The AI Tutor That Makes Kids Actually Think | E2298

A lot of VCs behaving badly stories trending all week. Matthew Prince from Cloudflare, I think, started this all off. I keep presenting my Series A slides to an unconscious man in a Herman Miller chair. Oh, I was on the other side of the table raising money myself. I'll give you a couple stories. This other motherfucker begs a friend of mine to give him the meeting. I'm finding out that the meeting's been canceled, that instead of just letting me pitch, you're so dumb that you would cancel on me when you knew I was flying up from from Los Angeles. You're going to be the worst venture capitalist of all time.

This Week in Startups is brought to you by Shopify.

Turn those what-ifs into sales with the e-commerce platform powering millions of businesses. Sign up for your $1 per month trial today at shopify.com/twist. LinkedIn. Thanks to our partners at LinkedIn. Post your job for free at linkedin.com/twist, then promote it to get access to LinkedIn Jobs' new AI assistant. And Northwest Registered Agent. Get more when you start your business with Northwest.

In 10 clicks in 10 minutes, you can form your company and walk away with a real business identity.

Learn more at northwestregisteredagent.com/twist. All right everybody, welcome back to TWIST, This Week in Startups for June 8th, 2026. Got a big docket today. There's a bit in the news, a lot of VCs, uh, behaving badly stories, trending all week on Twitter. I'm gonna get on into that at the end of the show when I go off duty and talk about the social media buzz. And today I have two amazing guests. One, an old friend, uh, and, uh, an incredible entrepreneur who has, gosh, been running an incredible company called Brilliant.org. We'll get into that. And before I do anything, I like to just take my Plaud. I turn it on, boom, I press it, a little haptic. Now I'm recording everything I say and I can take notes. I can say, hey, action item, catch up with Lon about the launch of this new This Week In Show we're doing. Let's make a strategy session, blah, blah, blah, blah, blah, A, B, C, D. I do that when I'm rucking. That's like a hike with weights, or I'm skiing around the mountains. And it's just been amazing. Even when I'm in the airport. This product, Plaud, has got millions of users now. It is the perfect product. When you get home and you're on your desk, you just plug it in, it syncs, and now you've got all your notes. Easy peasy lemon squeezy. It'll make a mind map for you, a template. All the great ways to just summarize information and then use it and never lose one of your great ideas or one of your to-do list items. And you never have to take your phone out of your pocket. That's the best part. You're just ambiently recording stuff. And you know me, I was like, hey, I don't know if I want this stuff in my life. You know what? Everybody's doing it. Everybody's recording everything. Everybody's making summaries out of everything. So the horse has left the barn. This technology is here. Everybody's using it. So I have to adopt and adapt. And I've adopted Plaud if you would like to get on the Plaud train with me. Hey, very simple. Uh, go to plaud.ai/twist, plaud.ai/twist, you'll get, uh, 10% off. I am so excited to have Sue Kim, the CEO and co-founder of Brilliant, back on the program. She's been on this program, gosh, she's got to be in the 5-timer club. She's an incredible founder. I met her, uh, when she launched a little company called, hmm, AllTuition. I think that was the name. And then she pivoted it to something called Brilliant.org. My bestie Chamath and I have been investors in the company for over a decade and they are crushing it. Welcome back to This Week in Startups, Sue Kim.

Hey, Jason, you're so famous now.

I know, I was like micro famous and then now I'm like somewhere between micro and whatever the next level of podcast fame is. But it is true. I go through an airport or I go to a Knicks game and people want to take a selfie. Yeah, world's number one podcast. I mean, it's kind of weird. We're manifesting here, uh, but you're on the OG podcast, This Week in Startups. Was it All Tuition?

Was that the original name? Yeah, All Tuition was the original name. 3 nerdy kids, uh, out of Chicago trying to fix student loans.

And that was like 12 years ago, I think. I think my first investment is when I was a Sequoia Scout, so it had to be 12 years ago before I had a fund. It was a great idea. What was the original idea, uh, for All Tuition?

Yeah, so the original idea was a sort of lending tree or bank rate for private student loans. Uh, student loans are the second largest loan that people take out in their lifetimes, and there was no way to shop for it. And so students were getting stuck with these, like, usurious rates that were hidden under all of these terms and fees. And so we read through thousands of pages of docs and rationalized all of that and, uh, built a, built a system to compare rates and then, like, immediately got cease and desist letters from all of the biggest lenders who are like, no, no, no, no, No, don't do that. But yeah, at peak we were processing $100 million of student loan consolidations and helping people get cheaper rates. And so there's a long story to why we ended up working on Brilliant instead. But yeah, it's a really important— student debt continues to be a huge problem.

Yeah, you found a huge problem. You gave it a shot. You got some level of product market fit, but you found a better idea. I remember Chamath was one of the judges, I think, on your panel, fell in love with the team. I think he had a follow-up meeting with you, whatever. You were brainstorming and you came up with Brilliant.org. And I think you even camped out at the original Social Capital office for a couple of months or something. Is that right?

Yeah, we worked out of that office. I mean, credit to Chamath for, uh, for, for Brilliant becoming what it did. I, I think that he looked at altuition and felt that we were going to end up in the same spot as, you know, Intuit or, uh, H&R Block. And he said, you know, you're just going to end up in the spot where you are going to— your incentives are to lobby the government to keep student loans really complicated and expensive so that you can make money saving consumers, uh, on their interest rate. But, you know, even if you do that and you make a lot of money and it's a great business— and he's like, I believe this can be a great business, I believe you can make a lot of money, uh, the problem is you're not going to actually fix student debt. You know, student debt needs to disappear. And, you know, helping people find cheaper rates and shop isn't going to make the problem go away. And the problem isn't that the rates are expensive. The problem is that they shouldn't exist in the first place. Uh, and so Brilliant was born out of a really first principles thinking exercise of, you know, the question he asked was, if you could build anything in the world to make the biggest difference you could possibly make, what would that be? And then that began the journey of, you know, gracefully helping people manage the loans that we were already managing on All Tuition, but also, you know, starting up a new business that today is Brilliant.

Yeah. And so tell people a little bit about Brilliant.org, what the mission is and how it's evolved over the last decade.

Yeah. So Brilliant is putting a world-class tutor in every home. We launched this product last week. Thank you for helping us amplify it. And, you know, it went incredibly viral. Like, this thing had almost 5 million views on X. We had no expectation that it would go as viral as it did. You know, we were launching a learning product on a Friday. You know, everyone's starting summer break. All the commencement speakers that are talking about AI are getting booed off the stage., you know, AI is immensely uncool and unpopular with consumers. And at some point it's like, well, you know, the product is good. We just have to push the button. And so we launched and then I, I think what we learned is that, you know, AI that makes you think is an incredibly popular idea.

Uh, AI that makes you think, that makes you, that challenges you, the human. Yeah. AI as a coach, AI as a guide by your side. That AI people seem to like.

Why? Yeah, yeah, I, I think that, you know, everyone's seeing the data, especially American parents see the data that our kids can't read or do math anymore. AI is making this worse. We are building AI that says it can do the thinking for you. And I think in that, uh, environment you still have parents who overwhelmingly believe that learning hard things still matters. And they're worried that their kids aren't learning hard things at school. And they're worried that this assumption that we used to have that you could send your kids to school and they're going to, you know, learn to read, write, and do arithmetic, like maybe that's not true anymore. And so we went out with this message that we are building an AI that makes you better at thinking, uh, was incredibly resonant with parents. And, you know, our takeaway from this is that people aren't anti-AI. That is like a very broad brush to paint people with. You know, people are anti-idiocracy, they're anti-IP theft, they're anti being replaced in their jobs, they're anti-slop. But if you show them a product that will help them and their kids, like, they can tell and they love it. And like, you know, it seems obvious when you say it, but like, if you use AI to help people, uh, it turns out that's incredibly popular.

It does get painted with one large brush, and if it was paper You'd be like, yeah, some of this paper is really good to write poetry on. And yeah, some of it you can make a flyer for some horrible cause and pass it out and get people to do terrible things. You can write a bank note to rob a bank with paper. You can do a lot of things with paper as well. Toilet paper, write poetry on paper, write a, I'm robbing this bank on a bank note and rob a bank with it. So yeah, people do. Paint with a wide brush with it, and it's much more granular than that. I, I completely concur. So this new product's come out, we're about to show it, but previous to that, you made a site where people can learn difficult math problems, science problems, level up, gamification, and all of that. And over this 10-year journey, my understanding is a large number of people use this product, but an, an very significant number of people pay for this product. And now people have told me, oh, you're in brilliant.org, they have that at my school. So before this AI launched, after all tuition, there was 10 years in the middle where you were building this incredible website, brilliant.org. I guess the best way to describe it would be a website that people can do self-paced learning to learn math and sciences, level up, get awards, and parents could pay $100 a year for it to feel great about their kid getting better at math, science, STEM, et cetera. Yeah? Okay, so you have identified a real problem and you put together a solid solution and a business model that you believe in. So you're all set to launch your new company, right? Uh, not so fast. If you want investors and potential customers to take your new business seriously, you need to consider forming a Delaware C-Corp. And that's where Northwest Registered Agent comes in. They're going to give your new company a real identity. That means an address for your public filings, a domain, a custom website, a business email, and of course a phone number. And that's going to take just 10 minutes and 10 clicks. They don't charge hidden fees, customer service is available around the clock, they're not overwhelming your inbox with spam, and they make it easy to cancel at any time. So get all the advantages of a Delaware C-Corp, independent regardless of where in the US you're operating from. Visit northwestregisteredagent.com/twist for more details, and the links are in the show notes.

Yeah, totally. Uh, and you were saying, you know, you've met parents and teachers and people in the wild who use this product. And the real differentiator is that school math, what you're doing is you're learning formulas and you're learning a procedure, and then you apply the procedure and then, you know, maybe you do great on the test, but then you sit down in front of a problem where the procedure that you learned doesn't apply and you have no idea what to do. And the difference is knowledge versus problem solving. And what Brilliant teaches is problem solving. Problem solving is a very transferable skill, uh, much more transferable than, you know, knowing how to compute, which is a very brittle way to learn things. It's a very difficult way to learn math. It's kind of, you know, memorizing the procedure works for a little while, but then there's more and more procedures, and then it just feels harder and harder. Um, when you learn problem solving What you're learning is how to see. You are learning how to see the structure of problems. And, you know, people develop these systems brains. You know, kids can develop these systems brains. And that's what learning math or coding trains. You get really good at seeing how to apply AI to your life and work. You have intuition for what kind of problems can and can't be solved, what's easy, what's impossible. How to get started. And so, you know, we started as kind of like the world's largest online math club where we treated math as this conduit to becoming a strong thinker and problem solver, getting kids to like reason through hard things. And, you know, just the stories we hear from parents are incredible. Like, we get so many emails from parents being like, you know, my kid wasn't that into math and then they got into Brilliant and they skipped a grade at school. You know, my 7-year-old, this guy, you know, actually they tweeted about this too. Vinita and Parag. Parag was the former CEO of Twitter before Elon took it over. And, you know, he reached out to me and he's like, I just wanted to meet you because, you know, my 7-year-old is doing, you know, 5th grade math on Brilliant. He's like obsessed with this app. He's like learning every day. And this was pre-Tutor. And then I guess he ended up in the Tutor beta, and apparently he ran to his parents and was like, guess what, Mom and Dad, I have a personal tutor now. And like I can ask Koji questions, you know, if you're busy. And so, you know, it came out of a belief that problem solving is really important. And kids, if you, if you call them to a high place and set high standards, can do hard things.

Yeah, I've learned that. I got 3, 3 daughters. I have learned that challenging stuff. I typically do it with a hike in the rain. There's no bad weather is what I tell my kids. There's only bad clothes. So you've got the wrong clothes, but humans can walk in the rain. Uh, when we get back from a quick, uh, sponsor break, two questions for you. How did you choose going after consumer versus trying to sell into schools, which I see a lot of education companies do? Number two, how did you pick the pricing? Because you picked the pricing point, $20, $30, $40 a month depending on which plan you get. That's not cheap when you compare it to a typical app, which They typically cost $60 a year. We'll answer those questions when we get back. Right now, I want to tell you about row.co. They got GLP-1s. How do I know? Look at me. When Sue met me, I was 40 pounds heavier, 43 to be exact. And now I am svelte. I'm back to my running marathon weight. Listen, it might be for you. You should check it out. That's my best advice. I don't give financial or medical advice, but let me tell you something. Being 40 pounds lighter and then putting on a 35-pound weight vest and skiing 30, 40 days a year and feeling great. My Lord, I should have done it earlier. And, um, nearly half of row.co members get their GLP-1s covered by their insurance, and you can do that with their free insurance checker, row.co/twist. R-O-W dot C-O slash twist. And, uh, see if your insurance will cover it. I am a big fan of being thin and fit. All right, Sue, when we, uh, left you we were talking a little bit about pricing and who to go after. You went after the consumer and you went with premium pricing. Tell us about those decisions and the debate maybe with your co-founders and the team that went into making those two very specific decisions.

Yeah, I think that we have always at heart wanted to get as close to the metal of solving a problem for the user of the product. Uh, as we could. And one of the problems with selling to schools is that you're pretty far from the actual learner. You know, maybe you sell to a school district and then the school district assigns the teachers to use your product and the teacher assigns it to the students. And now you're multiple steps away from the learner. You know, the nice thing about a scaled consumer app, you know, people solve millions of problems on Brilliant every day, is that you have a direct relationship with the learner. And you are learning every day, are we getting better at teaching these topics? And I think that's really— and are we getting better at teaching these topics in a way where students will self-direct and continue? Um, in a classroom, kids are forced to use the thing. And so I think it almost gives you license to build a product that, uh, is not inspiring to use, that kids don't want to use. And if they hate— end up hating math and never want to touch math again, so what? And we really wanted to align our incentives with the people who we were building for, which is the learners. And I think that's been a great place.

B to C, really big mistake because you're distancing yourself. You're putting a B between your business and the consumer. When you have an app in the App Store, you get these like 2 or 3 really interesting pieces of granular feedback. Obviously you get like engagement stats, great. They're very granular, but you also get people who write ratings. Yep. And use it, use the rating system as like customer support helpline.

Yeah.

So you're like, I can't get this to work, 1 star. And then you also have people unsubscribing, you have people asking for refunds, you have that commerce part of it, you have trials. So talk a little bit about that feedback loop and, and making the product have market pull, which you did at some point. You reached not just product market fit, but the market pulling., you know, the product out of your data centers and onto their phones?

Yeah, the App Store reviews are gold. Uh, we still have multiple humans reading every single App Store review. Uh, we, you know, get tons and tons of email from customers, parents, you know, telling us exactly what's working, exactly what's not. And that's another thing that you wouldn't get if you did a B2B business here is you're not getting that real-time feedback on how is this experience working for my kid. Uh, and let me tell you, like, parents are very happy to tell us exactly what they want to see in the product. They, you know, make extremely detailed feature requests. Like, they will, you know, send us photos of their kids and videos of exactly what's working, what's not. So, you know, just from like a product roadmap development perspective, like, nothing beats scaled consumer. Like, the, the volume of feedback is incredible. And also, you know, I think that it provides a really good North Star of what should we build. You know, we launched this tutor positive vision of AI learning really resonating and just reinforcing, you know, the cost of tutoring is a huge pain point for students or for parents. The cost of tutoring is a huge pain point for parents, but every parent wants to help their kids learn.

Adding a new member to your team is a crucial decision. You don't want to rush into a hiring situation that you will regret. But if you're a busy founder, you also don't want to spend a ton of time in the trenches looking for that perfect candidate. Instead, you need a partner and you need a trusted partner. And LinkedIn Hiring Pro is that partner. How do I know this? Because I use LinkedIn Hiring Pro. You want a real-world testimony? All right, here you go. We recently hired a new customer success manager. They handle all the advertising accounts here on This Week in Startups, and they ensure that we're keeping all of our wonderful partners happy. LinkedIn helped us connect with an amazing candidate right here in Austin, and He had exactly the combination of experience we were looking for. It's magic. It's alchemy. It's everything for me to find a great team member. So hire right the first time. Get started by posting your job for free at LinkedIn.com/twist. Terms and conditions apply. That's LinkedIn.com/twist. They'll do anything to help kids learn. They'll do anything.

They'll do anything.

That's, I think, the thing we had an early discussion. I was like, listen, parents, will work an extra shift to pay for a tutor. Like, literally, a tutor in Silicon Valley or even here in Austin— I have like 2 or 3 different tutors that come every week. Yeah. And I think it's $150 to $300 per like 90-minute session. We tend to have like 2 or 3 of the daughters go to the chess tutor, the math tutor, whatever. Yeah. Uh, music tutor. Uh, it's not cheap. What's the average tutors in like cities?

I think the average tutor, you know, let's call it $80 in Middle America if you're not getting a super high-end tutor. But the problem with that is, you know, you need high-dosage tutoring for it to be really effective, meaning you're spending that $80 for an hour, you know, maybe 3 times a week. You know, families can't afford that, especially if you have 2 kids. What are you going to do?

Yeah, it's— I mean, I look at it and I'm like, huh, is my tutoring budget exceeding the budget for tuition at, you know, the school. And it's like, yep, it is. Um, so that makes pricing easy. You priced Brilliant versus a tutor as opposed to versus casual games or, you know, other apps in the app store that are typically, I think the going rate is $60 to $100 a year for an app experience.

Oh yeah. We don't, we don't benchmark to games. I think that it's very telling that the main pricing question from our customers is not why is this so expensive? It's why is this cheap. Oh, like, we, we are, we are benchmarked to tutors, and people understand a tutor for a kid is $10,000 a year. And, you know, we have been working so hard on this because we wanted to bring that cost down by 95%. Like, we wanted to be able to offer a product for $30 a month that could do most of the things that you would hire a tutor to come to your house to do, uh, make it way cheaper and make it way lower hassle.

All right, so AI emerges a couple years ago. Everybody's like, oh my God, it's the death of everything, everything's gonna die. I mean, if I had a nickel for every single technology that was going to kill Robinhood, Calm, Brilliant, and Uber, you know, all these breakout companies, Thumbtack, it was cloud computing, mobile, blockchain, now AI, it's going to kill everything. Um, the truth is you're thriving more than ever with AI. I'm assuming internally you can build the product even better.

Yeah, I think that AI has been a huge tailwind for us. It's created this divide where AI clearly is going to either make you smarter or it's going to make you passive and dumb. And everyone's going to have to make this choice, like, are you going to use AI to do less or are you going to use AI to do more? Are you going to become a better thinker or a worse thinker? And, you know, being able to launch a product that makes you a better thinker in a time when you know, we think that AI used well can make kids into geniuses is like a very exciting mission to be working on.

Let's do a demo of the new AI tutor. This is the copilot, I guess, is the way to think about it. And I'm curious, am I on the existing Brilliant website and you added the AI to that, or is this all new experiences?

I would say that we have significantly redesigned the experience around there being a Socratic tutor present in the lessons. So let me go ahead and share my screen.

Yeah. I'm cur— and also just as we get through this, how long did it take to build? I think everybody's always curious when a legacy company, which is crazy to say you're a legacy company, like, but you are an established player with, you know, significant revenue. You know, is it a pivot? Is it an addition? But you, you are, you have to maintain the existing business and then add this new business. You're basically quote unquote disrupting yourself. It's not exactly a pivot, it's just a new feature, but maybe how long did it take to build this and architect it?

Yeah. So, you know, the thing that you're going to see is that LLMs have a very constrained role in this product. Uh, Koji's designed around what LLMs do well, but all of the, you know, pedagogy, the mathematical correctness, that lives in Brilliant's interactive lesson infrastructure, which is more deterministic, and we've been working on that for years and years. And, you know, part of what makes this product sing with AI is that, you know, since 2019, literally since GPT-2, we have been designing an interactive, uh, canvas infrastructure where every interactive canvas has an API that LLMs can read and write to. And so, you know, this isn't just like AI came out of nowhere and then we bolted it on to product, this is, you know, at this point, 7 years of building infrastructure that enables a tutor to tutor you graphically on the page, see everything that you're doing, uh, understand your intent and provide good pedagogy.

Hmm. Okay. And the name of the tutor on brilliant.org is—

This is Koji.

Koji. Koji, made up word, sounds Japanese maybe, or Korean. What's the background here? Because I think there's— I've been to a Koji Korean fried chicken at one point somewhere in my travels in Seoul, but I can't remember.

Yeah, we have a lot of international usage. So about 40% of our users in the US, but 60% are all around the world. And so it was important for us to pick a name that feels accessible to everybody. This is actually also how I named my children because my Asian parents have to be able to say my American kids' names. And so, uh, you know, We did a big exercise where we wanted sort of short names that are easy to say, easy to remember, and landed on KOJI.

Love it. K-O-J-I?

K-O-J-I, yeah.

KOJI, I love it. It's a—

does it—

it doesn't mean anything? You just found a word? Does it have a meaning in a language?

It does have a meaning and a backstory, you know, maybe not worth getting into, but you know. We like the sort of scientific backstory of koji, the fungus, which we can talk about another time.

Awesome. Listen, you are definitely living up to your nerd credentials here if you named it after a deep pool of nerfungus who are completely underrated. Let's see, show and tell here. This is just like the old days when you debuted at the Launch Festival. Oh, yeah. 14 years ago. I think that was it, right? Was it TechCrunch 50 or was it Launch? I can't remember which event it was. At the time. Yeah. You launched that.

A decade ago. Um, all right.

It was over a decade ago. So, here we go.

Yeah. So, Koji is Socratic. Uh, this is, you know, somewhere in the middle of a quadratics, uh, lesson. So, I've been learning with Koji. So, let's, uh, let's start this lesson. Let's keep going with filling in both terms of the product, but this time, let's add a coefficient to the mix. Complete the expression. OK, so here I am, and I see— oh, man, there's y squares here. So I see 3y squared, and then I'm supposed to fill in these 2 blanks. All right, so let me try this. Oops. Let me try that. I see 3 y's here. Is that correct? I see those 3 y tiles in the bottom left. If you look at the whole rectangle, how many separate regions contain y tiles? I'm not really sure what you mean by regions. Let's start with what we know. The 3y² term comes from multiplying 3y and y.

Can you fill in the two linear terms?

Ah, okay. So there's, there's two of these. I'm going to put 3y there. And 2y there, and 2, and then combine those. OK. Exactly.

All right.

Let's remove the tile labels this time. OK. So now it's a little bit harder because it's not telling me what the tiles are. But I just did a problem like this. We got the 2z squared. I'll count these. So there's 7 of these, 7 of these yellow things, and 3. Okay, let me check that. Here's what I want you to notice. We're multiplying 2 binomials using the distributive property and then combining the like terms.

Neat, right?

I don't really get it. Can you explain that again? I don't see how the graphic actually corresponds to the stuff below at the equations. The main idea is that the area model and the algebra show the same thing.

Notice how we have two separate regions with z tiles.

In algebra, we combine those like terms into a single 7z. And so, you know, a few things to notice here. Again, Koji is going to guide you to understanding. The responses are tailored so that he's not just throwing very, very long explanations at you. The conversations are designed to be interactive and have students do a lot more of the work. He can see everything you're doing on the screen. He can draw and annotate on the screen. It's like having a tutor sitting next to you, able to sketch directly on the screen. And then as you go through the lesson, a lot of this visual scaffolding gradually falls away as you're getting to mastery. And then by the end, you're in like a, you know, test-like environment where it's just the problem. And you have to do it without help, and then CodeGuru disappears and you're on your own. And that's where we want to get you. We want to get you to the point where the tutor is there for teaching you the concept, but then disappears when, uh, when it's time for you to demonstrate that you can do it by yourself.

Did you know you can set up your own dropshipping business easily with Shopify? Imagine running an online store selling whatever you like without having to worry about inventory, shipping, or, you know, paying for space in a warehouse, which means, yes, all you need to do to get started is pick a product category, find a supplier, and set up your Shopify store. All the infrastructure you need, from third-party apps to Shopify's seamless checkout process, is already in place. So you can just start moving units in just hours, not weeks. That's why we're creating a This Week in Startups merch store, and we're going to use Shopify. Shopify is the e-commerce platform behind millions of businesses around the world. In fact, they're powering 10% of all e-commerce in the US, from household names like Heinz and Mattel to companies that are just getting started like yours and mine. It's time to turn your what-ifs into sign up for your $1 per month trial today at shopify.com/twist. That's shopify.com/twist. And there's obvious advantages here. The tutor is available anytime the student has energy or a window. That is key when you start having kids.

Yeah, scheduling.

The tutor's coming at 4 PM, the kid didn't sleep well last night, whatever it is, they're in a bad mood, their friend is— all their friends are going for ice cream, they're not, they got the tutor. Like, it sounds silly, but you have to be in the right context, the person has to be ready to learn. And, um, if they're going too slow here, I assume it can take them back a step, et cetera, and adapt this for them. The question I think a lot of people are gonna have is, is this like just throwing Claude or OpenAI or Grok or Gemini against Brilliant? Or what did you have to do to that AI? And I don't know which one it is, or if it's headless and you've tried Deepseek and Pub, you know,, uh, open source versus frontier model, and then how do you harness it to constrain it to not go off on tangents?

Yeah, so it's fairly, really, it's fairly model agnostic. Uh, and the thing that we have versus frontier models is frontier models don't have verifiable reward signals for tutoring, and we do. If you look at frontier models, and we benchmark this very carefully, their ability to tutor well hasn't improved much since like '01. Uh, these models are much better at following instructions and using actions, but you know, this core job of tutoring, which is diagnosing and fixing student misunderstandings, that's plateaued. And our belief is that until you have a source of real verifiable rewards for RL, we're going to be stuck here. And what we have with our tutoring session is that we have actual learning loops, and the reward signal for us is whether the student actually understands. And we have now tons and tons of that data.

And is that how you trained it? Is you said, hey, look, here's a bunch of data for this problem, here's a bunch of data for this problem, these are the questions we typically get, this is where people get stuck, and it just magically came up with a tutoring, um, methodology? Or did you have to teach it a certain methodology?

We had to teach it a certain methodology and it was a huge schlep, you know, literally for every concept we had expert teachers in that concept essentially instruct the model on which tool calls to make, what the common misconceptions are, how to help a student in the situation. And not only that, but just work on the entire UI/UX of the lesson. So, you know, there, there's very little of this that we get for free from LLMs. What we do get for free, you know, as the intelligence gets better, the natural sort of real-time conversation, the latency, you know, the intelligence— like, we do get something with model advances. But, uh, you know, the architecture to self-improve and hyper-personalize in real time based on a dense user model and learning graph, uh, we don't believe that that's going to come out of a model company anytime soon. Like, you have to vertically purpose-built it to serve specifically the tutoring use case.

Yeah, that makes complete sense to me. How has it been received by users? And does this change your business model in any way, or is it just, hey, we just added a feature, enjoy? Or do you think this is a new paradigm that will allow you to teach people English literature philosophy, you know, I mean, you can be brilliant at many different modalities. I, I know you guys started in STEM, but if you did train a tutor and it's a big heavy lift, one would think there's another vertical or 100 verticals to go after.

Oh yeah. Yeah. Our vision is a tutor in every home, in every language, in every subject. And that means you have to have a tutor that works. It has to be at a price point that everyone can afford or even start out freemium so that you get some tutoring for free. It has to be in any language. We do get that for free with the models. So thank you to the people pouring hundreds of billions of dollars into that.

And how is it? Is localization accurate enough to feel confident selling it to other regions?

The localization is astonishingly good. And actually, some of these voices are employee voices. The tutor voices. And it is uncanny to hear, you know, people that I work with every day teaching me algebra in perfect Korean. The localization is very, very good. I mean, this is the stuff that LLMs just shine at. So in every language, in every subject. And that's what we're working towards. We cover math and coding now. We are getting very close to curriculum complete in math for middle school and high school. And then we're tackling college-level material. We're also going younger. And, uh, coding— we are getting complete on foundational coding this year, and then next year we're expanding into science and a lot more technical topics.

We have a portfolio company, MicroOne, uh, that does this learning for language model companies. How do you think about, you know, having the learning occur inside of Brilliant or working with those vendors who are currently helping, you know, uh, and there's a bunch of them, train the frontier models, or the frontier models now, I would suspect, are coming to you and saying, hey, you're filling this in and perfecting it, maybe we could plug that into our surface.

Certainly, I think there is deep recognition that we have a really unique data set. Uh, you know, we are— we have scale, we have tutoring sessions at scale with real students. And, you know, I think that this is like the ChatGPT moment for AI and learning. You know, we are a week into this, um, the progress has been extraordinary. And I think that, uh, the day when we are training models to be— training and fine-tuning models to be specifically very, very good at tutoring is near, and it's going to completely transform however learn.

What I love about what you're doing at brilliant.org, Sue, is you are taking a very world-positive view of AI. I believe the reason people booed AI during this recent graduation season— and it wasn't one, it was 3 or 4— yeah, um, is because they were the first generation to use AI in school, and I think they have some resentment for it in some ways that Everybody was kind of using these tools to— I don't want to say cheat, but to hack their way through school. And the teachers aren't using it. And there's no resolution to their future. Is their future that they just secretly use an LLM to phone their work in and have this big charade with the professors who are also using AI to make lesson plans? And they just feel like their entire college experience was university slop. And they're going to go into a world where their job will be job slop and they're not important in this equation. What this does is say, hey, we're going to teach you to be a systems thinker. We're going to help you learn how to solve problems with AI. You're going to become brilliant, dare I say. One of the great branding exercises ever was naming this company Brilliant. You're going to become brilliant. And when you become brilliant, the AI is your tool. You are not spoke in the AI wheel. I think it's, dare I say, brilliant what you've done.

Thanks, J Cal.

That was my— I sometimes have to give my little soliloquy and my little speech at the end so I can feed the Shorts algorithm. So excuse me if my performance seems a little bit canned and is less like podcast-esque. I try to do one of those every interview now to get the Short for my team. Um, how many people in the company now? I mean, you don't talk publicly about revenue, but it's, it's, you got a pretty big team.

Yeah, yeah, like, uh, about 70.

70 is great. I mean, for the amount— I'm— you don't talk, you don't release your revenue numbers or membership numbers, I take it?

Uh, no, I mean, we may hit some major milestones here soon and where we do that, but, uh, so far hasn't been public.

Got it. Uh, so yeah, I would, man, I would love to see some 9. I'd love to see a figure with 9 in it. I don't know, like, if that would be a revenue number or users or whatever, but I love when there's 2 commas and almost a third. That's one of my favorite moments in any startup.

Sue Kim, well, hang tight.

It's, uh, great to be on the cap table all these years and just to know you and watch the team crush it like this. And this is a true accelerant for the business. And this is what's important for founders to understand. If you have a company and you've been building all this time and you understand your customers, AI is jet fuel, you know, for your rocket ship. So just build a second booster, a third booster, whatever, and just embrace it fully and you won't get disrupted. You'll be the disruptor. And my Lord, if you can figure out how to teach philosophy or writing or other subjects And then instead of it taking 10 years to build a vertical, it takes you 2 years or 10 months. This business is going to accelerate dramatically.

Absolutely.

A lot of lessons there from today's interviews for founders. Number 1, uh, really understanding your customer and finding spaces where maybe it's not clearly defined, which means idiot VCs. And we're going to get into idiot VCs behaving like idiot VCs in a moment. Or entitled VCs. The issue with venture capitalists in a lot of cases, not all, and, you know, people like to dig on VCs, which is kind of lame in and of itself because they dedicate their lives to supporting founders and getting a bunch of capital from around the world and providing it. So it should be a little more symbiotic and there should be a love of VCs and You know, the, the part they provide in the ecosystem. But there is a valid criticism that if it doesn't fit into an existing TAM, very hard for, I think, some young inexperienced VCs to understand this. When you look at these two companies, like online tutor doesn't exist, uh, digital board game that doesn't exist, and the people who like the existing board games hate this, and the you know, serious gamers, it's not serious enough. And the Dungeons Dragons people think it's sacrilege. You have to ignore all that, and you have to ignore people who are trying to create a TAM for something that doesn't exist in the world. The TAM for Airbnb was basically zero. They induced the market to exist. The TAM for ridesharing, the TAM for eBay, uh, you know, eBay was like flea markets, but you couldn't really assess it to flea markets because it was global. All of this is to say, uh, listen to your customers and build a product that delights them, and everything will be fine. High-order lessons here. All right, um, let's get into bad VC behavior. For every story of bad VC behavior, you're gonna find an equal number of entrepreneurs doing crazy stuff. I've invested in 700 companies. I'm sure I've made mistakes over the last 14 years with them. I've gotten frustrated with founders doing crazy stuff. I've matured as an investor. I'm at peace with certain things. But VC has changed over the years. It went from the VCs replacing founders, putting in professional management like they did to Google. They were like, hey, Larry and Sergey can't run this company publicly. We've got to put Eric Schmidt in. To like Founders Fund being formed with the explicit purpose of Hey, we're going to be in the founder's corner and everything back and forth in this crazy debate. This crazy debate. Who's more at fault for bad behavior? Listen, anybody who's in the company building business is going to have sharp elbows at time. There's a lot of money at stake. It's a trust-based system. VCs, when you think about it, give millions of dollars, tens of millions of dollars, hundreds of millions of dollars. To founders and then trust them to do the right thing and report into them every quarter, maybe monthly, get some reporting. But you're basically taking this huge leap of good faith in giving founders money. And then founders are taking this other leap of faith that the investors are not going to block a sale down the road or block an investment down the road. And this is why standard documents and standards exist in our industry. I am a huge fan of standards, and when I see a non-traditional deal, I flag it in our database and with our team and I say, man, this is a non-traditional deal structure. And I beg the founder, please don't do this non-traditional deal structure. Please do not take money from some high net worth individual who then wants the ability to, you know, control the board or get 3 board seats. Just keep it easy breezy. If somebody owns 5%, you can give them a board observer seat. If they own over 10%, give them a board seat. You know, it's standard stuff, standard pro rata. If the person invests early, maybe they could get super pro rata and warrants. Oh, we want to keep our pro rata and we will put in money now and take this risk. This is what I do. Hey, we're going to put this money in now, take a little bit of risk. All we ask is that in the next round we can put a little more money in and then a founder can make that decision. Okay, do I want to let Jake Al come in for $100K now with the assumption he's going to put in 20% of the next round or $250K? Small price to pay because nobody else is doing early-stage investment. Anyway, Matthew Prince from Cloudflare, I think, started this all off. I'm gonna ask my team to just pull up the history of these tweets in the order in which this whole thing happened. And there is a very funny thing about these pile-ons that happen is they act as a cathartic release for the community. Everybody's got bad VC stories, everybody's got bad entrepreneur stories. Sometimes things bubble up and people just can't take it anymore and they have to explain how poorly they were treated or how well they were treated. And it's all being tweeted now and all the laundry is going out there. This happens every 5 to 10 years, by the way. My friend Adeo Ressi did a website. I forgot the name of it, but oh, I think it was The Funded. And he did an anonymous message board where people would wail on VCs. Okay, Greg Eisenberg. On the Twitter, I was once pitching in a boardroom at a top 3 VC firm for a $15 million Series A, 12 people in the meeting. One of the GPs fully fell asleep, out cold for 30+ minutes. Nobody acknowledged it. Everyone keeps going. I keep presenting my Series A slides to an unconscious man in a Herman Miller chair, and somehow that was considered normal. That's venture capital. You might fly across the country to perform for people who may or may not be conscious. It's a dance. And sometimes you lead and sometimes you follow and sometimes your partner is unconscious. If you're raising right now, just know every founder has a story like this. The process is weird. The power dynamic is weird. You're not crazy for thinking it's weird. No one talks about it because they want to continue raising, but I'm happy to stick my neck out. It is weird. Uh, 9.7 million views because everybody started retweeting this. Okay. Um, you have to consider when these tweets came out. VC behavior, like I said earlier, in the '80s and '90s was, hey, we're going to replace the founder, etc. Then in the 2000s, it started to change. How do I know? I was on the other side of the table raising money myself. I'll give you a couple stories. I was, I was pretty hot coming out of the Weblogs, Inc. sale. I sold the company for $30 million to AOL 18 months after starting. It was called Weblogs, Inc. They did Engadget, Autoblog. Those blogs still go to today. So very proud of that. And my partners, Peter Rojas, Ryan Block, Sean Gold, and of course Brian Alvey and our first investor, Mark Cuban. Went great. Then I did my next company. I emailed 3 people: Mark Cuban, Michael Moritz, John Doerr. I said, hey, I got my next company. I'm looking to raise $3 million. It's gonna be called Mahalo.com. I want to use wiki software to try to create a human-powered search engine. Very similar perplexity today. A lot of lessons there. I was way too early. It didn't work out. But I emailed Michael Moritz. He immediately emailed me back, called my two phone numbers, my mobile and my desk phone in my SIG, and, um, I was in his office two days later. That just shows you how great an investor Michael Moritz was. He introduced me to this new guy who had just started working there that year named Rudolf Botha, who became a lifelong friend. They funded it famously. But I also went to see John Doerr because I said, who are the top two VCs? I've never raised from venture capitalists. The last business I raised $300,000 from Mark Cuban returned $7 million to him. And we became great friends. And we were friends before that. But I never raised from venture capital. So I just said, let me just email the top two and see what happens. Short, two sentence. Hi, my name is Jason Calacanis. I sold my last company for $30 million in 18 months to AOL. It was called Weblogs, Inc. I've got my new idea. I'm looking to raise $3 million. It's a human-powered search engine. Boom, they both get back to me. Michael Moritz personally calls the phone numbers. John Doerr forwarded it to one of his people called and had them call me. I was in his office a week later. Just those two observations. One, the person I emailed called me back on my phone immediately within 1 hour. The second one was like a day or two delay. Still incredible. But John Doar didn't call me back himself. So that led me to believe Michael Moritz is on it a little bit more than John Doar. Then when I went to see Michael Moritz and Ruloff, Man, it was just very fast. They had me in a partners meeting the next week, et cetera. Lo and behold, I go, I get the meeting, and John Doar is there and two of his other partners. John Doar falls asleep in the meeting, and I told the story in my book Angel, and he's wearing a sling and he's not like fully knocked out. He's like a little groggy. And I'm looking at him and I'm like, I got to just power through this. But he's got like a scrape on his face that's like kind of not bleeding, but like raw. And he's got his arm in a sling and he can't move it. And he looks banged up. And so he leaves the meeting, whatever. And one of the guys pulls me over, grabs me by the elbow and says, "Listen, I just want to apologize for John Doar. He was out biking this morning. He flipped his bike in Woodside, went flying over the handlebars, went to the emergency room." came directly here from the emergency room. He's on a painkiller. He's got to go back to the emergency room to get an X-ray, but he didn't want to miss the meeting with you. Let that sink in. John fucking Doerr flipped his bicycle. Dipshit Jason Calacanis, with one or two wins under his belt, is at the office. An easy ability to say, I'm not going to make the meeting. He made the meeting. He did not off during it. I joked with him about it later. To me, this was the greatest compliment I've ever gotten. But I'll tell you two other stories since we're giving stories. This other motherfucker, this other melon farmer goes to a friend, says, hey, I hear J. Cow's Raising from Sequoia and John Doerr, he was on Sand Hill Road. Can you get me the meeting? I go, I'm supposed to— I said, listen, I'm in the final stages here. I had an offer from John Doerr. I had an offer from Sequoia. Mark Cuban had already committed, and so I'm 3 for 3. Let that sink in. I'm 3 for effing 3. More Davidow Ventures, I'll say the name of the firm, I don't think they even exist anymore, begs a friend of mine to give him the meeting. My friend's like a consultant to the firm. He's like, "Would you do it for me as a favor?" I said, "Fine, I'm going to be there." And they said, "Okay, come to the Monday morning meeting." I go to the Monday morning meeting. I get off the airplane in San Francisco. There's no internet on the planes 20 years ago. I'm driving in my rent-a-car. There's no Uber. I had to get a rent-a-car. This is like significant expense. I'm driving to San Diego. They have the office next to Sequoia. And I look at my StarTAC phone, just to date this, or it might have been my Nokia N95. I look at the phone, there's a message. There's 3 calls to 2 messages, whatever. This is before SMS. I listen to it. Guy says, listen, the meeting's canceled. I just wanted to let you know. Couldn't get my part. Now this is Monday effing morning. I got up at 5 AM to take a 6 AM flight to land at 7:30 to get a rent-a-car to then get to their office for this 10 o'clock effing meeting. I am on fire now. And this has got my Irish up. The kid from Brooklyn, like, I'm finding out that the meeting's been canceled on the way down the 101 freeway. I said, you know what? I'm going to the effing meeting. I'm going to the meeting. I walk right into Moore Davidow. I walk right to the receptionist. I'm Jason Calacanis. I have a meeting at 10 AM. And I look and I see the guy in the Monday morning partners meeting. It's 9:30. They've started already. It's 9:45. I'm here 20 minutes early. He turns white. He comes out. Did you get my message? I said, "Oh, I got your message. I'm here to present." He says, "I can't. I tell my partners, they told me the valuation, it's not a fit for our firm." I said, "You realize I got up at 5 in the effing morning for this meeting. I came up here and you're such an idiot that instead of just letting me pitch, telling me that I'm a genius, and then telling me your partners couldn't get around it, you had a conflicted investment, instead of just doing that." You're so dumb that you would cancel on me when you knew I was flying up from Los Angeles. You're going to be the worst venture capitalist of all time. I can't even remember the guy's name. And I said, I'm going to tell everybody the story. And I did. For 10 years, this is all I talked about with Mordechai Vidal. I would never recommend that firm. I don't think the guy's there anymore. If somebody says Mordechai Vidal to me, I just say, disgracia. And that's that. It is what it is. Like, uh, sometimes people will make a mistake. The guy came up to me 2 years later at like one of the Recode or D conferences. Hey, I just want to apologize one more time. I was wondering if I take you out for sushi. I said, I don't know if you're paying attention, but I'm rich now. I'm, I'm super rich. I could buy the sushi restaurant. I don't need to go to sushi with you. I don't need you to buy me sushi. When I needed help, you stabbed me in the back. I was just so, I mean, this is like 30-something-year-old J.Cow. I was a little crazy at the time, but I needed an enemy, so I took it out of him. That's what you're seeing on social media. You're seeing a 32-year-old J.Cow whose Irish is up. Everybody feels offended by something, and there seems to be people falling asleep in meetings. That's like a recurring theme. He had it happen. I had it happen. John Doerr, still greatest venture cap, one of the top, greatest venture capitalists of all time, great human beings. He had an excuse. I mean, he was probably on a Percocet or an Oxy or something. I don't know what they gave him, but, you know, he was banged up. He should not have taken the meeting. Um, but there's other stories here. Let's pull a couple more of these up. Travis, my boy, Kalanick. In 2001, I intercepted a partner at a VC— this is long before Uber, this is during Red Swoosh, I'm assuming— a VC who was trying to escape his office before our meeting was supposed to start. Same as me. I ended up pitching him in his parked Lexus from the passenger seat. At one point, he grabbed my laptop, placed it on his large belly, which was pressed against the steering wheel, and rapidly flipped through slides himself. 2001 fundraising hit different. 100% true. Um, I saw this actually happen. A friend of mine did this where he just took somebody's, um, laptop in front of me at like one of my events and just started rifling through the slides. You know, like, got it, got it, got it. And so when I made my firm launch and I have 11 people on the investment team, we were doing over 100 first meetings. I have every first meeting we do, 2 or 3 days after we do the meeting, it automatically sends from me, the head of the firm, you met with Bianca, Lucas, whoever it is on my team, was wondering if you could give us some feedback on how we did. And you can click here. And by the way, this is an automated message, but this is my real email. If you hit reply, you'll, you know, I'll engage with you or feel free to reply, whatever I wrote. So we get scores. If the scores are 7 and below, they go into Founder Feedback Tomatoes. And if they're 7 and above, they go into another room, which is like kudos. And we rank everybody in our firm. We stack ranked everybody. Now, first calls are only done by the 3 highest scored people. I told everybody else, oh, can I do first calls? Nope. These 3 people do the first calls because they are the great 3-point shooters on the team. If you would like to get back into that rotation, get in the gym, do it, and maybe I'll give you 10 meetings and let's see what your scores are and feedback. Then when they give that feedback, man, somebody will tell us in that feedback, this person didn't understand my business. That kept happening over and over again. People said that about me. People said I was not understanding the business. So what I did was I added to every time I meet with a founder, the final question I ask after they're done is, "May I repeat your vision back to you so that I can make sure I understand it perfectly?" Very simple sentence. What does it do? It lets the founder know that I've been paying attention all this time because I can repeat back to them what their business is. Number 2, if they forgot something that they wanted to add, they can add it at that moment in time. I've given them permission. Number 3, if I misunderstood something, whether it was my brain or their lips, somewhere between those two, I didn't get it, we can work that out. That changed everything. And then I made Everybody who works for me answer every single first meeting with that. And we said, our first meetings are 20 minutes long. You present for 10 minutes. We ask you 2 or 3 questions for 5 minutes. Then you can ask us about our firm for 5 minutes, and then we can decide mutually if we want to go to another meeting. Why do I frame the first meetings as 20 minutes on Zoom? Because it shows the founders that we respect them and we make sure we have all the proper information and we make sure they understand how our firm invests. And this has led to us getting extraordinary scores from founders, but you have to be deliberate with it. And then I had hired some people who had experience in venture capital. They brought all these bad habits. They wanted to do one meeting a day instead of doing 3, 4 meetings a day, and they just weren't hardworking. So I created an associate in training program because I noticed that the people who were in venture capital for more than a couple years absorbed all the bad habits, which is they think that their power dynamic is that because they write a check, because they write a check, they get to dictate these things. They don't. Okay, I'm Jason Calacanis. Thanks for coming to my TED Talk.

Bye-bye. Thanks for watching This Week in Startups. If you liked this episode, check out more.

If you're a startup founder, Founder University Cohort 13 kicks off this fall. It's a 12-week program that provides guidance on building your product, launching to real customers, and pitching to investors.

Top startups receive $25,000 or $125,000 in investment. Apply now at founder.university/twist.

Already have traction? The Launch Accelerator invests $125,000 and connects you with 500+ investors to help you raise your next round.

Apply at launchaccelerator.co. If you're an accredited investor looking to gain access to quality deal flow, apply for Jason's Angel Syndicate at thesyndicate.com. We find 2 to 3 deals a month. And check out This Week in AI, Jason's experts-only roundtable with top AI founders and operators every week. Find it thisweekinai.ai. Check out the TWIST ticker, our daily newsletter, at thisweekinstartups.com/ticker. Thanks again to our sponsors for making today's show possible. Follow the show on Instagram, follow the show on X.com. This Week in Startups publishes 3 days a week, Monday, Wednesday, and Friday at 5 PM Central Time. You can submit an audio or video file question by emailing it to thisweekindotcom.