This transcript was prepared by a transcription service. This version may not be in its final form and may be updated.
Zoe Thomas: Welcome to Tech News Briefing. It’s Monday, August 28th. I’m Zoe Thomas.
Julie Tang: And I’m Julie Chang for the Wall Street Journal.
Zoe Thomas: Over the past few weeks, we’ve been looking at how generative artificial intelligence could impact education and shape learning, but there is also an area that’s making money off this boom. Education technology companies and their investors. This school year, students and teachers will have new AI tools like ChatGPT and Bard at their disposal. Julie, you used to be a teacher, and now we’ve done some reporting on this, how much are these AI tools expected to change the basics of learning?
Julie Tang: Well, these AI programs are unlike tools we’ve seen before because generative AI can do just so much, and those abilities mean that it can essentially touch every area of education, even areas as basic as reading, writing, math.
Zoe Thomas: So today we’re going to hear from a startup that’s trying to ride the AI wave and from a venture capitalist who knows this space well and can tell us why edtech investing is entering a new frontier.
Julie Tang: By the way, this is Episode Three of our series, Reading, Writing, and Algorithms. You can find the first two episodes linked in our show notes.
Zoe Thomas: One way edtech startups are using generative AI is to create new or personalized versions of tools that schools have always used, like books.
Julie Tang: LitLab.ai uses artificial intelligence to make individualized stories for children with a goal of helping them learn to read. It was founded a year ago and used to be called Koalluh. It released a product for parents last September, and then at the start of the year, it pivoted to teachers. It’s founder and CEO, Varun Gulati, says it’s currently working with thousands of students.
Zoe Thomas: What makes LitLab stories different from something written by programs like ChatGPT is that it’s built to generate what’s called decodable content. That means the stories are easy for children learning to read because they can sound out the words, think Dr. Seuss’s Cat in the Hat or Hooked on Phonics. We wanted to understand a bit more about how that works with AI, so we invited Gulati to come on the show and tell us more about what decodable content is.
Varun Gulati: Decodable books typically repeat a phoneme or sound over and over again. It aligns with the phonics and other foundational literacy instruction that’s grounded in what’s known as the science of reading. They’re super effective at teaching kids how to read, but the challenge has always been to make them engaging and to make them relevant for kids. And other than Dr. Seuss, a lot of the decodables out there are frankly quite terrible. They’re boring for kids to read. Would you like a decodable story?
Zoe Thomas: Absolutely.
Varun Gulati: Okay.
Zoe Thomas: Go ahead, put Dr. Seuss to shame.
Varun Gulati: Let’s see. And we’re going to practice a story with Leo. Leo’s going to be a penguin that lives in a castle.
Zoe Thomas: How are you choosing those things?
Varun Gulati: This is part of our story creator. You can choose the character, you can choose the setting. You can choose a phoneme as well that you want to practice. So we’ll pick the phoneme A-R. Okay, and it’s generating. So in a far off land, that’s the phoneme right there, ar, there was a castle. It was not just any castle, it was a grand castle with a tall tower that reached up to the stars. So we’re seeing that phoneme again and again. In this castle lived a happy, playful penguin named Leo. Leo was not just any penguin, he was a kind and clever penguin who loved to have fun. One day, Leo found a jar near the castle’s gate. He was curious about what was inside the jar, so he opened it. Inside the jar, he found a marker along with a map. The map led to a hidden treasure buried far away in the castle’s garden.
Zoe Thomas: Is it also generating pictures as well at the same time?
Varun Gulati: Yeah, it’s generating pictures at the same time.
Zoe Thomas: Oh, why not use Dr. Seuss, though? Why spend money on a different program?
Varun Gulati: Yeah. One thing I want to reframe over here is this is not instead of Dr. Seuss. Dr. Seuss is fantastic. I love Dr. Seuss. But what’s interesting is that Dr. Seuss sits in a category of what I would call literature. And there’s a lot of curriculum and content and instruction that exists outside of that, and that’s typically not personalized or differentiated for kids. And that’s the challenge. A lot of educators spend a lot of time and energy tailoring instruction and tailoring content to meet each student’s individual needs. That means their interest, that means their particular skill profile, and that’s very time-consuming. If we can differentiate content in a way that’s deeply personal to every student and meets their particular skill profile and needs, then you can really accelerate reading and make reading far more engaging for kids.
Zoe Thomas: And does the underlying system work in the same way as large language models like ChatGPT?
Varun Gulati: We do use large language models in our system, but there’s a lot of engineering we’re doing beyond just the standard out-of-the-box large language models. And the reason for that is that the LLMs today are not aware of the progressive skill buildup within a phonics instruction scope and sequence. That’s where we come in. We specifically engineer for this use case of decodable content.
Zoe Thomas: Is this program something that a school, school district, or individual classroom would have to purchase for themselves?
Varun Gulati: Yeah. So, our primary revenue source is a subscription-based model for schools and districts. We don’t believe in charging teachers directly, for equity reasons.
Zoe Thomas: I want to ask about that, because it is a bigger question when it comes to a lot of technology and particularly generative AI, you’re still going to have school districts that have more money to afford programs like this. So how do you get students to engage with reading, to get that motivation, as you say, if they’re not going to have access to these tools?
Varun Gulati: Yeah, absolutely. And there will always be a portion of our product that will be free for teachers and for students to access. We don’t want this to be something that is inaccessible to learners, because ultimately the goal over here is not just to improve reading engagement, but to improve reading proficiency as well.
Zoe Thomas: In this effort though, to attract more schools to your subscription model, your paid model, is there concern that some of these school districts are going to be cutting back on their budgets and that edtech and particularly new edtech might be the thing that goes first?
Varun Gulati: Absolutely. In fact, I’ve spoken to admin who are in this mode of cut, cut, cut. But here’s what we know to be true. Technology that is effective, that teachers use, that is integral to their workflows and valuable to teachers and students, stays. We’ve heard this again and again. For the products that teachers love most, the admins find a way to pay for it.
Zoe Thomas: That was LitLab CEO, Varun Gulati.
Julie Tang: Coming up. Investors are also betting schools will find a way to pay for these tools, and they have lots of startups to choose from. After the break, we’ll hear just how many AI pitches a longtime edtech investor has been getting since the beginning of the AI boom.
Zoe Thomas: Welcome back. I’m Zoe Thomas for the Wall Street Journal, and I’m joined by TNB producer Julie Chang.
Julie Tang: LitLab is not the only company using generative AI to try and reimagine traditional teaching.
Zoe Thomas: And the capabilities of these tools aren’t just catching the eyes of schools. They’re also attracting investors. Jennifer Carolan is a partner and co-founder of Reach Capital, a VC focused on education technology. Reach is actually a backer of LitLab, though Carolan wasn’t a part of that investment.
Julie Tang: Before becoming a VC, she taught middle and high school in Chicago for about seven years, and she’s been funding edtech startups for over a decade.
Zoe Thomas: Since Carolan has been investing in this space for so long, we wanted to get her thoughts on how investors are approaching the AI boom. So one of the first things I asked her was about the pitches Reach has received this year.
Jennifer Carolan: So, in the last six months, we received about 260 generative AI companies coming into our pipeline. Teachers spend about eight to nine hours on average per week creating content, organizing it, presenting it, getting it ready to deliver to their students, and generative AI has just such great potential in helping teachers do this. And then also in assessment, the ability for generative AI to automatically grade or assess content is also very exciting, because that’s another task that teachers spend a lot of time on.
Zoe Thomas: You mentioned over 200 companies with generative AI in your pipeline. How many were you getting before ChatGPT came out?
Jennifer Carolan: None, actually, in the generative AI space.
Zoe Thomas: How do you separate good investment in this space from hype? Because there is a lot of excitement around generative AI right now.
Jennifer Carolan: One of the things that we’re really looking for is proprietary datasets, and does the company really understand the educational process? Do they understand their end customers deeply? And so that’s in a lot of ways not different than how we would look at these companies prior to generative AI, but I think it’s going to become even more important.
Zoe Thomas: Are there specific risks for edtech that you wouldn’t need to think about if you guys were investing in, say, delivery or the medical field or something else that wasn’t education based?
Jennifer Carolan: One of the biggest risks that I come across is that children are a vulnerable population, period, and at different parts of their development, they are experiencing different cognitive changes. So we think a lot about the talent that’s in these startups, and do they have expertise and behavioral psychologists and experts in child development that understand the unique needs of the developing mind? We see a lot of companies that are developing AI tutors for kids, and they are essentially trying to replace teachers and teach kids how to read, for example. Will it be harmful? Probably not. But is it the best tool to teach a child to read? Is it better to have a human there with a tech tool alongside them to support them? Probably. Because again, there are different needs of a developing child.
Zoe Thomas: Jennifer, thank you so much for joining us for this conversation.
Jennifer Carolan: Thank you so much. It’s a pleasure.
Julie Tang: As we’ve been hearing from educators throughout the series, they’re planning to teach their classes differently, and students will have access to these new tools. That means schools will likely have to rethink their learning goals, because AI has launched academia into a new era.
Zoe Thomas: And that’s it for Tech News Briefing’s special series Reading, Writing, and Algorithms. This episode was reported by me and producer Julie Chang. We had additional support for this series from Chastity Pratt. This episode was mixed by Jess Fenton. Our supervising producer is Melony Roy. Our development producer is Aisha Al-Muslim. Our deputy editors are Scott Saloway and Chris Zinsli. And Philana Patterson is the Wall Street Journal’s Head of News Audio. Join us tomorrow and the rest of the week for regular episodes of TNB. Thanks for listening.
