How AI Helps L&D Scale

[00:01] Tom Moriarty: Welcome back to the Secret Society of Success. In this not so secret podcast, we explore the changing landscape of corporate learning and development so that you can bring successful L D to your organizations. Here in season three, we're taking on a very hot and controversial topic generative Artificial Intelligence. In each episode we'll be talking to different L&D experts about what generative AI is, how it is already being deployed for learning design administration today, and frankly, whether or not you should be scared. Oh, by the way, we use Chat GPT to write this intro. Hey Sourabh, thanks so much for joining us today, man.

[00:42] Sourabh Bajaj: Yeah, thank you so much for having absolutely, absolutely.

[00:46] Tom Moriarty: So before we dive too deep into our conversation today, why don't you tell our audience a little bit about yourself, your background.

[00:53] Sourabh Bajaj: Yeah. So I'm Sourabh, I am currently the co founder of Uplimit. We just changed our name from Uplimit to CoRise. So if you've heard of us, that's CoRise before and we're like a startup that's about like two years old now. But prior to that, by way of background, I was an early engineer at Coursera, kind of did everything on the engineering side that kind of was required for the product. Learned a ton and then took a detour. Went to Google research, primarily working on AI research and some of the applications that you see today, but felt very compelled to come back to education and startups. So excited to be back in that land with awesome, thanks.

[01:41] Tom Moriarty: So I mean, at a high level you've been working in development and engineering in both education and then specifically Google in the AI space, really focused. What inspired you with those two experiences to start CoRise?

[01:59] Sourabh Bajaj: So what inspired us was if you look at education, traditionally there is content that you're teaching and then there is a lot of services that are involved. So these can be like how do you give feedback to learners, how do you answer student questions, how do you create content, how do you generate checks for understanding all these pieces that were kind of slightly harder to scale. And what made learning special was this motivational human connection aspect. And 2012 when we started working on online education, the trade off we had to make was things that were scalable stayed and things that were not that scalable kind of were pushed a little bit to the side. And so that's why you saw a lot of rise of content as the king that was defined as learning. So this would be like rise of content libraries and obviously there are certain fraction of people who are extremely successful with that model. But over time, if you fast forward time like twelve years, you see that there is tons of content available now. But there is this interesting opportunity with generative AI to rethink a lot of the services components that came with education, that make education very meaningful. And interesting. So how can you give feedback to every learner in a very personalized manner? Can you create content for every learner in a very personalized manner? Can you answer student questions? Can you create content like 100 X faster? We felt like there was, like a leap that needed to be had in education. Overall, we're still focused on adult education as a space, but if you extrapolate that to higher ed, can we make a real dent in cost of college or things like that? So felt like the two worlds I had been part of could intersect in a unique way in the world we are in today now, where there's a lot of change happen.

[04:06] Tom Moriarty: Yeah, it sounds like it was kind of experience meets a market, opportunity meets something that I think you're motivated to do. I really like the way that you describe the way to look at education. Right. So there was this rise of focusing on content and content being the key to education, but there's all these other services that an educational or learning provider actually brings to the table in order to even be able to deliver the content. And I think the interesting thing is how well you noted that the thing that makes it special is actually the human interaction. Right. It's what the human is able to bring to that learning experience in the form of the content feedback. Check in some of the other things that you highlighted, and I think you're probably right. Right. In the time that we're in now, with the capabilities around generative AI, there's probably an opportunity to allow trainers to spend even more time on sort of what is special by spending less time on other things. You mentioned scale a lot. I'd love to get into in a little bit more detail how as an expert with generative AI, sort of how that solves for scale. Before we do, let's set a groundwork. So from your perspective, what is generative AI and why should educators not be horrified and shaking in their boots?

[05:39] Sourabh Bajaj: Okay, so is the world shifting and changing 100%? Right. And we would be in denial. And change is hard. Like, change is hard no matter who you are, where you are. Traditionally, what we used to call AI was think of it as a pattern recognition problem, where every time someone did this, i, like, predict this. And the realm of options you predicted on was kind of defined. So think of it as like, when I call an Uber, I know the price can be from zero dollars to one thousand dollars. Nobody's taking an Uber that's more than $1,000, at least I don't know of anyone yet. So you kind of knew the universe of predictions. So let's say but now with generative AI, you go to Chat GPT and you say, write an email to Catherine talking about XYZ and inviting her to the party on Saturday. Then that email has never been written before. So what the model is outputting is something very unique and new and that is why we call it kind of like a shift in what we can do because it's kind of generating something that it didn't get exposed to in the past. What it's still doing internally is…what it's learning is given these words what is the next word? And it's kind of like generalized saying when people talk about an email, it's kind of roughly this structure. When people say this reference question to me, I can kind of patch this answer here. But underlying the model's core job is to given a set of words, predict the next word. So now I think I still feel like there's an opportunity with AI to think about, how do you elevate yourself to the most strategic, higher value work? In some sense, where what makes teaching or learning special for anyone is the moments you have with other humans, where it's motivational. It feels like personal. It feels like somebody had a life changing impact on you. But there is a lot of logistics work we are used to doing. Okay, so take an example. Let's say like you're a fifth grade teacher. The questions you answer in year one, very likely, very same to the questions you answer in year three. But the current system doesn't propagate like work you have done last year to this year. Like maybe you wrote a guide, maybe you wrote some explainers. But generally there's a lot of information loss year to year, year to year or class or cohort to cohort supplement as a platform helps you run live experiences at scale. So this can be like inside the company you doing onboarding. This can be like customer education. This can be you leveraging some of the courses we have on the platform. If you've done the work once, we don't want you to do the work again. We want to make sure that this is getting easier for you so that you can think about how do I make this course more richer? Are there more interesting topics to be included? How do I focus on every student in a more interesting manner? So I think it's like for me it is like we will shift roles to more strategic roles and also to what makes us more human. And those both seem like positive outcomes.

[09:20] Tom Moriarty: Yeah, for sure. Yeah, I think so too. And I appreciate that it's helpful to get a background from someone who's worked in it and lived in it before. And I think that that hopefully helps bring barriers down for our audience in terms of understanding where this fits for them as a tool. I love your takeaway of what it enables a teacher, a learner, a trainer, an instructor to do which is really scale themselves, right? Try to use this tool so that you do not need to repeat work or spend the same amount of time doing work that you historically have done, so that you could take that time and spend it on the more strategic, valuable exercises, allowing you to spend more time working on the important and less time working on the urgent. I'd like to understand how you guys are using it AI at scale, right? So if you could, why don't you walk me through a specific course that you guys build. I think when we were preparing, we talked about a course that you guys have building on building AI products for software engineers. Paint the picture, walk me through the background of what that course looks like and how it's run and the resources you guys use to run that course today.

[10:37] Sourabh Bajaj: Yeah, no, that's a great example. And the class you mentioned is in collaboration with OpenAI on basically how do we educate more people to come into the AI space. So let's take the lifecycle of a course as our starter example. So the first phase generally for anything is like creating content. And where today people sort of try to think about is like first order effects. Like I can ask it to write an outline for me, I can ask it to write a section for me. But imagine if you extrapolated that to what we are pushing towards is like you tell us what course you want and if you have some seed material you can give us that and then we can get you to an 80% solid first draft where the entire course is laid out for you. It comes with a ton of checks for understanding, it comes with educational content, video, images, whatever you need. And those are kind of like auto generate. So it is a seamless experience that gets as a first draft now where we think it's strategic for humans to get involved is like adding in personal nuggets, adding in stories. When I was like here this happened. If I'm teaching a class on AI oftentimes I'm like back in the day when I did this job, I did this and that's what actually the audience finds extremely meaningful. Then there is the part of creating content, then there's part of delivery where if you think about delivery, I could give content out to people. But then there is interesting things around like how do you scale personalized nudging? How do you nudge every person? How do you make sure that every person can get feedback on the things they're submitting? How do you make sure that everybody is getting their questions answered? So we have this tool called Cobot, which is kind of think of it as like an AI teaching assistant where what it does is it's good at answering questions that might be about the class. It'll exactly cite references from the course content. It'll send you personalized nudges, it'll send you reminders if you miss the lecture, it knows that you missed the lecture and it'll send you an automated message saying here's the recording. So it kind of offloads a lot of work that you do with this one character that is kind of like doing work in the background for you. For numbers we ran this class with over 6000 people and the instructors were mostly just doing the live teaching. So think about in a typical L&D role, that would be like the SME time, and then there was like one course manager and a couple of TAs so overall I would say like 6000 people with twelve to 15 human hours for the entirety of a week. So it's just like there is so much we can offload to AI in interesting manner. This was the largest class we have ever run so we learned a lot and we also there are obviously things that we want to improve but we do see that we can make an order of magnitude leap where the first classes we ran were like 30-40 people. Then we tried like 100 people class, then we tried like a 500 people class and this was like the biggest we have tried till date.

[14:06] Tom Moriarty: How many people were supporting this?

[14:11] Sourabh Bajaj: Five people in the overall team but two people are just subject matter experts doing like 1 hour of live lecturing. Then there are three human course manager TAs who are basically working with the AI to assist them and you can imagine they spent like 10 hours total combined.

[14:31] Tom Moriarty: And then talk to me about the course itself.

[14:34] Sourabh Bajaj: Right?

[14:34] Tom Moriarty: So what's the structure of the course? How long does it take?

[14:37] Sourabh Bajaj: It's a week and a half milestone where you build like a real world project. So the project in the class is let's say I subscribe to my favorite podcast. Oftentimes I miss episodes so it automatically listens for the podcast releasing a new episode, creates a summary, writes a newsletter, email for me and says this was the speaker, this was what they talked about. Here are key takeaways if you want to listen to it, go here. I can do that for either YouTube channels where I'm subscribed to 20 YouTube channels and don't watch everything they post. I would love to get like a two minute summary before I decide if I want to watch this or not. So that was kind of like the goal for the students, and people went very creative and thinking about how do they auto connect with other podcast tools or can they spread this across YouTube, spotify, et cetera, et cetera. And so obviously with that a lot of questions come up, et cetera. But there is where we lean a lot on AI plus like a community experience since we are cohort there's like a rich community there that can answer. So we think about how does AI work in tandem with this community aspect of learning.

[15:53] Tom Moriarty: Yeah, let's go back and I think you addressed this already but in the design phase of the course. So this is now the biggest one you've run. There was 6000 people. In the design phase, this team of five people, two are the main lecturers. Then there's the three assistants. Where did they inject themselves and then where did they let AI get them that 1st 80%? Talk to me about from the content design for the week long course. Week and a half long course.

[16:26] Sourabh Bajaj: Yeah. So in general, how we think of this is you never start from scratch. It's very hard to stare at a blank piece of paper. So you basically start with this first draft. Then what you're doing is you're checking, you're kind of reading and personalizing it to your personality and tone a little bit where like, hey, I want to talk about this. I don't want to talk about this. Or maybe I can paraphrase what I wrote here. So imagine you got like an 80% draft and then you were playing more of the editor role rather than the writer role. It's like an elevation from the writer to the editor.

[17:07] Tom Moriarty: From a design perspective. So let's say there's 100% of the time that it took this group of five people that would normally, whatever, ten years ago, build this course.

[17:21] Sourabh Bajaj: You're probably talking about weeks normally. Like two, two and a half years ago, when I was manually writing a course, it would take around like twelve weeks to create like a three week class. I am not the fastest writer, things like that. And I would write, rewrite, get writer's block, things like that. Now when I'm editing, it is so much faster and I get the first draft in five minutes. And then I'm like, oh, this structure makes sense. I don't want this. I'm just kind of like shuffling things around and saying like, oh, I want to paraphrase this. I can still use the AI to paraphrase it. So it's like, even in the editing experience, there's a lot of AI opportunity, but it kind of shrinks the time from twelve weeks to kind of getting ready in less than a week. Even with like, oh, I want to change things. So you definitely have an order of magnitude leap there.

[18:28] Tom Moriarty: Yeah, I would say twelve weeks to one weeks is definitely an order of magnitude. Okay, so from just focusing on the areas where the people delivering the course in your platform are using AI, we probably went from, let's just say as an example, we went from a twelve week period to a one week period for content generation. Right. Building the course outline, building all the content. And the reason for that is because you let AI get you to 80% and you spend the time on the remaining 20%, rounding out the edges, adding the right contextual components. I thought that was a huge takeaway earlier that you mentioned. Right. Adding in that last 20% where you're giving examples, you're telling stories, you're adding in things that can contextualize the general content, but make it specific to the audience or to what's relevant to them at that moment. Which in my experience is often when some of the best learning takes place. Right. So really in terms of content creation, you're enabling the team to produce the same output probably in one twelfth of the time, essentially with the math that we just covered, but also probably as we're potentially even more effective. Because that one twelfth of the time that's being spent is really being spent on that last 20% to 25% of the content, where it probably resonates most with the audience.

[19:49] Sourabh Bajaj: I think that's the other thing is like you don't have the fatigue of the eleven weeks.

[19:53] Tom Moriarty: Yes, that's a great point. Yeah. Whatever you're outputting in week eleven and twelve is not nearly as good as it was in week one and two. That's for sure. We've all been there. Yeah. In this class in particular, there's this rather large project. There's some administrative burden involved in most classes, right. Like making sure the people are there, that they have the right access to the right information, that they're in the right groups. You said you guys use communities and things like that. So where does team of five leverage, where do they insert themselves and where do they leverage AI as it relates to the administrative parts of the class?

[20:30] Sourabh Bajaj: So let's think about it. So let's start with traditionally when you wanted to do like a live teaching experience, you would have caps of like 30 people, 50 people, because beyond that, the administration of did someone get a calendar? Did someone find a zoom? Everybody asking, where's the recording? Can you send me the link to the recording? Too hard. So we kind of built guardrails on I'm not going to do this for more than 50 people. Now, if you take the system basically is automatically figuring out is the event on everybody's calendar if they miss the live session, are they getting a personalized message saying here's the recording, is the recording coming with very naturally spliced up where you are saying here in this chunk, the instructor talks about this part. So people are having a much easier experience watching the recording itself. Then the student has a question somewhere. But the question could have been answered in the lecture live, it could have been answered in the content, it could have been answered previously in the community itself. So the AI TA can pick up all the information that is being created and say, oh, I know how to answer this, if I don't know how to answer this. It's surfacing the 10% of questions it can't answer to the TA saying, hey, can you answer this one? I don't know yet. So if you teach me how to answer this one now, I'll learn it for the future. And then it can handle both logistics questions as well as academic questions, which is very interesting in the sense that the logistic questions actually are kind of equal part compared to academic questions that come up in a large classroom. And then I think what you want is the AI to not jeopardize community. So if people are talking to each other, people are having a rich discussion, we make the AI pull back on saying hey, Tom and Jonathan are talking, don't intervene because that's a much richer human experience. And I think that's where today you could imagine going much further. How you think about AI's involvement in the live session itself? How do you think about when you run the class again? What does the AI carry over? But also knows that the logistics have changed. So the answer of the logistics question last time is not the answer of the logistics question this time around. So there's just being aware of where in the class people are who has gotten stuck, where let's say people have asked they're doing a check for understanding they've gotten it twice wrong. Can you surface the right hint for them? How we've historically done it is we simplify the questions we asked, which is like generally multiple choice so that you could hard code the hint that was like the pattern. But can you do the same for an open-ended question and say like, hey, give feedback, this is the desired outcome. So we right now have this system. We can give you a hint on an open-ended programming question so people can make mistakes in hundreds of ways and we can still automatically surface you the right hint as a learner because that kind of unblocks you and makes you more likely to succeed. Like people get frustrated and drop off and can you proactively catch them? And then I think the last piece is like how do you check in on students? Who do you check in on? How do you nudge people to make them aware? Those are all places where in a 6000 people classroom that would be impossible to scale without AI and a ton of automation.

[24:29] Tom Moriarty: There's so much in there I love that as it relates to the logistics. I think you're right. I think you said it perfectly in the beginning, right? If you try to think about the logistics associated with organizing an audience of 6000 people to get them in the right place, to know where they are, to minimize the questions about the technology not working in the first 35 to 40 seconds of the class. That's why you were, as you said, traditionally limited with class sizes that usually max somewhere between 30 and 50 and you never see that anywhere. And I think the obvious scale that AI is able to deliver is increasing the size of the audience that you can serve.

[25:12] Sourabh Bajaj: Important to keep in mind is you could counters by saying hey, if I just put it on YouTube, it's infinitely scalable, like a million people can watch it together. But then you want the engagement of the 20% classroom to be the same as the 6000 people classroom. So how do you keep the bar high on the engagement and pedagogy metrics or the learning success? While scaling is, I think where I think there's opportunity because almost YouTube is the most scalable format. Whatever baby shark has been watching like 9 billion times.

[25:52] Tom Moriarty: Yeah, no, but you're right. I think that where the gap here, the gap you're trying to bridge is how do you scale the interactive classroom experience?

[26:01] Sourabh Bajaj: Right?

[26:02] Tom Moriarty: And that's really what uplimit. But I think where the opportunity lies for anybody in the learning and development space is to replicate the experience that really does make an impact and makes a difference for your learners of that live facilitated experience, but start to be able to deliver that at a scale that was just previously unimaginable. Right.

[26:30] Sourabh Bajaj: Two things I would add to that is like if you're a learning and development leader, the two problems you run into often are resources you have on your team. There's never enough resources to go around. And some of the side effects of that is that you create exclusivity unintentionally for the learning experience you can provide to people. So you coach every manager in the organization is much harder and you can coach every employee in the organization on AI. Right. Now, I'm sure there's a lot on people's minds around that, but being able to scale effectively has been like a challenge. And that's where we felt like that was an interesting area of work.

[27:14] Tom Moriarty: Yeah, Sourabh, that totally resonates. We for many years, Mimeo would put together a state of learning and development survey almost every year. One of the most consistent challenges for learning and development leaders, professionally speaking, is that they are underresourced under resourced in some way. Whether that's underfunded or a smaller team size, the resources they're given in an organization to drive an impact do not equate to the size of the audience or the size of the problem they're trying to solve. Right. And I think the exciting thing if I'm a learning and development leader listening to this, is that the primary takeaways that I'm hearing for them is that tools like Uplimit and generative AI in general is something that will allow you to scale yourself. Right? That's the first thing we talked about by spending less time on tasks and elevating yourself to spending time on the more strategic tasks. Like the example of using generative AI to generate 80% of your content, spending the rest of the time on the 20% and being that much more effective because you're not beaten down from spending twelve weeks on that one course outline. Right. So you could scale yourself, you could also scale your audience. Right. You could scale the size of the audience that you're able to support in a classroom session like you're talking about in this 6000 learner example. Because practically you can because you have a tool that can help out with making sure the right people are in the right place.

[28:54] Sourabh Bajaj: I think a couple of things I would add there is the thing is we're also in this unique situation now with learning and development where the roles of everybody in the organization are changing. Like how marketers did marketing pre generative AI is going to look a little bit different from how marketers do marketing with generative AI. How many unique assets can you create? What levels of personalization can you create? How do software engineers work? How do sales teams work? And there's just going to be implications. How do you measure for performance when say like 80% of the code is being written for you? How are you evaluating a software engineer then? So they're also suddenly on the hook for supporting all these roles in the other organization that they support to be able to make that leap. So the expectations are higher, the resources are still lower. And I think that's where we think it's like all collectively get creative on how do we solve these problems. Right? We are like one example but it's just like a lot of creativity needed in figuring out highest impact, highest ROI based of doing things.

[30:16] Tom Moriarty: Yeah. If there's a one sentence takeaway that I think our audience can have in terms of why they should be figuring out how to leverage generative AI and tools like Uplimit, it's because it will exponentially increase the total business impact that they can have. Right. Practically speaking, if you can increase the amount of content that you generate because you're spending one twelfth of the time generating it and you can increase the size of the audience that you can effectively serve because the logistical components are brought down. But you can do that while maintaining an environment that allows you to have feedback, have prompts, and have an awareness of where your audience is and help guide them through an experience that still feels like that special human. Interaction where there is a true learning moment that you can create the scale of the business impact that a learning and development professional can have when leveraging generative AI tools versus someone who is not. It's going to be orders of magnitude larger.

[31:27] Sourabh Bajaj: One last thing I would love to plug in is we would love to continue the conversation and we run this AI for Learning and Development course on Uplimit. That's a free course. It's done in partnership with Karie Willyerd who used to be the CLO at Visa, and Jenny Dearborn who used to be the CLO at SAP. And they're both phenomenal leaders who talk about how they have thought and supported their organizations through technology shifts and also just kind of dive deeper into some of the things we were touching upon.

[32:03] Tom Moriarty: Sourabh, we'll definitely get that link from you and include it in our show notes so the audience can easily access that specific course and jump in. And learn a little bit more if they'd like to. Where can our audience go if they want to learn more about you and more about Uplimit, specifically, where's the best place to point them to? Where can they get more content from you?

[32:24] Sourabh Bajaj: So I'm on LinkedIn, Twitter, whatever social media platform there is, I have an account, so you can find me there pretty easily on Uplimit, the best place to find us is Enterprise. That's the best place for anyone to connect with us.

[32:44] Tom Moriarty: Awesome. Well, Sourabh, thank you so much for your time. I think that there's a lot here. This is obviously a very hot topic these days in all walks of life, but especially in learning and development. And I think that there are some real key takeaways that our audience can have that hopefully will make them think more seriously about using generative AI and have some very specific ideas on where in their life they can use it and what kind of impact it can have. So thank you for your expertise and your time, and we really appreciate you sharing. We shared today.

[33:17] Sourabh Bajaj: Awesome. Thank you so much for having me.

[33:20] Tom Moriarty: Thanks for listening to The Secret Society of Success, a podcast by Mimeo. To find out more about how corporate L&D teams use Mimeo for smarter content distribution, visit www.Mimeo.com. Also, don't forget to subscribe to get our episodes as soon as they launch. Enjoy your day.