become a standard. And even if it does, I think there will be a factor by which the market incentives will be to minimize that awareness as much as possible, because we are in an engagement based economy where more engagement leads to more revenue, either in subscription models or advertising models of business. And so, you know, I think that there's going to be a real reckoning on that one, when I see what the kind of coming back to the headlines when I see what the folks at photo study are doing with got it, it is interesting to me that there are now b2b players that are essentially creating a text voice and video layer of AI enablement, that then they're selling to other businesses. And if that ends up being the predominant way of spreading AI, to your midsize and large size companies, that actually might have a better chance of safeguarding against like over engagement, things like you know, with tick tock, now you get a warning, when you've been watching it too long, you could imagine like a warning system where you've been engaging with AI too much like reality check, right? So if everyone is it's wild, less, and everyone's developing and building their own, just simply on top of a cloud stack and open AI, or, you know, other LLM integration, I think that it's going to be it's going to play out in a zillion different ways. If we do find that b2b winners that sit on top of the LLM stack, kind of think about Salesforce, Salesforce basically has its full stack, but then there's this whole ecosystem of vendors that make Salesforce work in your company, it ultimately means that every company can have a personalized AI ability. But there are some like core guidelines that you know, Salesforce has basically made every CRM have these kind of core flows. So very interesting to see, we should probably move on for the benefit of our listeners, and shout out to Sarah who helps us with production, we should go other topics.
We need a dedicated podcast to AI so we can not cover it not take up all the
time. And we will be at ASU GSV April 17 or 18th. We're doing a roundtable at Tech insiders with a bunch of other folks. focus really on AI, you know, it's social, it's a happy hour, but it's also bringing people together to chat about this in many different topics. So please be there. All right, Topic number two, let's go we're gonna go round the world. Global edtech. Let's start in India, go to Africa, then we're going to Europe, we're hitting all the continents. Well, Asia is the kind of for India. Tell us about what's going on in India, Alex?
Sure. So there were two headlines that caught our eye from India this week. One is the completion of a process that started a little while ago, but it's very interesting, which is that we in school, which is one of the big Indian Ed Tech unicorns, has now completed the acquisition of Pearson's K 12 learning business in India. So this feels like the completion of a very postcolonial for lack of a better word, you know, movement in education, where British Pearson is returning the Indian K 12 business to an Indian company, which is now which is a unicorn is doing incredibly well, which is now going to shepherd it. It's a very interesting moment, I think, for Indian Ed Tech on the download, but I think so. And then we also saw an interesting press release from upgrade, which is, you know, started in 2015, one of the large workforce development and skilling platforms that is basically says, they're really moving. And they basically just topped off there 22,001st Time candidates have gotten work through their platform. So they're working with 400 companies. And basically, they're just tooting their own horn a little bit. This is a press release. But I think it's worth noting, because the scale of this in India is so big. And they sort of know, if you look at the number of students who have taken boot camps, for example, in the US, it's still in the 10s of 1000s. It's not even after 10 years of many companies doing it and already have one company with 22,000. You know, already I think the scale of this upskilling movement in India is really something to keep our eye
on. Yeah, that's my India is that there's been a kind of culling of the herd, you know, during the tech winter, and you're starting now to see who the winners are, and I think up grad, and you know, all of these companies have had fits and starts just given the exponential rate of growth, but it does seem like upgrade is successfully positioning themselves. And there's a really good product market fit in the Indian ad tech market like you just said, going to Africa. We are just seeing so many innovative stories out of Africa, and Northern Africa, Sub Saharan individual countries. But the idea of rapidly opening up access has been the theme. And you lesson this week announced an Open University with courses for undergraduates that focus on technology. The basic deal here is that anyone across all of Africa would be able to get an undergraduate degree set up for technology skills, and there's a booming offshore business in Africa, supporting Europe, Asia, US to a smaller extent. But the goal here was to make high quality education accessible, and to help students succeed in the digital marketplace. This is both a combination of like, career based education, as well as, you know, universal access through internet I liken this to, you know, their moment of skipping the telephone poles and just going straight to mobile. They're skipping building the physical universities and going straight to digital online learning. And it's exciting because so many young people live in a few countries in Africa. I mean, a friend of mine is an entrepreneur in Nigeria, and just the sheer quantity of young people there, they get an incredible addressable market. So very exciting stuff in Africa. Any thoughts there, Alex, just really quick. We
recently interviewed two past winners of the ASU GSB cup, including our friend of the podcast, Opay Buccola. And she is doing chemo school in Africa, which is tech training. We've also seen, you know, AltSchool, Africa we have next for university. The thing that stands out to me is that there was this dream at the very beginning of the sort of MOOC movement that if the Harvard's and Yale's and MIT is in Cal Tech's opened up their learning, it would go all over the world, and everybody would immediately learn it. And there was some truth to that. But it was sort of people with a huge amount of cultural capital all over the world would find out that that was happening. The big realization of that movement is that people still prefer and know a lot more about their local universities than ones across the world, even if they're Harvard and Yale, believe it or not. So the fact that there are these homegrown African online universities and boot camps and tech training programs is really exciting, because I think they're going to grow faster and hit more people in those countries. The other thing that's worth calling out, I learned this myself just a few weeks ago, but the African continent is enormous. There's four countries that are considered the Big Four, because they account for more than 70% of the venture capital investment in the continent. I didn't know this. This is from a TechCrunch article there, Nigeria, as you just mentioned, Ben, Kenya, South Africa and Egypt. Now, that doesn't mean that that's the only places that are booming, but it's probably good to keep our eye on things that are happening in those four countries. And things that are spreading to the rest of the
continent. Also, like Egypt is very centralized in terms of, you know, at tech innovation and follows a little bit more of a playbook for Mena and, you know, Saudi Arabia, UAE, whereas Nigeria and South Africa and Kenya are market economies where it's going to be very bottoms up. So very interesting moment here, and probably one to dive in even more with experts on the African and tech landscape. Let's talk a little bit about Europe, a great article about the rise of ad tech innovation in South Eastern Europe. You know, we covered the Photomath acquisition by Google, which by the way, it was not in any press release, it was discovered because of a lawsuit in Europe. But this idea that Croatia that, to be honest, there's an outsourcing engine and Poland, Ukraine, we're seeing Italy, Greece, lots and lots of innovative companies, especially in southeastern Europe, the kind of lower cost to develop products, the lower cost of developers has been a real opportunity. This article really points out that it's an engine of this region is an engine of innovation on the continent.
Yeah, it's really exciting to see a tech blooming in regions that are that are sophisticated and have a lot of tech but have not traditionally been ad tech players. We saw a few weeks ago a Moldova and ad tech raise a million euros in funding. This article goes through things in Romania things in North Macedonia, which is a tiny country. It's really, really exciting that ad tech is becoming so global. And I think this creation, you know, Photomath being in Croatia has drawn some spotlight to that region, the same way that goes student has drawn spotlight to Austria and Poland is brainly is in Poland. That is sort of having an effect when people know about these companies are Kahoot in the Nordics one want like Best Place to Work in Norway contest, and it's like when things like that happen. It makes edtech as a whole field more appealing to recent grads. It makes it more appealing for entrepreneurs and it's just really exciting to see Southeast Europe As a region starting to embrace it, so we definitely wanted to call that out. There's a few more headlines. We're running really low on time. Do you want to do a little lightning round?
Lightning round? So let's go big tech tic toc is adding a dedicated feed for STEM content. You play the game. Is this for promoting learning using its viral platform? Or is this strategy to fend off regulation by the US government? Also LSU issued a statement on academic misconduct after a college athlete influencer Olivia Dunn promoted AISC helped her on ticked up. So ticked up in the news, it's going to be part of our learning ecosystem. Ban or no ban Tiktok is here to stay. Next topic, Alex.
Yeah, that headline, it has to be the winning headline for like most modern headline in all of tacrine. It's a college athlete who's an influencer promoting an AI tool on Tik Tok. You Yes, the next topic is the teacher shortage. We've been talking about this all year. And there people are, you know, now here we are in March coming towards the end of the school year, and people are starting to think about what the next school year will look like. Some interesting headlines around this. One is that, you know, the four day school week has been something that some districts and some schools have been trying for a while. And there was an interesting headline this week about, you know, the sort of from the Washington Post about like, teachers like the four day workweek, it gives them more flexibility in a time they really need it. Does it affect student learning? And I think it's an open question that we're going to be seeing some more data about at the end of this year. Because if it doesn't, if that's good for teacher flexibility, but it might mean something a little strange about it. Well, you know, what does that mean, for schools? If it does, does that mean that online learning is working as well, or hybrid learning is working as well, there's a lot of questions in there. And I think we want to keep our eye on that four day workweek policy. Another one is editorial from Ed surge, basically saying, the way to keep the teacher shortage down is to give teachers their time back, which is something that we as edtech, entrepreneurs and investors and all of that can really, really take to heart how do we give teachers their time back. And then another was about it's been International Women's Day last week, and there's been some really interesting articles about the gender disparity in education. One, there was a terrific Ed search article that basically said, Let's compare the amount of superintendents that are male and female, it took a little inspiration from fun, I mean, dark and awful, but funnily framed article that basically says there's more, you know, CEOs, I think, named John than women, CEOs, something that basically says like that disparity is so big, and they basically found that men with one of 15 names, so that's, you know, Matthew, Kevin, Steven and Michael are 27% of superintendents, and all female superintendents are also 27%. And so you have three quarters of superintendents in the US being male. And there's a real disparity there that I think we all need to think about. The other thing is female teachers are paid less than male teacher. So this, there's a lot to think about here. And we don't need to breeze over it. But we will definitely link to these articles. In the show notes, it's really interesting to think about the teacher shortage and how pay can be useful, how respect can be useful, how flexibility can be useful, how giving teachers their time back, just and removing gender disparity are all techniques that we could use as a country to keep the teacher shortage from being a national crisis. And
this is why our opening around the potential for AI is so important, because we really tend to focus on AI directly for the learner. But as Kristin was talking about the ability for AI to actually be a teacher's assistant, expand the capacity of the educator, that may be the killer use case in formal education, and these kind of moves to have four day work weeks, or, you know, augmented pay, or give teachers time back, the rest of the capacity and capability has to come from somewhere. And you know, the promise of AI really needs to be focused on giving teachers that capacity to be great educators, and not have to go through all the hours of drudgery that I remember, as a middle school teacher from back in the day,
instead of Ben Franklin, maybe they just need a killer teaching assistant to do all the paperwork for them all the lesson planning and they get to do I've literally
had three conversations this week about people who want to build that thing. So I'm very optimistic that that's coming and when it does, and when anything happens in edtech. You will hear about it on this weekend ed tech, we're going to wrap the show. We're not going to do m&a and investment highlights today. But we'll include that in our newsletter. See you all at an upcoming event. Subscribe and listen to our other episodes. A thank you and shout out to Kristin from Khan Academy. Such an insightful interview is so timely. And thank you all for listening. Thanks, Alex. Talk to you all soon.
Thanks for listening to this episode of Ed Tech insiders. If you liked the podcast, remember Number Two read it and share it with others in the tech community for those who want even more Ed Tech Insider subscribe to the free ed tech insiders newsletter on substack