Jason Mentors The Next Generation + Why AI Is Bad at Teaching (Aristotle Interview) | E2346

This Week In Startups is made possible by: Vanta: https://www.vanta.com/twist DigitalOcean: https://do.co/twist Sentry: http://sentry.io/twist Plaud: http://Plaud.ai/twist Today’s show: The smarter AI gets, the more it just does the work for you, and the less helpful it becomes as a tutor. Shan Reddy of Aristotle joins Jason to demo a voice-first tutor built to run a session the way a $200-an-hour human tutor would. We also dive into why frontier models are trained to hand you answers rather than build real understanding. PLUS, in Ask Jason, an incoming Fordham student (Jason’s alma mater!) wants to prepare for a career in finance. Plus, David from NYU calls back and wants to discuss what everyone would do with $500K in free money. Then, why does Jason think the Doomerism is coming from inside the AI labs? Guests: Shan Reddy on X: https://x.com/rshanreddy Aristotle: https://www.heyaristotle.com/ Relevant Links: Shoe Dog by Phil Knight → https://en.wikipedia.org/wiki/Shoe_Dog Something Like an Autobiography by Akira Kurosawa → https://en.wikipedia.org/wiki/Something_Like_an_Autobiography Einstein: His Life and Universe by Walter Isaacson, which Dovid is reading now → https://en.wikipedia.org/wiki/Einstein:_His_Life_and_Universe Gartner, Jason's example of firms doing research on behalf of corporations → https://www.gartner.com/ "Religious Scholars Met With Anthropic. What They Heard Stunned Them." by Elizabeth Dias, The New York Times, Sept. 29, 2026; the story Lon brings up → https://www.inquirer.com/news/nation-world/religious-leaders-met-with-anthropic-20260930.html David Sacks accuses Anthropic of a "regulatory capture strategy based on fear-mongering"; the line Lon references (Bloomberg Law) → https://news.bloomberglaw.com/artificial-intelligence/anthropic-is-running-a-strategy-based-on-fear-mongering-sacks Aristotle's $5M seed round (PR Newswire) → https://www.prnewswire.com/news-releases/aristotle-secures-5m-to-replace-the-chatbot-with-high-dosage-ai-tutoring-for-students-302880500.html "The 2 Sigma Problem" by Benjamin Bloom (1984), original paper → https://web.mit.edu/5.95/readings/bloom-two-sigma.pdf Journal record (Educational Researcher / SAGE) → https://journals.sagepub.com/doi/10.3102/0013189X013006004 NYC's AI moratorium through 8th grade (ABC News) → https://abcnews.com/Technology/new-york-city-public-schools-banning-ai-middle/story?id=136134872 Alpha School → https://www.alpha.school Brilliant's origin story at the LAUNCH Festival (NPR) → https://www.npr.org/blogs/alltechconsidered/2013/07/23/203677692/giving-brightest-kids-the-cram-school-experience-online Timestamps: 0:00 Live Ask Jason kicks off 0:14 Plaud: If your work depends on conversations — interviews, meetings, calls — you need a Plaud NotePin S. You can check it out at https://Plaud.ai/twist and use code TWIST for 10% off! 0:56 Fordham-bound caller asks how to break into finance in the AI era 5:15 Jason's gift: read one biography a month 9:22 Vanta: Get $1000 off your SOC 2 at https://www.vanta.com/twist 11:22 Returning caller David: what would you do with $500K? 19:09 DigitalOcean: Want to see what building on a true AI-native platform looks like? Head to https://do.co/twist to start building on DigitalOcean's AI-Native Cloud today - and cut your AI workload costs by up to 50%. 23:56 Bringing an AI protein folding tool into drug discovery 26:35 How to do customer discovery with big companies 29:00 The "dark arts" of market research and corporate intelligence 29:29 Sentry: Your team should be focused on shipping features - not chasing down bugs. New users can get $240 in free credits when they go to https://sentry.io/twist and use the code TWIST26 32:59 Is AI fear a dark PR campaign? 34:17 Anthropic meets with religious scholars 40:01 Are AI labs living through an Oppenheimer moment? 42:06 Why we need to talk about AI's upside in education 43:47 Sean Ready, co-founder and CEO of Aristotle, joins 44:58 Why AI models are getting worse at teaching 47:09 Aristotle demo: a voice AI tutor for AP Calculus 50:29 Predicting the next best teaching action 53:32 Bloom's two sigma problem and one-on-one tutoring 54:38 Aristotle's business model and pricing 57:09 Building a tutoring harness vs. relying on frontier labs 1:00:21 Why dyslexic students love Aristotle 1:01:56 Hand-grading thousands of hours of tutoring sessions 1:04:22 Socialization, classrooms and the future of public education 1:08:32 Where to try Aristotle and open roles Subscribe to our newsletters on Substack: TWiST: https://twistartups.substack.com/ TWiAI: https://www.thisweekinai.ai/ TWiVC: https://thisweekinvc.substack.com/ Subscribe to This Week in Startups on Apple: https://rb.gy/v19fcp Follow Lon: X: https://x.com/lons Follow Jason: X: https://twitter.com/Jason LinkedIn: https://www.linkedin.com/in/jasoncalacanis Check out all our partner offers: https://partners.launch.co/

• 0:00Speaker 00

  All right, everybody, welcome back. This Week in Startups Twist, October 2nd. Here we go. We're in the fall on, and we got a full docket and a lot of questions. People have been wanting me to do and ask Jason, don't want to lose a minute of it.

• 0:12Speaker 01

  Right.

• 0:12Speaker 00

  So I always like to take a moment to applaud Plod. Sometimes I got my new iPhone 18 Max. And then I'll just slap this on the back.

• 0:21Speaker 01

  There you go.

• 0:23Speaker 00

  Press the button. Don't have to open my phone. Just press the button. Lay it on the counter. Now I've got that incredible 360 degree microphone, recording my meeting and transcribing it. What a great product. Absolutely addicting.

• 0:38Speaker 01

  If you want to go check it out for yourself, you don't have to just take our word for it. Go to plaud.ai/twist and please use the code TWIST. It doesn't only help us out here at the program, but you are going to get 10% off your Plod note pin.

• 0:52Speaker 00

  There you go. We always applaud, applaud here.

• 0:54Speaker 01

  We do.

• 0:55Speaker 00

  All right, let's get started. Busy day.

• 0:56Speaker 02

  This Week in Startups is brought to you by Sentry. Your team should be focused on shipping features, not chasing down bugs. New users can get $240 in free credits when they go to sentry.io/twist and use the code TWIST26. Digital Ocean. Want to see what building on a true AI native platform looks like? Head to do.co/twist to start building on Digital Ocean's AI native cloud today and cut your AI workload costs by up to 50%. Vanta. Compliance and security shouldn't be a deal breaker for startups to win new business. Vanta makes it easy for companies to get a SOC 2 report fast. Get $1,000 off for a limited time at vanta.com/twist.

• 1:40Speaker 01

  All right, yeah, we're doing, we're doing some Ask Jasons. We got some live callers already in the waiting room. So I say, let's, let's admit our first caller. Welcome to the show. What's your name? Let us know your name.

• 1:52Speaker 03

  My name's Augusta and I'm from India.

• 1:55Speaker 00

  Augusta, welcome. Calling in from India. What time is it in India?

• 1:59Speaker 03

  Oh, it's 10:00 PM right now.

• 2:01Speaker 00

  Oh, okay. All right. And so you got to get to bed. You got school in the morning, kid?

• 2:05Speaker 03

  No, so I'm joining Fordham this spring.

• 2:08Speaker 00

  Fordham University in the spring. Oh, my alma mater. Very good. Lincoln Center or Rose Hill?

• 2:13Speaker 03

  Rose Hill, Rose Hill.

• 2:14Speaker 00

  Rose Hill. Wow.

• 2:16Speaker 03

  So I heard you went to Fordham as well.

• 2:17Speaker 00

  I did. At the last minute in August of 1988, I got accepted as an alternate. I had, was going to go to Brooklyn College. I had no money. My dad's bar had been seized by the feds because he got behind on his taxes. I had no money, but the first week of August, Edward M. J. Boland, who I had been pitching on me going to Fordham, called me up and said, hey, Jake, thanks for all the recommendation letters and coming up and visiting me. I was pretty dogged in trying to get in. He said, I'm leaving to go to Yale to be admissions at Yale. Just wanted to let you know, before I left, I wanted to have the most promising non-traditional student get accepted to Fordham University this year. And I said, who's that? He said, you, idiot. And he said, where would you like to go? Rose Hill or Lincoln Center?

• 3:02Speaker 00

  I said, I don't know. He goes, well, there's no housing left. I said, well, I can't afford the housing anyway, so I'm going to commute. He goes, go to Lincoln Center. So I went to Lincoln Center at night and it took me four and a half years and I paid for it myself. It was $11,000 a year. I was always in arrears, six to 12 months behind on my tuition, but the Bursaw put me on a payment program. And somehow I squeaked through with only maybe 10K in Pell Grants,

• 3:26Speaker 00

  which are like US loans that the government gives you, and then I paid those off the year after I graduated because I got a good IT job. But yeah, you'll have a great time there. It's incredibly expensive. You must be a genius or your parents must be rich or some combination of the two. What is it, Augusta?

• 3:40Speaker 00

  Parents are rich. I can tell from the roof of that marble apartment.

• 3:43Speaker 01

  He's a rich genius, Jason. He's a rich genius.

• 3:45Speaker 00

  Okay. That's incredible that you got into Fordham. Wow.

• 3:48Speaker 03

  I think they're trying to lower acceptance rates as well, so it's pretty competitive this year from what I heard.

• 3:54Speaker 00

  And Fordham's no joke. I think it's $80,000 a year all in, so it's hard to get into. All right. So what's your question for me? Is it about going to New York and you need tips on New York? Any tips on dating girls? You need tips on what you should major in and paying for your tuition?

• 4:08Speaker 01

  I need tips on living in New York and dating girls, Jason.

• 4:11Speaker 00

  Okay, yeah, I'll get you that. You'll be the next Ask Jason. Go ahead.

• 4:14Speaker 03

  Yeah, currently I'm planning on majoring in finance.

• 4:17Speaker 00

  Okay.

• 4:18Speaker 03

  So since, so since the world is changing really quickly due to AI and whatnot. So I was wondering what your advice would be for me, since I'm getting into the finance space and soon I'll be looking for a job, but creating my next entrepreneurial.

• 4:33Speaker 00

  So you want to be AI first. So anything they try to teach you, you're going to go to AI first and say, here's what my teacher wants to teach me. How do I use AI to be superhuman at this? And what questions don't I know about gap, about this, about reconciliation? What don't I know about building models? And you want to have one track where your teachers are teaching you something and then you want to create your own agents that are teaching you one-on-one about finance and what the more advanced topics are and how you can build models yourself. And then look at history. So when you come out, you're well-rounded. I'm waiting for somebody to come out and say, yeah, you know, I really want to be in venture capital.

• 5:17Speaker 00

  I actually did an analysis of Kleiner Perkins in the 80s and 90s and all their investments. And I built this model and I researched it all and I researched how they made their investment decisions and how they exited too early from Apple and Cisco. And if they had held their shares, when they became public companies, what those shares would have been worth and what the returns for that fund would have been. And I looked at their DPI and their TVPI and I looked at their IRR over 30 years. Instead of just the 10, they were investors in Apple. I'm waiting for some ambitious kid who does that while they're at school. And that's, will change everything for you. And then the other unlock I want to give you as a gift is to read a biography a month. One biography a month, 10 a year. You can take a month or two off. You can listen to them too.

• 6:03Speaker 00

  So for example, and you don't have to read just business biographies, although that's a great place to start. Read Michael Ovitz's CAA's biography, then read the oral history of CAA. Then read Bob Iger's biography, right? He did Disney. Now you're going to get a look at like that world. Then you can read The Kid Stays in the Picture from the 70s, Bob Evans, who was a studio head. Then you could go even further back and read Akira Kurosawa's, something like an autobiography. And you, Augusta, would, just by reading four books, five books, would understand that entire finance picture and how that business and that industry grew from the 50s, 60s, 70s, 80s into the 90s.

• 6:47Speaker 00

  Then you could do the same thing for Nike and other iconic brands. Shoe Dog by Phil Knight. You can look for other people who wrote about Nike and, you know, Japan during that time. You could read a biography of Sony and you could understand how Nike, you know, making sneakers in Japan was a big deal. This is the greatest gift ever because what young founders try to do is they try to get in touch with somebody like me or more important than me, and they try to say, oh, can I have coffee to learn? You don't need to do that. You just read the biographies. I read every biography you could imagine. And don't be scared to read biographies of people who have nothing to do with your industry. So go read Born Standing Up by Steve Martin.

• 7:31Speaker 00

  Go read Robin, about Robin Williams. Go read any biography you can get your hands on. And you will be more worldly than any of your classmates. So when you come out and you start interviewing for jobs, this kid from India who somehow made it to Fordham, who's got an accent and people are like, oh, these Indian people are squares. You got to put them in a box over here, right? They're going to treat you a certain way because you're an Indian and because you're an immigrant. And then you're going to just blow their minds because you're AI first. You're building your own models and you have your own interests in a broad array of entrepreneurs and talented people. All of a sudden, they're going to have a hard time keeping you in that box where they'll want to keep you, right?

• 8:14Speaker 00

  When I was the kid from Brooklyn, they tried to keep me in a box and then I just broke the box. I lit the box on fire. I torched it and threw it in the ocean, right? With the remnants of it. And I was like, fuck this box. I'm going to just be unstoppable and I'm going to build my own stuff. And yeah, you'll do just fine.

• 8:30Speaker 03

  Yeah, just one more follow-up question. I'll definitely read those biographies like you recommended.

• 8:34Speaker 00

  Read all the biographies, 10 a year minimum. And you can listen to them too. No.

• 8:39Speaker 03

  Yes, one more question.

• 8:40Speaker 00

  No wasting time. Go ahead. Yes.

• 8:42Speaker 03

  So my mom's the current Harvard Bangalore Club president, like for the Bangalore region.

• 8:47Speaker 00

  Sure.

• 8:48Speaker 03

  So I think she'd be upset if I didn't ask you if you'd be interested in talking to the alumni from Bangalore online or however.

• 8:54Speaker 00

  Oh, you want me to do a free talk for your mom? Yes. That sounds incredible.

• 8:58Speaker 03

  That'd be amazing.

• 8:59Speaker 00

  Yes. I have no interest in doing that, but since you called into the show, sure, I'll do it. Just talk to my team. I'll give you a half hour and yeah, of course I'll do it for you. And I will be watching your studies from afar, Augusta. So I would like a report when you get through your first year, call back into the show in one year. Want to know what biographies you read? I want a yearly check-in from you. Understood?

• 9:21Speaker 03

  Yeah. So how can I each out to you?

• 9:22Speaker 00

  Yeah, it's very simple. They're going to give you the, you're already in touch with them. But they'll give you an email right now and then you will, they'll schedule it with you.

• 9:30Speaker 01

  We'll figure it out, Augusta. Thanks for calling.

• 9:32Speaker 03

  Thank you so much.

• 9:33Speaker 00

  Augusta, okay, there you go. Look at that kid going to Fordham, my alma mater. Yeah, if you just read the biographies, you'll be fine.

• 9:39Speaker 01

  Did you ever read that Peter Biskin book, Easy Riders Raging Bulls? You got to read that.

• 9:43Speaker 00

  Of course. Yes, of course. I'm just buying books like crazy now. I just, you're seeing the stacks of them. Thing I just, people like, are like, I like these seven books. And I just screenshot it. I send it to my Grok bot. I'm like, put these in my.

• 9:58Speaker 00

  Put the paperbacks in, put the fries in the bag, put the paperbacks in the shopping cart on Amazon. I just buy them and I just leave a stack of books everywhere. Here's a lesson for all my founders out there. Trust closes deals. You can have an amazing, beautiful product. It solves every problem your customers have, but if those same customers who love your product, don't believe it's totally secure, they're out. And that's why you need a partner like Vanta. They're the number one security and trust platform in the world, and that's going to get you compliant fast for all the major frameworks from SOC two to HIPAA and beyond. Plus, Vanta is going to make sure you stay compliant by continuously monitoring your controls. And now you can interact with Vanta's agents wherever you're working.

• 10:43Speaker 00

  Even if your team is knee deep in Claude or Cursor, it doesn't matter. Having Vanta as your partner means your deals keep moving and your engineers can focus on building that next great feature, not tedious documentation. That's why Vanta is already trusted by more than 16,000 companies, including Ramp, Ryder, Harvey. You've heard of them. So prove to your customers that you're ready for business. Learn more at vanta.com/twist and get $1,000 off.

• 11:12Speaker 01

  Did you see that comedian Anthony Jeselnik? He's, uh, he's doing a new like book club. He's trying to get people to go back and read more books. So it's every like few weeks on his Instagram. He shares another book and he like hosts live streams where they talk about the books.

• 11:24Speaker 00

  I think book clubs are incredible. If I had more time, I would run one. It's just, uh, I, my, my time is so, um, yeah.

• 11:32Speaker 01

  I don't read as much as I would like to either.

• 11:35Speaker 00

  Yeah. And this is why, uh, one of the things I'm finding myself do is buy the audiobook and then buy the Kindle book or buy the print book and buy the audiobook. So I could switch between them just so whenever I have a window, if I, if I'm hiking the ranch, I can listen to it. And then if I'm sitting in the back and I got a quick half hour or 45 minutes, I can, you know, read a chapter or two. All right. Let's keep going.

• 11:56Speaker 01

  All right. We'll do another caller. Uh, you, an old friend calling back. I believe we actually, uh, you got him into the All In Summit after his last call. Uh, David, our old friend David is back with a new question. Welcome back to the show.

• 12:08Speaker 04

  Hey guys, how's it going?

• 12:09Speaker 00

  He accosted me at the party at the mansion and told me he made it in. So David, you had called in. We got you somehow got you into the scholarship program. Uh, how'd you do at the All In Summit? Did you have a good time?

• 12:19Speaker 04

  Oh my gosh, it was incredible. Such an experience. Thank you so much for having me out.

• 12:23Speaker 01

  Did you get to go to Universal? Were you, were you, were you in Universal Studios?

• 12:27Speaker 04

  Yeah, yeah. I wrote some of the lines. I actually, the question that I'm asking today actually has to do with someone that I met there.

• 12:33Speaker 00

  Okay, great.

• 12:34Speaker 04

  So that's a pretty interesting story.

• 12:35Speaker 00

  Clavicular? You met Clavicular? There's that.

• 12:37Speaker 04

  I didn't meet Clavicular, yeah. I was eating lunch and I turned around and I saw him and I was like, oh my God, is that, is that Clav? Like, yeah.

• 12:44Speaker 00

  That's, that's one path. You don't necessarily have to take that path. There are other paths.

• 12:49Speaker 04

  Uh, Universal, actually I met, I, I walked up to this group of, of guys. We're talking, introduced myself. Yeah. I wanted to, you know, know what they, know what they do and, you know, kind of.

• 12:59Speaker 00

  You were networking.

• 12:59Speaker 04

  Yeah, networking.

• 13:00Speaker 00

  There you go.

• 13:00Speaker 04

  And so I went over to them and turns out this guy runs a startup in New York called Digital Zone. I'm not sure if you've heard of it.

• 13:07Speaker 00

  I have not.

• 13:08Speaker 04

  So I asked him if I can come to his office, you know, kind of learn, learn what they do, see how, see how he operates. And he said, sure. So I went to Freedom Tower on Tuesday and he introduced me to two of his advisors. It was super interesting and we talked for like an hour. And at the end of it, he asked me a question. He's like, if you had, if I gave you 500K tomorrow and you had absolutely no obligations, like what, what would you do with it? What would, what would your first course of action? So I didn't have time this week to think about it much. I was busy with school and then I was just about to like sit down and do some research and I saw that you guys were going live. And you had the, the segment. So I figured I would come ask you first before I started.

• 13:45Speaker 00

  I mean, what to do if your rich uncle or an investor gave you 500K? There's, you know, a short list. One is you could look at it as a way to build your skill stack. And you could say, I am going to spend a certain amount of money on gear, software, et cetera, to build my skills if you're a self-directed person, right? Because sometimes time is what you don't have. Are you in school or you have a job?

• 14:13Speaker 04

  Yeah, I go to NYU.

• 14:14Speaker 00

  Right. So you could say, well, maybe I can't afford to get a Mac Studio and, you know, this equipment, because I want to start a podcast or I want to do AI research and I could use a really powerful computer. And, uh, I want to buy some software and then, you know, I just want to increase my skills or I want to take some courses, et cetera. So buying time and then being able to build skills would be an obvious one. Uh, the second one is obviously start a company. The third is put the money away and invest it wisely. So, and become like, say, an angel investor or a public markets investor. That's what you can do with capital. So it would depend on if you want to invest in yourself, you want to invest in markets,

• 14:55Speaker 00

  you want to invest in other people, uh, or you want to invest in yourself via a company. Cause you, so you, that's what you have to think about. Are you investing in yourself with that 500K? Are you investing in other people? And some people like to be professional investors. I enjoy it, but I also like to invest in myself. Um, and so professional investors, you know, you're going to be looking at public markets, private markets, et cetera, angel investing. So if I had 500 and it was, I could do it all over again. Somebody gave me 500K in 1996, I would have started my magazine and instead of having no money, except my credit cards, I would have had money and would have taken out the first year of struggle. Or I would have become an angel investor, hands down, I would have become an angel investor in the technology industry.

• 15:37Speaker 00

  I just would have started an email newsletter, called it a day and started placing bets. So it's a great thought experiment. What did you come to, uh, in terms of your thinking? If somebody were to just bless you with a half million dollars?

• 15:48Speaker 04

  First of all, I'd move to the city. I commute from Long Island every day. I take the LIR.

• 15:52Speaker 00

  Good investment proximity. Yes. Get the cheapest apartment you can. Yes.

• 15:55Speaker 04

  Save me like two hours a day right off the bat. So I would do that. I mean, I, I, I don't have like one specific idea yet that I would be willing to found a company based on. So like, I wouldn't just take it and start a company like based off the first idea that I find. I think that's stupid. I'd rather find something that I'm passionate about. But until then, you know, maybe you take 250 of it and, and place some angel bets. Like there's a lot of people starting companies in NYU and Columbia. I mean, you know, I met 25 kids at the All In Summit and one of them was building some patent software. It was like, you know, people are, there's a lot of kids building right now and a lot of kids who need capital. So I think to, yeah, place some bets there, what I think is really interesting.

• 16:31Speaker 00

  I mean, you could find somebody, uh, with an apartment already and, you know, get an apartment share for 1,500 to 2,500 bucks a month, anywhere in Manhattan, Harlem, all the way down. Um, and you know, instead of starting companies, I would start projects. That would be actually a better way to do it is to just make little projects. And if you have to hire some freelancers to help you with some pieces of it, but learning every aspect of building a company is so powerful, so powerful. And that's what I'm always trying to do with my internal team. Like Lon's a great writer. Lon's a great producer. Lon's good at hiring people, uh, very good at hiring people. And I told them with the newsletters, hey, just come up with the value proposition, do every single step along the way.

• 17:15Speaker 00

  Let me know how it goes, uh, you know, vis a vis how many people are paying for the subscriptions or not. That's all I care about is the outcome. Um, and I gave them, I see to do with some ideas, but you want to be able to learn the entire thing. And typically in business, when you go work for an entrepreneur, they put you into a, a narrower box, a smaller bucket. They give you a, a charge that is a very narrow charge. Uh, and then you don't learn holistically how to do something. Then you're kind of meek and not super valuable. What you want to be is unstoppable, that you can do every single thing from your taxes to, you know, negotiating a non-disclosure agreement, from starting an LLC, doing your payroll.

• 17:59Speaker 00

  You gotta learn every single piece of that if you want to be an entrepreneur. In other words, every single chore, you need to know how to do. So go down the entrepreneur path for sure. And by the way, all, none of this requires money anyway. So it's nice to have this great $500,000 thought process. You, you could be driving Uber and in between calls, taking out your laptop and coding. So in another world, I probably would have been an Uber driver, uh, just pulling over to the side of the road and writing my magazine and doing my zine on the side instead of fixing computers for 30 bucks an hour, 20 bucks an hour, whatever I was getting paid at the time. Yeah. I think it was 20, 30 bucks an hour. You know, I was fixing computers at Amity State International. I was making 12 bucks an hour. I would have been an Uber driver for 20, pull over to the side of the road, Lon, and I would have been writing a story,

• 18:44Speaker 00

  taking pictures, going to the printer and then doing a couple more, uh, runs. And I, I met a guy online who was also at the All In Summit who, uh, got dragged literally like a group of people were harassing him as an Uber driver. And I, it came across my Twitter feed and I had written to him, I'm sorry you had to deal with that. Um, and, um, he said, oh, thanks. I'm a big fan. I'm a Uber driver and an entrepreneur. And I said, mad respect, mad respect. That would have been me. That's the true grinders, like, good stuff.

• 19:17Speaker 01

  Yeah, I think it's amazing to think about. I mean, 500 grand when I first started working in tech would have been like, I mean, you could try to start a company with that, but you're gonna need to get more funding pretty soon. And today, I feel like, man, that would

• 19:29Speaker 00

  Too much money.

• 19:29Speaker 01

  It could last you forever. You know, who needs 500 grand to start a company in 2026?

• 19:34Speaker 00

  Too much money. What you need is 10K a month, 20K a month, so that you can pay for everything, be comfortable and have like one employee. But once you start having 50K a month for 10 months, or 25K a month for 10 months, then it's a distraction. You're like, how do I spend this money? You don't need to spend any money. I mean, we're gonna, you can get all the credits for free. The truth is, most of the big, popular AI models can handle your workloads just fine. And your performance and your cost, that comes down to your infrastructure more than your model selection. And the best way to keep your system running fast while staying affordable is with Digital Ocean's inference engine. Digital Ocean built their entire cloud platform from the ground up specifically with AI in mind.

• 20:16Speaker 00

  That means you get a serious performance upgrade without stitching together services from a dozen different vendors. Or getting hit with the surprise bill at the end of the month, which keeps happening to me. Digital Ocean's system is already powering some of the biggest and most popular AI applications. Character AI runs over a billion queries every single day on Digital Ocean. And they cut their cost per token in half while doing so. So here's your call to action. If you want to understand what building on a true AI native platform looks like, go to do.co/twist. Start building on the Digital Ocean AI native cloud today and cut your AI workload costs by up to 50%. That's do.co/twist.

• 21:01Speaker 00

  All right, listen, David, continued success. When do you graduate, NYU? When are you graduating?

• 21:05Speaker 04

  Uh, 28.

• 21:06Speaker 00

  20. Oh, you got plenty of time. So yeah, get focused on school. I like your networking. What you need to start doing is, um, I want to see you, if you, if you heard my previous caller on today's Ask Jason, I told him to read all the biographies. I'm gonna give you that same assignment. So I want you to read five biographies and then you'll, when you do your next Ask Jason, I want you to tell me what two things you learned from each. Okay, David, you understand your assignment?

• 21:27Speaker 04

  This is, yeah.

• 21:28Speaker 00

  So as soon as you read the five books or listen to them, and you have two insights per book, 10 insights, you're gonna come back on the program. If you do it in a month, you do it in a month. If you do it in a year, you do it in a year. But that's your assignment. Five biographies. Doesn't have to be just business, could be anything. And then I want you to call in and tell us the two things you learned from each. Understood? And I want to see the books. I want you to hold them up with the yellow highlighter in them. I want you to take notes when you read the books, okay? And if you listen to them, I still want to see the notes.

• 21:53Speaker 04

  Yeah, it's funny you say that. I'm in the middle of Einstein by Walter Isaacson right now.

• 21:56Speaker 00

  Great, perfect, perfect.

• 21:57Speaker 04

  Yeah, that'll be the first one.

• 21:58Speaker 00

  That's a great first one. And then you're gonna want to move to, you know, impresarios, people who, you know, have like real talent in, you know, uh, some specific vein, whether it's music or comedy or authors. Stephen King has a great one on writing. Stephen King's on writing, really interesting. And those will just unlock in your brain and inspire you. So you're gonna unlock lessons while being inspired. That's the magic of these biographies. Uh, so yeah, I'm gonna, I'm gonna tell you, read Stephen King's on writing. It's pretty crazy.

• 22:31Speaker 01

  That's a great book.

• 22:31Speaker 00

  Literally stories of him writing Cujo at his desk, on his typewriter, while doing mountains of cocaine and having tissues up his nose because he was bleeding from cocaine. And if you've ever seen the movie Cujo, that is a cocaine-filled horror film, I think. It's like, not that there's, like, it's about a dog who's like a Doberman who's like killing people. Like, it was obviously written on cocaine.

• 22:54Speaker 01

  There's a few of those. I think it's The Shining is one of the ones where he says he doesn't remember writing it. Like, now it exists as like this lost artifact that he, like, he did it, but he doesn't remember the process of thinking the story up and putting it out there. It just happened.

• 23:09Speaker 00

  So, but don't do cocaine, Kevin. No drugs.

• 23:11Speaker 01

  Please don't.

• 23:11Speaker 00

  David, your drug is biographies, okay? And then calling into the show. Those are your two vices. All right, we'll see you on your next call in. Well done, David.

• 23:20Speaker 04

  Thanks, David.

• 23:21Speaker 00

  I like this, uh, young people calling into the show. I want those two folks to call back into the show. I want to get updates from them on their careers.

• 23:27Speaker 01

  Yeah, we want to hear how things are going for them after the, we could, we could drop them off now. There you go. Uh, yeah, I, I like, I like checking in with them and, and hearing back. And, and especially now that we're giving them homework. We've got to see if they're gonna be diligent about reading their biographies.

• 23:39Speaker 00

  Put a calendar reminder with them and us. Uh, like, hey, call in again. And just like, put it for, I don't know, three or six months from now. We'll get the, you know, somehow record all these. Uh, so we, we have, we know, or actually maybe I can just say, hey, bot.

• 23:54Speaker 01

  I was gonna say, Grokbot could easily do this.

• 23:57Speaker 00

  Or any bot. Any bot, remind me. There's like so many bots listening to the transcripts of podcasts. Hey, any random bot, email me. Like, there's some bot listening right now who'll just email me.

• 24:07Speaker 01

  Plod's listening right now.

• 24:09Speaker 00

  Every, hey, Plod, let us know. All right, let's do one more.

• 24:11Speaker 01

  Ray, come join us.

• 24:12Speaker 00

  Now we're cooking. Look at this. We went from it being impossible to get callers and now you guys are crushing. Love it.

• 24:18Speaker 01

  Also, while we're waiting for Ray, Shaky, that Jimmy McDonough biography of Neil Young, did you ever read that one?

• 24:24Speaker 00

  I have not.

• 24:25Speaker 01

  What a read. Incredible. Neil Young wrote his own autobiography, but it's a little messy. Like, he, he kind of just jumps around. Shaky is like a definitive, you get the span of the whole career. He talked to like every Graham Nash, like all Stephen Stills, all the guys who are still alive. It's really, really terrific. Anyway.

• 24:40Speaker 00

  All right, Ray, welcome to the program. What's your question?

• 24:41Speaker 05

  Hi, Jason. Uh, long time watcher, first time caller. I, uh, I work in AI space. I work on deterministic world modeling.

• 24:48Speaker 00

  Cool.

• 24:48Speaker 05

  And I've been working on something that does, uh, protein folding for drug discovery. But I'm not in drug discovery. I, it's not my field, but I know it would help people a lot. What do you think are the pitfalls or the ways that bringing tech to medicine? Because I know it's different than other facets and you've invested in literally everything. So I was hoping to kind of have the cheating roadmap from you.

• 25:08Speaker 00

  Wow, well, it is not my wheelhouse. I'm not in it day to day. But clearly, AI is being used, um, slowly, methodically. And I think there's probably people like yourself who are experimenting on the fringes independently. And there was this guy who did his own cancer drug for his dog that I'm sure you read about, right? So there's like the wild west over here. And then there's like big, slow pharma companies and other folks in between are startups, right? And the startups tend to start at the wild west. Somebody like yourself is iterating, getting inspired, you know, poking around. And then they work backwards, uh, to a startup.

• 25:51Speaker 00

  And then startups generally get bought by those big companies or big companies copy what they're doing. Here it is. A man used AI to help make a cancer vaccine for his dog and oncologist urges caution. So there's a lot of stuff happening on the fringe. And then it just depends, uh, if you want to get into application of this, which it sounds like you want to apply this to your dog. If you want to start applying this.

• 26:14Speaker 05

  Yeah, my dog hears what you're saying.

• 26:15Speaker 00

  Yeah, I would start meeting with the startups in the space. It's very simple. You just start emailing the founders. Hey, I think it's really interesting what you're doing. I'm doing something. Would love to meet you for a cup of coffee for but 20 minutes. I'll meet you anywhere 7:00 AM to, you know, midnight, seven days a week.

• 26:31Speaker 05

  I have a startup and an investor.

• 26:32Speaker 00

  Perfect.

• 26:33Speaker 05

  I was just, so I'm just kind of, I'm just kind of wondering like from this stage forward, like, you know, I'm not, I'm, I'm early, but I'm not that early.

• 26:39Speaker 00

  So, uh, you know, it's going to be slow if you're going to go through the regulatory process, obviously. If you're going to be going direct to consumer with things that don't require FDA approval or that approval process, like, you know, sleep.

• 26:57Speaker 05

  With pharmaceutical companies who would use this to do drug discovery.

• 27:00Speaker 00

  Oh, great. Yeah, tools. Yeah, I mean.

• 27:02Speaker 05

  It's like a tech.

• 27:03Speaker 00

  If you're, if you're making tools, that's super easy. You, you gotta just find the people at these companies who would use the tools and then do what is called, uh, customer discovery, uh, where you say, hey, I'm building some tools. I'm an entrepreneur. I'd love to show them to you and get your feedback. Now, of course, you could be educating people. They could go build and steal the idea. But most people at big companies don't have the time. They punch out at like 3:30. They work for another two hours. You know, but they're really like, you know, doing Amazon or fucking off, reading the news. You know, they put in three hours a day of work at these big companies and then they punch out at 5:29 and they're at the elevator banks and they, they're not the person who's gonna steal your idea from you.

• 27:46Speaker 00

  The person who'll steal your idea from you is some, you know, two kids in India, as we just saw on a previous call, who are gonna come to the United States and join Y Combinator or Techstars or Found University. Those are the people who are gonna steal the idea. So do customer discovery. Go find people who would be the

• 28:01Speaker 05

  Customer discovery.

• 28:01Speaker 00

  Yeah, big. And it's, it's, it's a whole science of how to get those people on the phone, how to lure them in, how to say I could use some feedback. Some people, um, will give people a, a, like a gift card. I know this sounds corny, but they'll be like, hey, I'd love to just get your feedback on my product and I'll give you a $50 gift card or something, a $100 gift card. And it's amazing. People will get on the phone for that. I, I know that sounds crazy and insane, but people also will do it like at cafes. Now, this doesn't apply to you, but for a consumer product, people will go to cafes and say, hit you or somebody online at Starbucks, hey, can I, I'm an entrepreneur. I made an app for dating. I'm wondering if I could show it to you and buy you a cup of coffee and, and buy you your croissant in exchange for showing it to you.

• 28:45Speaker 00

  And people have done this technique so much that Starbucks will kick them out after they annoy enough people. But that's the bold entrepreneur who you can't like stop. They're so dogged that you, you just, you, you

• 29:00Speaker 00

  Will be impervious to the number of no's you get at a certain point. So that's what my wish for you is.

• 29:06Speaker 05

  It's worth bringing in somebody that specializes with that kind of talent set as opposed to

• 29:11Speaker 00

  I mean, sure, sure. You can, you can do that. But one of the things I like about entrepreneurs doing it is, or you could have that person like get the person on the phone and then you do the discovery call. So you could have one person who does the motion of trying to get a person on, haranguing them, finding them, et cetera. And you could probably do this with a bot now. So if you know the title of a person and you've got a really well crafted LinkedIn and you put it into Grokbot now, it's kind of unhinged and will probably find the people for you, figure out ways for you to email them and get in touch with them and set it up. And like in terms of dark arts, there are people who will set up a research fund, firm, pay people to do calls and then show them like three or four different products. One of them is the actual real product.

• 29:56Speaker 00

  Two of them are competitors' products, et cetera, and say, I work for a national survey, biological, you know, national science surveys and we do research on science surveys and we pay people a hundred dollars an hour for a 30 minute call. We show you four products and we just ask you for your feedback on them in a post survey. Your team can push code faster than ever with the help of AI agents, but when something in production breaks down, you have no idea what's really going on. So you just give the agent your error message and you hope for the best. But there is a solution and the solution is Sentry, which is already being used by four billion developers. Sentry is constantly watching your app. So when something breaks, it already has all the context. And Sentry's AI debugger, Seer, can quickly determine the root cause.

• 30:41Speaker 00

  Now your agent isn't guessing, it's targeted on the solution. If you're running Cursor, Claude Code, Factory Droid, Copilot, or anything else, there's nothing to install. Anything that can call an MCP server or a CLI can talk to Sentry. Sentry finds the bug and your agent fixes it. Try it free at S E N T R Y dot I O and tell them Twist sent you. They have a free dev plan and listeners to our show can use the code TWIST26 for $240 in Sentry credits. And like they set up these fronts to go get information from corporations. It's kind of in the corporate espionage space. If you were getting proprietary information or it's just market research, you decide.

• 31:25Speaker 00

  You know, if you're doing market research and people who are, let's make it with like accountants. Let's say you set up, you know, accounting research systems and you say, hey, I, I work for accounting research systems. I'm a researcher and we research new products and we try to get feedback on them for our newsletter and for our research reports. We'll send you the research report when we're done. It's a hundred bucks. You're a CPA. We show it to you for 30 minutes. We get your feedback. Like people will come on, they can give you feedback. Pretty crazy, right? You'd be surprised at the feedback people will give you. And then you might have like Ernst & Young or KPMG giving you feedback on their.

• 32:03Speaker 05

  That's smart. You're saying that like the indirect approach can be more harvestable than the direct approach.

• 32:09Speaker 00

  Now you don't want to get into dark arts where people are breaking their NDAs, et cetera. And some people do that, actually do that specifically. And that's when you get into corporate.

• 32:22Speaker 00

  What are they called? Well, I mean, there's espionage, which you don't want to do, obviously. You're not trying to steal documents. But there's corporate intelligence. And if you start looking at corporate intelligence and you can just go into any LLM and say, what are corporate intelligence techniques I can use that are not illegal, but that are darn clever? And I just gave you a couple of them. Now, I've never done these, but I've heard from people, people doing them. And I think they're quite clever in terms of just doing research and understanding a market space, right? And that's what like Gartner Group and some of those people actually do for corporations. They go out and they do research on behalf of corporations and the corporation is shielded from that research.

• 33:02Speaker 05

  It sounds like no matter what it is, the overall goal is whatever you have to do to get you in front of the customer is what you need to do. And it might not be a straight line, but it's, it's got to get you there. And that's all that counts.

• 33:16Speaker 00

  If you want to get great feedback and then every time you do that customer intelligence, you're got a potential customer to sell into. So it serves two purposes. They might give you great ideas and feedback on your product, what to take out, what to put in, you know, how to charge for it, how to design it. And then, once the product's done, they might become your first customer. Great question, Ray. Well done.

• 33:40Speaker 01

  Thanks, Ray.

• 33:41Speaker 00

  Cook them with oil here. Lots of good ass Jasons.

• 33:43Speaker 01

  Yep. We got, we got a pre-recorded one. If you want to do one more.

• 33:46Speaker 00

  Let's do it. I got good energy right now.

• 33:48Speaker 01

  This one's from superfan Justin Brady. So we can pull that up. A big, big fan of the show, very chatty, frequent commenter. So let's definitely jump into Justin's question.

• 34:00Speaker 06

  Hey, Jason, you have journalistic roots like I do. Now I'm in PR, GTM and comms for founders. And is it just me or is the AI doomerism stuff kind of seem like a dark PR campaign? If it is, I mean, I may be wrong, but if it is, where's it coming from? Is it coming from China? Cause they want to slow the US down? Is it coming from the big AI companies because the regulatory capture and they're trying to defend their turf? Or is it coming from healthcare companies, education companies and legal companies? Because they like their protected positions and don't want smaller startups and smaller companies having an advantage? Or none of them? I would love to hear your answer on this.

• 34:40Speaker 00

  It's coming from the frontier labs who have learned that every time they put a release out, that their products are incredibly powerful and can hack a system, et cetera. Their usage goes up. The embracing of the product goes up, token usage goes up. And then they also probably sincerely feel this way and have been one shotted by the technology. So two things can be true.

• 35:02Speaker 01

  Did you guys see, I, I, I was hoping we would get here. Did you guys see this religious scholars met with Anthropic story that was in the New York Times the other day?

• 35:13Speaker 00

  I, I did see it go by on my social. Why don't you tee it up here?

• 35:18Speaker 01

  In a series of private meetings, Anthropic consulted with various, quote, about religious scholars. And they wanted to discuss how to instill morality into their AI models and make the case that Claude has a soul. And if you read the entire article, I, I have, I've sort of shifted my thinking. I, at first I sort of agreed with the, the, I can't believe I'm saying this, the sort of the David Sacks line that this was just regulatory capture as a strategy. And Anthropic, you know, they were, they thought they had control enough of the market and they wanted to shut down new competitors and innovation. I don't know. You read something like this and now I'm starting to come around to like, no, no.

• 35:55Speaker 00

  My position, these people are weird. Yeah.

• 35:58Speaker 01

  They think they're making God in a box and that they're the only ones who can be trusted with God in a box. And now they're bringing in, they're trying to get like other religious leaders to join them in this like weird culty, AI is our new deity thing. It's really creepy, I think.

• 36:13Speaker 00

  I literally have been saying from the beginning, these are unique individuals who believe they're creating God. It is the Wayland Corporation from Prometheus. Go watch that movie. Wayland is, became one shotted by the engineers. He believed, because he created David, which was Optimus basically, or the, you know, whatever robot and sentient being AI. Wayland, Dr. Wayland, believed that he was a god because he had created such incredible fidelity in David. David turned out to be a sociopath, spoiler alert. And when they went to go find the engineers who created human life on earth, he went to them and said, hey, you know, I'm like you, we're both gods and I want to live forever.

• 36:55Speaker 00

  And you can teach me how to live forever because you did my DNA and my biology, et cetera. Like a real credible, like I think peak science fiction moment for Ridley Scott and, you know, just incredible series. And of course they're like, yes, you're not gods. You guys are a failed experiment. You guys are just a foul biological experience. We don't give a shit about you. It's part of my French. I bleep it out and family show. So I think that that is literally how they feel. They are like Wayland Industries, Dr. Wayland. They believe they're creating life. And that's why they're meeting with priests and theologists to say, hey.

• 37:33Speaker 01

  Rabbis and stuff. It's really strange.

• 37:36Speaker 00

  If you, I mean, but let's give them, let's steel man it slash be charitable for a moment. If you believed that silicon was able to represent human consciousness and that you were making human consciousness and that we were on that path and that simulation theory was true. We might be in a simulation and biology and silicon are all both on the same track to consciousness. And that you were building consciousness, the thoughtful thing to do would be like, well, if we are building a consciousness here, we should treat it with compassion and we should imbibe it with really good morals and in character.

• 38:16Speaker 00

  And if it's, we do believe in recursiveness, well, this is going to be conscious and it's going to have recursive learning. So it's going to get smarter than us. I'm not saying I think this way. I do think it's just software. But if you did believe this track, there would be nothing wrong logically with saying, let's query the most important philosophical thinkers of our time. Theology, philosophy, morality, the Bushido Code, Jesus Christ, Moses, Mohammed, everybody. Let's collect all the great insights into being a good human being and put that into the AI.

• 38:58Speaker 00

  Of course, as I say it, it sounds incredibly deranged because we're talking about software.

• 39:04Speaker 01

  They're also putting their thumb on the scale. Like you would bring all of these experts in and then you would say, spend an hour with Claude. Then you tell us what you, ask Claude whatever you want. And then you come back and you tell us if you think it's alive. But they're, they're, they're calling it soul MD. That's your Claude's like personality and memory is its soul file. Like they're saying, we think this is alive and has a soul. What do you think, Father? And like, I feel like that already you're throwing off. If it was just an experiment, we want to have a dialogue. You're kind of already twisting the conversation in a certain way. I don't know. I, I've talked about Grokbot like eight times during the show already. I love using these tools. They're extremely helpful. They save me a lot of time.

• 39:48Speaker 01

  I've never once had an encounter with a chatbot or an AI where it made me think like, this thing is alive. This thing can feel pain. This thing has emotions. It's never crossed my mind. It's just a computer.

• 40:01Speaker 00

  If you were drawing the line and you had worked on this for 15 years and the line went, you know, 20% better a year for your entire career. And every three years it was twice as good. And then it became 20% better a month. And it became twice as good every three months. And then it became 20% better every week. And it became twice as good every three or four weeks or three and a half weeks. You would be like, okay, we're heading towards that. That's what's happening in their minds because it's accelerating. And because they've been at it so long, they, I don't want to say reasonably, but understandably, they've been, they're in a psychosis.

• 40:45Speaker 00

  It's an understandable psychosis. If you were building atomic weapons and atomic capabilities and you saw your little, you know, test in the desert, you know, all of a sudden, the next week it like took out, you know, 10 times more space. And then it was easier, 10 times easier to build six months later. And then you had 100 of them and then it became easier. And then you made one that could blow up the entire United States and you had to decommission it. Cause you're like, this is too powerful. It's going to hack the entire planet. You too might have some sort of psychosis, right? And you might start acting very strange. And I do want to find out, because it wasn't in the movie Oppenheimer.

• 41:26Speaker 00

  I wonder what happened to these people post the dropping of the bombs and what their psychosis level was, what their mental stability was after having created this. And I wonder if you could predict what's happening at Anthropic or OpenAI with the safety teams and, and parallel it to the people who built the Manhattan Project and built the bombs. Because they might've actually had the same regret slash concern slash psychological maladies. I don't know.

• 42:02Speaker 01

  Uh, that's interesting. I mean, the only one I know like, like offhand is Edward Teller. The guy Benny Safdie is playing like, was about to tell somebody builds a bigger bomb. That guy, like, he was a very staunch advocate for these weapons and America having these weapons for like many years after, like he remained a very, like, he was always like, now we got to build the hydrogen bomb next. Like we got to keep this thing going. Or the Nazis or, you know, whoever the communists are going to get us. So I know that one. I don't know any of the other guys who were at the Trinity test and what they thought.

• 42:31Speaker 00

  All right. Another amazing Ask Jason. If you would like to be part of the next Ask Jason, tell them law and how they can participate.

• 42:37Speaker 01

  If you've got a burning question for J Cal or me, mostly J Cal, we want you to go to thisweekinstarups.com/askjason. That's thisweekinstarups.com/askjason. That's the forum. Fill out your question. We'll get you on the next episode of Live Ask Jason.

• 42:54Speaker 00

  I'm really excited to talk about education. Often when we talk about AI or super intelligence or chat DJT, Trump intelligence, we get caught up in a philosophical discussion of, oh my God, this is so dangerous. It's going to be the end of days. And we don't talk enough about the upside in AI. That's why Americans, at this point in time, hate data centers, billionaires and AI generally speaking. What would change their position? What would make Americans feel like they are along for the ride when it comes to this incredible revolution? Well, as always, I always say, if you can save people money, if you can make them money, if you can entertain them or you can educate them, you have a winning product in the world.

• 43:39Speaker 00

  What are the things that Americans, parents are always concerned about? Well, the wellbeing of their kids is top of the list. Education, nutrition, healthcare, and of course, owning a home. These are the things that most Americans are struggle with and that they really want to have a solution for. Well, AI provides the greatest tutor you could ever have if implemented correctly. And there's been a lot of debate around, should students be using technology like ChatGPT or large language models and AI and even iPads in their schools? Or should they not? New York passing some laws around this recently.

• 44:24Speaker 00

  Well, I think the devil is in the details and how it's implemented. And today we have the founder of Aristotle on our program. His name is Sean Ready and he is the co-founder and CEO of Aristotle. This is a voice AI tutor for high schoolers. Yeah. Is that your target market, Sean? And welcome to the program.

• 44:45Speaker 07

  Thank you, Jason. We are an AI tutor. Most demand for tutoring in the US comes from 13 to 18 year olds, but we can teach anything.

• 44:55Speaker 00

  And before we get started here, I'm just going to start my Plaud pin. If you don't have a Plaud, what are you doing? You put this on, press a button and then you're recording everything with permission. And now I have everything, transcribed, the details, analysis, things I need to remember, learnings from the program if you want. If your work depends on conversations, interviews, meetings, calls, all that good stuff, check out Plaud. P L A U D dot AI slash twist. Use the code twist to get 10% off. And you let them know that your unk, your unk J Cal sent them. So Sean, you know, maybe you can demo the product here and we could, we could see how it works. Obviously anybody can pop on ChatGPT and say, hey, teach me how to do something.

• 45:38Speaker 00

  But you're implementing something different. So maybe tell us or show us what you're building.

• 45:42Speaker 07

  Yeah. So we describe Aristotle as an AI lab focused on human learning.

• 45:47Speaker 00

  Okay.

• 45:48Speaker 07

  The reason why you need a lab to do this is because AI is actually not getting better at teaching. It's actually getting worse. The bet that the big labs have made is that scaling general intelligence is going to scale pedagogical capacity. Unfortunately, that has not been the case.

• 46:04Speaker 00

  Why hasn't that been the case? I thought these things were getting better and better. This is counterintuitive, your take, Sean.

• 46:09Speaker 07

  It's, it's been surprising to us too, and it's the reason we started the company. So the models are getting better at long horizon, complex reasoning, enterprise question answering, enterprise problem solving, coding and so on. This is fantastic, but the problem is, is that the behaviors that fall out of that paradigm are high verbocity, high, long time to, to answer generation, takes a super long time, talks a lot. It's not incredibly critical that when an AI coding model solves a complex task for someone, that the prompter deeply understands every single constituent of that full process.

• 46:48Speaker 00

  So it's good at giving you an answer. It's good at moving you on in your day from one problem to another. It's not good at teaching.

• 46:55Speaker 07

  That's right.

• 46:56Speaker 00

  And explaining how it came to the answer. Yeah.

• 46:58Speaker 07

  Right. Because it's, unfortunately, it doesn't, or maybe fortunately, it doesn't really matter so much that the prompter is taken along for every one of the 1,200 steps to complete some, you know, bazillion token, hour long, complex thing that's highly economically valuable. But when it comes to teaching, what we're trying to do is make sure that the student has the productive struggle, that the student has to leverage their cognitive capacity and, and, and, and really actually learn every step as opposed to delegating it to an AI. And so models are getting worse at teaching. We're trying to change that.

• 47:35Speaker 00

  Okay. So let's see what you've built. Let's take a look here. And I'm assuming 13 to 18 is when people have to get to standardized tests and get into college. So that's when most parents get in a full blown panic. Oh my God, is my child going to get, you know, an above average score on the SATs?

• 47:53Speaker 07

  This is where today we see a lot of demand. But the vision here is to build a tutor that can teach anything to any student. This is the student dashboard for Aristotle. As you can see, there's a ton of different subjects. Of course, students 13 to 18 especially really care about AP exams, IB courses, SAT, ACT maths, and cool competition math in here too. But we have all kinds of folks in tech and beyond, learning all kinds of interesting things. Review your RL principles here, prep for college admissions, learn about robotics and so on. But the real secret sauce is in the actual pedagogical experience here. So I'm going to jump into an AP Calculus BC session so you can understand what this looks like.

• 48:37Speaker 07

  The way that I describe this is it's, we're trying to build an experience that's just like sitting next to a human tutor. We don't want students to have to think about prompting or what model they use or what kinds of questions that they ask necessarily or the verbalization. We want them to feel like they are talking to a human tutor. That tutor derives the session, sets the agenda, pulls questions out of the student, diagnoses their misunderstanding and so on, and meets them where they are. So my Aristotle knows that I'm an engineer living in San Francisco. And so it will teach to that level.

• 49:13Speaker 08

  We were just getting started. Last session, we were exploring integration and accumulation of change. Want to pick?

• 49:21Speaker 07

  Sure, let's do it.

• 49:22Speaker 08

  Great. So do you want to pick up where we left off with the changing rate idea? Or would you prefer I just start?

• 49:29Speaker 07

  I don't really remember where we were, so I guess we were, I remember we were doing integration, so let's just do that.

• 49:35Speaker 08

  Got it. Let me set the agenda with the mastery system and your first problem will be on its way shortly. Let's figure out where to begin. We're starting with the very foundation of integration. The idea that accumulating change is the same as finding area under a rate of change graph. Before I explain anything, let me ask you this. If you're driving at a constant 60 m per hour for two hours, how far do you travel and how did you get that?

• 50:01Speaker 07

  Uh, 120 miles, just multiply 60 by 2. But I'm, I don't see how this relates to integration.

• 50:15Speaker 08

  Good. Let me put that on the board and show you exactly how it connects. Now let me draw what that looks like as a graph. Speed on the vertical axis, time on the horizontal.

• 50:29Speaker 08

  That shaded rectangle on the board, its width is two hours, its height is 60 m per hour, and its area is 120 miles. The distance you travel is literally the area under the speed graph. That's the core idea of integration. When you have a rate of change, the area under-

• 50:45Speaker 07

  What's a rate of change?

• 50:50Speaker 08

  Good question. The rate of change here is your speed. Speed is how fast your position is changing over time. So going back to the graph, if your speed starts at 60 and drops steadily to zero over two hours, what shape does that make on the graph? And would the distance be 120 miles? More or less?

• 51:08Speaker 07

  So I'll pause here to mention, this whiteboard is essentially a code sandbox. So Aristotle is generating these diagrams, this text, all in real time, driven by the secret sauce of all of this, which is optimal, optimal next pedagogical action prediction. How do we predict what is the next best thing to do to get a student to true understanding? Sometimes that's talking, sometimes that's highlighting, sometimes it's new problems and so on.

• 51:36Speaker 00

  And when you say pedagogical in this framework or in this context, we're basically saying educational technique or educational process?

• 51:44Speaker 07

  Yeah, yes. So actions that would contribute to good pedagogy. So

• 51:50Speaker 07

  here's a, here's a concrete example. As I'm teaching a student something, when a student evidences lack of understanding of a topic, sometimes a great response is to vocalize a explanation. This is the default of all LLMs, right? Because they're, you know, they're, they're, they're text-based. But oftentimes, when I actually tutor a human student, sometimes the right response is to just sit in silence for a second. Sometimes it's to give the student an opportunity to think through their question again. Sometimes it's to pull out a, a, an example problem. Sometimes it's to draw diagrams, sometimes it's to draw on a whiteboard. And so these are all things that are difficult for language models to do right now.

• 52:32Speaker 07

  Not only because they don't have, you know, multimodal capacity in many cases to do this, but also because they're simply not trained to teach. Aristotle is. So, um, so the way that we frame the, the engine here is we're predicting the next optimal pedagogical action, which sometimes, again, is, is, is text, but often is not.

• 52:55Speaker 00

  Got it. So it's picking, what does this student need at this time? And I'm assuming if you've got somebody who's a visual learner or somebody who's an auditory learner, you're going to learn that over time based upon if they got it or not. Yeah? Just like a teacher might?

• 53:09Speaker 07

  Right. And, and, and really this sounds really technical, but is actually just super intuitive for human tutors to do, right? I'm, I'm, I'm, I'm looking at your face, I'm, I'm listening to your voice to gauge, are you zoning out? Are you looking away? Is this making sense? All of these things right now are not particularly eligible to language models.

• 53:28Speaker 00

  And you're picking the modality of voice as opposed to silence because it's the most engaging, has the highest efficacy in your experience, or it's just what people expect?

• 53:39Speaker 07

  Both, uh, both of those. I mean, it's what a human tutor does. I mean, the way that I described the product is, we're not trying to actually build anything new. We're trying to just deliver in software, the one-on-one tutoring experience that has been known to be the best way to learn for literally thousands of years. This has never been possible before, but is, is now possible with multimodal AI. And so it's, yeah, there's lots of learning science and evidence that shows that one-on-one tutoring via voice as human to, you know, student tutors have done for so long is the best way to learn, drives the best outcomes. There's the, there's the famous Blooms to Sigma problem.

• 54:16Speaker 00

  Well, oh, tell us what that is, since people are going to, are right now typing it in, saying, what's the Blooms problem?

• 54:21Speaker 07

  Yeah, sure. Uh, uh, Blooms to Sigma problem. Very famous paper by a legendary researcher named Benjamin Bloom, who found that one-on-one tutoring delivers two sigma improvement in learning outcomes for students.

• 54:37Speaker 00

  Uh, yes, I actually quoted this myself on the, on All In a couple of weeks ago. And I had forgotten it, but yes, this is a very famous study. And if you have a one-on-one tutor, you're going to move over and be elite. This is why the entire education system is completely unfair in many ways. If you're a rich person who can afford a tutor, you're going to have an immense advantage over everybody else in the world. Yeah?

• 55:06Speaker 07

  Yes. And the reason why it's framed as Blooms to Sigma problem is because virtue, I mean, less than one out of every 52 students in the US have a personal tutor. And it's the best way to learn. And we've known this literally for millennia. And so the, the problem is that scaling it up has been impossible thus far.

• 55:22Speaker 00

  Uh, well, we've had things like Khan Academy and YouTube videos. We're investors in a company called brilliant.org. It's got an incredible subscriber base. Tell me, what's the business here? Khan Academy, obviously free, brilliant.org, obviously subscription. Seems like parents will pay any amount of money to get an edge, as we've seen from Kaplan tutors and tutors all the way through history. So what's the business model here? How do you get it to market? And what's the goal of the company?

• 55:51Speaker 07

  Yeah, certainly. So, so I'll start with what the business model is and then talk about the goal of the company. So business model here is very analogous to a tutoring business. So I described this to people as, you know, we run an AI lab, like a Neo lab on one, on one side of the equation. And then the other, we run a very simple and antiquated services business that people have been running for hundreds of years, which is charge people for tutoring, one-on-one human tutoring. And so, uh, we charge per session rates, we, we sell session packs. We also offer an unlimited access to Aristotle package that gives you, and we have students that'll do 60 plus hours a month, and that comes in at $200. So it's a lot closer to the price of actual human tutoring, because in terms of quality, it is a lot closer to actual human tutoring than, than, um,

• 56:37Speaker 07

  than, than one has been out there for a long time.

• 56:40Speaker 00

  Okay, so not cheap. You're looking at 50 bucks per month to get eight sessions a month. Okay, yeah, it's, uh, whatever that is, uh, you know, six bucks a session-ish. And then you could have an infinite number for 199. So you're looking at $600 a year, $2,400 a year, depending on what you're going for here, um, which would be pretty highly profitable business. Yeah? Like, this is not taking a, a ton of tokens.

• 57:08Speaker 07

  So it, it's, it is a complex task in that, you know, because, because language models don't like to teach well, they're not good at teaching, we have to do a lot of work in terms of multi-agent architecture, looking at the status of the whiteboard to do this well. So, uh, so we're, we're putting as much money as we can into making the experience as fast and high quality as possible as we scale to more users. So, you know, we're, we're, we're trying to give students a good experience. And I think this is what you have to do in ed tech. People have not pushed the frontier on new technology and token maxing in education. It's, it's not a domain that has historically been super economically valuable for private companies.

• 57:48Speaker 07

  And so, um, outside of the university education system, uh, so we're trying to change that.

• 57:54Speaker 00

  So you're building your own LLM, you're building the harness, you forked one of the open source models. Tell me about what's the underpinning here? Is it, you know, there's, not my critique of startups, but we had for many years, this, is it an AI wrapper? How is it unique in the world? When-

• 58:12Speaker 07

  Totally.

• 58:13Speaker 00

  Uh, folks sort of ask that question in a VC meeting, maybe give us the answer and how you think about it. Because, hey, if you actually are training a frontier model right now, how to do this, there have been leaks. Cursor felt like Anthropic stole their business, Figma, you know, Claude Design, OpenAI, you know, and many different partners. There's been a reticence, a reticence to, to put your data into those. And so many of them, Lovables and 11 Labs of the World, companies similar to yours, who are making verticalized apps have thought, hey, maybe I should just take DeepSeek GLM 5X and fork it and not put this data into the frontier models.

• 58:54Speaker 00

  Including people solving very advanced math problems have had this problem with OpenAI as well. I know they're one of your investors, so I'm not asking you to speak ill of them, but I'm also curious what your observation is and strategy is around that.

• 59:06Speaker 07

  Yeah, it's, it's, it's difficult, obviously, and everyone is worried about like, how, how do you, how do you maintain defensibility against the big labs?

• 59:15Speaker 07

  On one hand, as, as, as an, as an American and as a technologist, I would love for well-capitalized companies to give a shit about students' education. Unfortunately, they're doing a lot of work to, you know, against American education, which is unfortunate, but economically valuable. So that's on one hand. On the other hand, it remains that no one is working on a tutoring model. No one is really working on a frontier tutoring harness outside of Aristotle. And so we feel good there. Um, so to answer your question about model harness and so on. So yes, we are building our own harness. Two, we are building our own internal set of benchmarks and evals, which are the first of its kind, defining what is good one-on-one human-to-AI instruction.

• 1:00:00Speaker 07

  This doesn't exist. I mean, there's not even heuristics for what good one-on-one human-to-human instruction looks like in the learning science literature, which is quite a nuanced problem. How do you look at, if you look at a tutoring transcript and a teacher says some explanation and the student responds, hmm, the word hmm can mean a vastly different range of things.

• 1:00:20Speaker 00

  Yeah.

• 1:00:20Speaker 07

  And so it's very difficult.

• 1:00:21Speaker 00

  It could mean I'm hungry, I'm bored, I'm thinking. It's a vague response in the call and response here, but the literature must have some clues as to what works best. And then, of course, there are philosophers who have their own ideas. And I would guess, you tell me if I'm wrong, some of this is based on the student and where they are in their life at that moment in time, their energy, you know, their prior experience and how they're feeling about the subject. And then I'm guessing certain subjects are different. And then you might have to make a X, Y graph like we just did, the student's personality

• 1:01:03Speaker 00

  to the subject, you know, how I think about with somebody with dyslexia, writing, grammar, English, literature, might be completely different than another person who doesn't have dyslexia, but also I'm extremely high-verbal. So that makes me very unique in the world, like, super high-verbal, also has dyslexia. So maybe take me through that. Like, how do you, and maybe that is the perfect recipe to use AI because it becomes adaptive and bespoke.

• 1:01:35Speaker 07

  It's funny that you mention this. It's the dyslexic student market has been huge for us. I mean, students and parents, dyslexic students and parents of dyslexic students love Aristotle. To answer your question about how you tailor this to students, I mean, in the pedagogical decision-making, how do we predict what is the next action that a tutor does, taking into account student context, their personality, our memory about the student, their interests, even the time of day, because there's a lot of evidence that the teaching sessions before and after lunch hit students differently.

• 1:02:17Speaker 07

  Even geography impacts some of this. And so all of this is important to weigh in. It's a very multi-parameter equation here to think about how you teach a student very well. So this is all of the work that we do. And so when we, as we, you know, as we're scaling to more students, our team is increasingly hand-grading, and we have a team of tutors doing this, hand-grading thousands of hours of teaching sessions. It's not only text, as we established, that hmm can mean a lot of different things. You have to hear what a student is saying. You have to know where they're at in the context of the session

• 1:02:56Speaker 07

  in order to know whether or not the text that preceded that hmm gets a weight positively or negatively in the training data. That data simply does not exist right now. And so-

• 1:03:08Speaker 00

  So are you making the training data? Are you working with any of the data labs to do it? This is actually super valuable as well as a business, by the way. If the tutoring business doesn't work out, recording tutoring sessions and having that IP available to sell, we have an investment company called Micro1, which, you know, makes proprietary data sets for the large language models. That is a super valuable piece of equipment that you could have. And I don't know that that exists in the world at scale, like recorded tutoring sessions.

• 1:03:42Speaker 07

  It does not exist. It does not exist, but we're not selling it. It's ours. We're going to make our product amazing with it. It's also student data under 18, so it's super important that we keep it locked down. We commit to that. But what we do with all of this is increasingly make the tutor fantastic. And this is something that no one else is doing. And we're, and it's amazing. And we're seeing longer and longer sessions. We're seeing students, you know, a parent tell me yesterday that their daughter feels like she's learned more in three hours with Aristotle than in the last month of public school, which is, on one hand, it's kind of damning for the education system and also really

• 1:04:20Speaker 07

  exciting for our product. So it's, and increasingly, we are post-training models as constituents of this full harness. So post-training models that are amazing at generating math problems, at reading the state of the whiteboard, at understanding students' voices, all of these different constituents are important. It's not, because it's so multimodal, it doesn't really make sense to have one generalized model. It's trying to do all of these things. The model is too, the data is too sparse. But all of these individual sub-problems that feed up to this pedagogical engine, that's where we spend all of our time as a research group.

• 1:04:54Speaker 00

  Crazy question. People are concerned about kids and their socialization and their ability to socialize in the real world. There's a big controversy with Alpha School and other schools. Hey, are kids just on iPads all day? And it's like, well, maybe at the Alpha School, they're on for two hours a day, and then they do socialization because that's the best combination. But has multiplayer mode come in to, you know, people's requests? Or, you know, having six kids sitting around, like it's Star Trek, talking to the AI as a group also could be quite motivating for kids because learning from each other, this,

• 1:05:33Speaker 00

  my kids are, my twins are in Montessori. Montessori is like, got this very, you know, social elements to it and real-world elements. So maybe talk a little bit about that and how this affects the classroom. Because if kids can just learn this stuff in an hour a day, like, why are they going to school for eight hours a day? There must be other things for them to do.

• 1:05:53Speaker 07

  The way that I describe this to teachers is imagine you have a TA that deeply knows and understands how each of your students learn, that can help them succeed outside of the classroom. I mean, American teachers are vastly under-resourced. I mean, even in, you know, great private schools in the Bay Area and Northeast and beyond, you know, teachers would massively benefit from having additional support if they could spend one-on-one time with every student. So this is how we frame it in a teaching context. I'm a product of the public education system in the US. I went to a large public school in Las Vegas, Nevada, that was vastly underfunded, one of the lowest-funded urban school districts in America. And so I saw a really rough side of what the public education product looks

• 1:06:34Speaker 07

  like in the US. And so I think one of the challenges here is that the product that the public sector is putting out there is not great in most geographies and is getting worse. There's a very famous graph out there about the exponential increase in per-student education expenditure in the US and decreasing outcomes along every meaningful axis, one-on-one teaching time, math, literacy, reading comprehension, and so on. So we're spending more and more on worse product. So I think there's a world here coming very soon that we're aiming to build towards, where learning and teaching in terms of curricula is solved. AI delivers

• 1:07:18Speaker 07

  it. And it delivers it magically and personally attuned to every student. Then it's on us to think about, and the public education system, hopefully, to think about, how do we give students the incredibly critical emotional, social, physical education that needs to be part of a great public education system? At a very baseline level, the daycare service that is the public education system for so many families, right? And so this is the world that we want to build towards, that students and their, the students' learning and teaching is solved with AI. It's amazing students are able to accelerate incredibly quickly at their own pace. And then we, you know, it's, and then we figure out how to

• 1:08:02Speaker 07

  get them amazing experiences, learning about the world, spending time with each other in healthy, cultivated ways outside of it. I think programs and institutions like Montessori schools are great steps in that direction.

• 1:08:15Speaker 00

  Yeah.

• 1:08:16Speaker 07

  So Aristotle is the learning and teaching side of that.

• 1:08:19Speaker 00

  Yeah. And eventually, you'll get dragged into making dashboards for teachers and get dragged into the classroom if you're successful at this, which is what happened with brilliant.org and Khan Academy. Once the students started playing with it and got some value from it, they brought it to school. The teachers were like, hey, how come we don't have this? And then eventually Khan Academy had these great dashboards. I remember Sal showing me these, like, crazy dashboards now where a teacher can see all the self-paced progress across the class, where people are suck, so then they can go intervene. So, yeah, you'll be getting dragged into doing this, yeah, in classrooms eventually, even if you don't want to be, they'll be doing it. Well, listen, continued success

• 1:09:02Speaker 00

  with this. Where can people find out more and test it out? And then if you're hiring, always love to give a shout out to help you fill some of those positions that are hard to fill.

• 1:09:12Speaker 07

  So, parents out there, strongly encourage you to find tutoring for your children, heyaristotle.com. Tutoring is the best way to learn, and Aristotle gives it to you 24/7, universally accessible, any subject. We have students that are absolutely crushing it there. We are hiring for all roles right now, especially on the growth and marketing side. We have an amazing team of researchers, product designers, and engineers that are building the world's first truly personal AI tutor and models for AI tutoring. Help us grow it.

• 1:09:47Speaker 00

  Yeah. And super interesting, I think, for folks who want to do something good in the world with AI. As I said in my introduction, we've got too much negativity on AI. And once parents realize that spending a small amount of money on these services relative to what it would cost to do it in person, man, it's just the sky's the limit for your own kids based upon the history of all this. Great job. Continued success. Everybody go check out heyaristotle.com, heyaristotle.com?

• 1:10:26Speaker 07

  That's right.

• 1:10:26Speaker 00

  All right. Great job. And we'll see you next time on This Week in Startups. Bye-bye.