Brilliant launches Koji, an AI tutor that teaches through interaction rather than explanation

there a couple rounds.

There we go.

That's hard. That's a hard one.

That's a hard one.

Anyway,

expert mode. Uh, great to catch up, Nico.

Awesome progress, fellas.

Thanks for coming on the show soon.

We'll talk to you soon.

Goodbye.

Cheers.

H, very fun. That is the hardest one to hit the glasses. Anyway, we have our next guest, Sue Kim from Brilliant. She's the co-founder and CEO and she's here with us in the TV panel. What's going on, Sue? How are you doing?

Hey, thanks for having me. I'm doing great.

Thanks for hopping on the show. Uh, since it's the first time on the show, I'd love an introduction on yourself and the company and the launch and uh for you to just take it through take us through it.

Yeah. Uh, so I'm Sue. I'm the co-founder and CEO of Brilliant. Uh, we just launched Brilliant's new AI tutor. Uh, his name is Cooji. He's a little green guy. He helps you learn how to think and problem solve in math and coding. Uh if you come check out the product, everything starts out uh very visual and interactive. And this tutor can see how you're interacting with all the concepts and the problems you're solving and point and sketch and annotate on the screen with you. Uh so we call it a graphical tutor because it's designed to feel like someone sitting next to you looking over your shoulder and ready to help. uh talk about how people have processed the intersection of AI and uh and education to date. It's been such a big uh such a big selling point overall. Uh and personally, I think it's most powerful as as a way to ask like the dumb question that you might be uh embarrassed to ask, you know, a friend or an actual teer teacher or even even a tutor. uh but how have you kind of processed it? Obviously, you can use the base models themselves to to learn, but there's a lot of missing functionality and opportunity for you to kind of um you know, extend out that kind of raw capability.

Yeah, totally. You know, I think that lowering that barrier of embarrassment is huge. That's something that we hear a lot h that you know the student who would never ever raise their hand in class and ask a question or you know even with a tutor it's just kind of like I still don't get it like can you explain it a third way it's hard to do and you know I think that these models have helped a lot with that like lowering that embarrassment and one of the things that we haven't yet seen them do is design for something other than the moment of explanation

you know when we designed this tutor, we designed it for that moment of understanding, not for the moment of explanation. And there's actually a lot of research on tutoring that shows that when h human tutors don't just sit there and explain stuff to you, the tutoring conversations become a lot more interactive. And then students just do a lot more of the work. And you know, these like magical outcomes of why tutoring is so broadly effective. You know, ultimately a lot of that just comes down to the learner spending more time on the material. uh and you know if a tutor talks too much and just like spits out tons of stuff for you to read uh that is worse than the student doing stuff and just sitting in the confusion. So I think it's a combination of how much can you engage the student to actually do the work and then you know lowering the embarrassment improving their motivation and engagement and just getting to the point where they feel sort of confident at tackling these things on their own. Do you feel like you have a duty to make learning addictive?

How would you something similar?

Because because every every you know uh there's a lot of big companies out there. I won't name them. They certainly want people to spend as much time in their applications as possible.

It's like every company, but yes.

Yeah. Not not every company. There's there's there's uh but let's say like the the algorithmic video feed, right? Like they want as much of our time as possible. I feel like you have a duty to take to try to make learning as addictive as possible but because it's a positive addiction right I want people to be obsessed with understanding the world furthering uh their skills etc. Yeah, you know, we talk about this a lot like a a big marker of success of the tutor is whether or not he can make himself unnecessary. And you know, the the the goal is that you scaffold the learner and allow them to ask questions until you get to a point where they can do it for the for themselves and then you want to

get out of the way. So you know one of our goals is that if we are doing a good job over the course of learning a concept coach should become totally unnecessary like you should be asking the questions that he used to ask you and you know one of the things that it like I find a funny parallel is that on our board is uh someone who started the dating company okaycid and then h went on and you know became the CEO of match was on the board of Tinder and he's like you know the funny thing about consumer dating apps is that if you're successful, churn is built into the product. Like you want people to turn because like you don't want people just in the app just going on dates all the time and not, you know, meeting that long-term partner.

Yeah.

And so, you know, we do measure whether or not people turn and but we measure it very differently from how most other consumer products would think about it.

But I feel like I feel like learning is much more learning is way more continuous. Like everyone should aspire to be a lifelong learner whether whereas like maybe it's not so healthy to aspire to be a lifelong dater you know somebody that's just like I never want to you know I don't feel like there's there's never been a

subject that I've like learned about where I've just run out of things to learn.

Well that's interesting that happens to me all the time. I feel like I'm like yeah but I'm talking about

I've done I know everything about I know everything about science I'm kidding. No, but I mean course you're joking, but there is specific things where you're like, "Okay, I've learned enough about nutrition." Totally. Totally. And I don't really care about it too much, you know, let's say, but like I've learned the basics and I can move on to the next topic. But there's so many topics. For me, that's nutrition. Every single point,

every single place that I've gotten to in terms of an overall understanding, I'm like, "Wow, this goes deeper than I thought."

Yeah. Uh, how are you thinking about multimodality? I feel like uh the the big labs all have different ways to turn out a result that's not just a big block of text. There's an image, an infographic. We saw with Google IO, there's the omni model that can show a 10 second video diagram or an explanation, a whiteboard lecture. And it feels like there's maybe a gap right now in that you have to come to the model or the chat app and ask for what you want. You have to say I want this as an infographic or I want this as a video. Or you might even have to go to a specific model or a specific company. Like if you want video, you go to Google. If you want an image, you go to chatbt. Uh, and I'm wondering if you're thinking about how that fits into uh your process of actually picking the right tool for the job.

Yeah, it's very uh, you know, working with a chatbot to learn is a very proactive process and it's one of those things where h when you use a chatbot, the chatbot gets very little data on whether or not you've actually learned the thing. Like the typically the interaction with the chatbot is you ask your question, you get your answer and the exchange is over. And so for a lot of these like homework help type learning queries, uh people answer uh people get their question answered and then they leave. And so the model gets back no data on did the learner actually learn this thing. And you know going back to the point about you know addiction the thing that we try to get people really immersed in is like tons and tons of like dense real time reps uh to uh in learning a topic so that we can self-improve and hyperpersonalize in real time and to do that you have to have a dense user model you have to have a learning graph you know that's not going to come out of a model company anytime soon uh the UIUX matters a lot too it has to be purpose-built or teaching with all the features that come along with that and also you know just some of the stuff also has to be purely deterministic like you have to be correct uh and you have to guarantee that correctness and so I think there's a lot of things that ultimately just add up to a lot of friction uh when you are asking the user to pull and sort of scaffold and structure their own learning whereas you know our philosophy is we're going to pull you in and give you things to do that are the right next thing for you to do.

Amazing. Uh, where can people get started?

Uh, go to brilliant.org. Uh, and we have apps as well in the App Store and the Play Store.

Great. Well, thank you so much for coming on.

Great to Great to finally meet you and congrats.

Have a great rest of your day. We'll talk to you soon.

Thanks.

Goodbye.

Bye.

Uh, before our next guest joins, we have a new uh a new release, new information from Sam Sulk. He has chimed in on the debate over should you work 5 days a week or seven days a week. He says, "You only need to work 5 days a week. Don't overdo it. You do not need to work 24/7. Overtraining reduces performance. Most of the time, you're not working on the Manhattan project. Work hard, recovery, keep some balance." So, wise words from