What's Going On With Google Gemini Deep Research? | E2072

This Week in Startups is brought to you by… PrizePicks. TWiST listeners, download the app today and use code TWIST to get $50 instantly after you play your first $5 lineup! Go to: https://prizepicks.onelink.me/LME0/TWIST PrizePicks - Run your game. Coda. TWiST listeners get 6 months at https://www. Coda.io/twist Fitbod. Get 25% off your Fitbod subscription or try out the app for FREE when you sign up: fitbod.me/TWIST Today's show: Jason and Alex interview Aarush Selvan, lead PM at Google Gemini Deep Research, and discuss their groundbreaking work. They also took a moment to cover TikTok's dire situation and AI investment to start the year. (0:00) Show Teaser (1:26) Jason Calacanis shares his Japan skiing experience (6:18) Arush Selvan from Google Gemini team background discussion (9:26) PrizePicks. TWiST listeners, download the app today and use code TWIST to get $50 instantly after you play your first $5 lineup! Go to: https://prizepicks.onelink.me/LME0/TWIST PrizePicks - Run your game. (14:23) Deep research demonstration and server resource usage (19:42) Coda. TWiST listeners get 6 months at https://www. Coda.io/twist (22:11) Google's business model and AI-generated research reports (29:55) Fitbod. Get 25% off your Fitbod subscription or try out the app for FREE when you sign up: fitbod.me/TWIST (32:00) Personalized data integration and content permissions in AI tools (34:22) Differentiation and hallucinations in AI-generated content (44:11) Business models for AI search engines and OpenAI's new task feature (54:30) Synthesia's AI-generated video tool demo and privacy in biometric payments (59:36) Investments in AI companies and TikTok's potential ban (1:04:17) Chinese government influence and US rule of law in tech (1:07:29) Irony of migrating to Little Red Book and upcoming inauguration (1:08:02) Staying updated with This Week in Startups Subscribe to the TWiST500 newsletter: https://ticker.thisweekinstartups.com Check out the TWIST500: https://www.twist500.com Subscribe to This Week in Startups on Apple: https://rb.gy/v19fcp Check out Google Gemini Deep Research: https://blog.google/products/gemini/google-gemini-deep-research/ Check out Open AI Scheduling: https://help.openai.com/en/articles/10291617-scheduled-tasks-in-chatgpt Check out Synthesia: https://www.synthesia.io/ Follow Aarush: X: https://x.com/AarushSelvan LinkedIn: https://www.linkedin.com/in/aarush-selvan-3a376a8b/ Follow Alex: X: https://x.com/alex LinkedIn: ⁠https://www.linkedin.com/in/alexwilhelm Follow Jason: X: https://twitter.com/Jason LinkedIn: https://www.linkedin.com/in/jasoncalacanis Thank you to our partners: (9:26) PrizePicks. TWiST listeners, download the app today and use code TWIST to get $50 instantly after you play your first $5 lineup! Go to: https://prizepicks.onelink.me/LME0/TWIST PrizePicks - Run your game. (19:42) Coda. TWiST listeners get 6 months at https://www. Coda.io/twist (29:55) Fitbod. Get 25% off your Fitbod subscription or try out the app for FREE when you sign up: fitbod.me/TWIST Great TWIST interviews: Will Guidara, Eoghan McCabe, Steve Huffman, Brian Chesky, Bob Moesta, Aaron Levie, Sophia Amoruso, Reid Hoffman, Frank Slootman, Billy McFarland Check out Jason's suite of newsletters: https://substack.com/@calacanis Follow TWiST: Twitter: https://twitter.com/TWiStartups YouTube: https://www.youtube.com/thisweekin Instagram: https://www.instagram.com/thisweekinstartups TikTok: https://www.tiktok.com/@thisweekinstartups Substack: https://twistartups.substack.com Subscribe to the Founder University Podcast: https://www.youtube.com/@founderuniversity1916

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  Deep Research is designed to be your personal AI research agent. The way it works is you sort of give your query. It'll first break it down into a research plan. And by the way, we should totally just share my screen. Actually, I can just talk while we're on this. Okay, perfect. Awesome.

• 0:16Speaker 00

  And then it researches the web. First step is it's going to put together this research plan. The plan looks pretty good, so I'm just going to go ahead and start the research.

• 0:22Jason Calacanis?

  How does it make the plan while we're sort of sitting here watching you do the research? Which I find these queries take five minutes or so, which makes me feel good. I don't know if it's actually taking you 10 seconds and you're making me wait four and a half minutes just to make me feel like you're doing a better job and it's more worth it, but.

• 0:40Speaker 00

  No, no, it takes us the full five.

• 0:46Speaker 02

  This Week in Startups is brought to you by PrizePix. Run your game, the best place to get real money action while watching your favorite sports. Download the app today and use code TWIST to get $50 instantly after you play your first $5 lineup. Coda. Coda empowers your startup by bringing words, tables, and teams together. Strategize, plan, and track goals effectively with all your valuable data in one place. Go to coda.io/twist to get started for free and get six free months of the team plan. And Fitbod. Tired of doing the same workouts at the gym? Fitbod will build you personalized workouts that help you progress with every set. Get 25% off your subscription or try out the app for free when you sign up now at fitbod.me/twist.

• 1:26Jason Calacanis?

  All right, everybody. Welcome back to This Week in Startups. It is Wednesday, January 15, 2025, and we're back. I'm back from Japan and I had the most wonderful trip. My second time, Alex, going skiing in an area known as Hokkaido. If you look at the map of Japan, you'll see there's like an island at the top of the island. You can only get there by plane, ferry. There's a tunnel, there's a train, but it's a little bit of an adventure. You got to do like a two hour drive from the airport to get to Niseko. Niseko is a ski mountain. It's actually five different mountains. Four of them are together called United Niseko, and you ski there.

• 2:12Jason Calacanis?

  The reason you go there is because the powder, the snow is, and there is Hokkaido, very good. And if you were to type in Niseko, you could go to the ski village of Niseko, N-I-S-E-K-O. There's your Niseko ski resort right there. And it is an extraordinary place to go. Why is it extraordinary? Well, if you look at something called open snow, I pay every year for open snow. It's a bunch of meteorologists. Couple of years ago, I was looking for powder because somebody told me you could powder chase, Alex. And so I said, okay, I'm gonna powder chase. I want to get some of this fresh powder. And what I saw consistently was, like, Canada, Japan, and the Rockies and other places got like

• 2:56Jason Calacanis?

  really consistent snow. But Niseko was always going up to the top of the list when you ranked by how much snow has come. Three years ago, I had a speaking gig. I told my speaking bureau, anywhere there's snow, Utah, Japan, Canada, I will do the speaking gig. Because then I could put a snow ship on it. And so they started looking. I got a tiny speaking gig in Japan one time, two years ago. I was like, yeah, I'll take it. It was at like one-fifth of my normal rate. I didn't care. I was like, if this pays for the hotel, I'm good when I go skiing. And I just went back again. And it is extraordinary. I went to a place called IWAANI, which is I-W-A-A-N-I. You can go look that up. IWAANI cat skiing.

• 3:39Jason Calacanis?

  And cat skiing is when they drive you up a mountain on a cat, which is like a tractor. The same thing that ran over that actor in Tahoe's leg. You remember that?

• 3:49Speaker 03

  I do not remember that, actually. But here is the Iwani Resort. Oh, my gosh, you went there?

• 3:56Jason Calacanis?

  I was there. It's my second time there, my third day skiing there. It's a bit expensive. It's on the pricey side to have a mountain to yourself.

• 4:03Speaker 03

  Yeah, for that, I'd pay whatever, right? I mean, look at that. That's insane.

• 4:08Jason Calacanis?

  Yeah. So what this is, Alex, is there was a ski resort there, but it was abandoned. And so all the villages inside of, and there's the cat, and you see that little greenhouse box on the back of it? You kind of go into that, and they drive you up the hill. You get to do maybe six, seven, eight, nine runs a day, depending.

• 4:28Speaker 03

  Is there a bar inside the greenhouse?

• 4:31Jason Calacanis?

  There is no, but actually interesting you say that. They serve homemade donuts. The guy, John, who's a friend of mine who runs it, um, he, his wife makes donuts, these like little stick donuts every morning and then grill a great coffee. And when you're in the back of that, they will pour you a cup of coffee and give you a warm donut. Absolutely. It's run her experience. Uh, you can see me smiling.

• 4:52Speaker 03

  And people say money can't buy happiness. They've never been handed a fresh donut on a cat on a Japanese mountain about to go powder skate. And turns out there's a price tag on joy.

• 5:01Jason Calacanis?

  There is. I am not a gatekeeper. I let everybody know the great things I'm doing. Uh, I want you to, I want to share it with everybody because, uh, you know, spread a little joy. Anyway.

• 5:11Speaker 03

  Yeah, spread a little joy.

• 5:12Jason Calacanis?

  Go to the

• 5:12Speaker 03

  Glad to have you back.

• 5:14Jason Calacanis?

  Yeah, it's good to be home.

• 5:15Speaker 03

  We missed you. Um, also it's much easier to film and we were one hour apart versus 14, turns out.

• 5:21Jason Calacanis?

  I mean, thank you for doing the seven, eight AM early, uh, shows when I was at 10 PM.

• 5:27Speaker 03

  We were both suffering. Let's be honest. That was neither one of us was like, this is fantastic. Let's do this every week.

• 5:33Jason Calacanis?

  It was, um, you know, sometimes I do things because I like to push myself and it was a strategic mistake. Uh, and I should have just, it would have been easier if I just did the, uh, I, I, recorded the show or I had somebody else co-host with you for a day or two, but it's great to be here. Great to see all the live folks coming in.

• 5:52Speaker 03

  Yeah.

• 5:52Jason Calacanis?

  Uh, and Wayne and Mark and all these great folks who are in the Nody gang. If you want to join the notification gang, it's very simple. Go to youtube.com and, uh, search for this week in startups and you can put the bell on and then you will get a notification. And if you get to watch us live, you, uh, get to be part of the Nody gang and say, hi, all the Nody boys are back.

• 6:19Speaker 03

  Indeed. The boys are in fact back in town. Uh, Jason, we have a lot to get to today, but I thought we'd start up top with our guest. We're very excited about this. You are a fan of something called deep research from the Google Gemini team.

• 6:30Jason Calacanis?

  Yes.

• 6:31Speaker 03

  I have subscribed and played with this. I am now a paid Gemini customer, upping my monthly AI spend to 40 from 20. Thanks, Google. Uh, but today we have, uh, Arush Selvan. He's the deep research product manager from Google. Let's get Arush up here and we're going to pester him about what he has built and how it's doing. Jason, I'm glad that you brought this up. I'm in love with it, but Arush, hey, how you doing?

• 6:52Speaker 00

  Hey, how's it going?

• 6:53Speaker 03

  It's good to go.

• 6:53Jason Calacanis?

  Arush, how are you, sir?

• 6:55Speaker 00

  Doing great, doing great. Thanks so much for having me on.

• 6:58Jason Calacanis?

  Yeah, I mean, I've been very impressed with what you've done with Gemini. Now, I wasn't impressed for the first, I don't know, year. You know that. I gave you guys like a C, a B minus. It was pretty janky, the original versions, but some things happened inside of Google where you guys are cooking with oil now and the deep research product is the best thing you've done in AI for consumers that I've ever seen. Or that's public facing. What's changed at Google in terms of product velocity and getting more product out?

• 7:31Speaker 00

  Yeah, I mean, you gave a bunch of people who care about their GPAs a B minus and of course, we're gonna pick that up, right?

• 7:39Jason Calacanis?

  So it's directly related to me making you.

• 7:41Speaker 00

  It was all you, Jason.

• 7:43Speaker 03

  Don't inflate his, no, no, no, don't inflate his ego more. That's not necessary. I have to hang out with him all the time.

• 7:49Jason Calacanis?

  Well, people are noticing. The world is aware in many ways of what's happening. And have you, how long have you been at Google? How long have you been doing AI?

• 7:59Speaker 00

  I've been, I've been at Google about five years. I've been doing AI the whole time. I started off

• 8:06Speaker 00

  working on sort of our next generation TPU chips. And then sort of just worked my way up the stack, going like working on compilers and then speech and NLP technology and finally working on Gemini.

• 8:18Jason Calacanis?

  Yeah, and you guys have started to release stuff more often. And I think that's really great to see out of Google, which when you're at a big organization, you have some downside. People expect Google to be perfect. They expect everything to be, you know, extremely elite, but let's face it, you got competitors out there, you got startups, you got big companies and people are releasing and people want to play with this technology. So there's been a little bit of a, I think, looking at it, a philosophical change at Google, which is with the AI stuff, we're gonna put stuff out there. Obviously, you're red teaming it and testing it, but it does seem like the velocity has gone up. Is that correct?

• 8:58Speaker 00

  Yeah, absolutely. I think a lot of us.

• 9:02Speaker 00

  Especially in the NLP community, we've always had this vision of building an amazing personal assistant, right? And finally, we're starting to have the technology that that makes that possible real. And so people are just super excited. Uh, and you know, uh, exploring whole different new avenues uh with this technology. And so, um, yeah, it's just been a ton of fun on the team getting to ship stuff.

• 9:25Jason Calacanis?

  I love sports. I watch every single Knicks game. I watch Knicks Fan TV after the Knicks game and I like to get some real money action while I'm watching the game. It keeps me engaged. It gives me something to root for. Aside from the binary win-loss for my Knicks, let's talk about PrizePicks. It's simple, it's fun, it keeps you locked into the action. Here's how it works. You register, you make a deposit. It's easy to make a deposit. It literally took me 10 seconds. And then you pick more or less on two to six player stats. The more of these players you include, the bigger the payoff can be, up to a thousand times your cash. And you know, obviously, if you're not into basketball, football, baseball, they got everything. You lock in your picks right before the game and then you watch in the app.

• 10:07Jason Calacanis?

  So I'm watching the screen, I'm looking at my phone and I'm seeing my progress. Will Jalen Brunson have more than this number of points, assists and rebounds? How about Karl-Anthony Towns on the rebound tip? What's my guy OG and Ananobi gonna do? And PrizePicks has Flex Play. That lets you cash out even if your picks aren't perfect. Hey, you win two out of three, three out of four, four out of five, you can get a little bit of money back and you can still double your money even if your picks don't hit. So you get a lot of action and I like a little action. I like the zest that comes from using PrizePicks. They now offer MasterCard. I use PayPal, it works. When my picks hit, I can get my money in as little as 15 minutes. It's fast, it's secure, it's exactly how it should be. PrizePicks is the best place to get real money sports action.

• 10:51Jason Calacanis?

  Join over 10 million members and get started now. Download the app today and use the code TWIST to get $50 instantly after you play your first $5 lineup. That's right, 50 bucks guaranteed. You don't even have to win. PrizePicks, run your game.

• 11:06Jason Calacanis?

  I was talking to your boss Sergey and he is really, really into it. He told me he's going to work every day. He's super engaged. He's in the office and he was like, give me all the notes you got on everything. And I sent him a note about, hey, this is really impressive. Deep Research, and I've known Sergey for a long time, the Deep Research is a really incredible and impressive product. So I thought maybe you could show it to us and explain how it works to the audience and then what you think the business case is for it because, you know, we do try to do some analysis here and we have researchers to prepare us for the, for the pod and Alex writes some great show notes. And what we're seeing is when there's a new

• 11:46Jason Calacanis?

  topic, whether it's like election integrity, or it's self-driving and how long it would take to put a robot taxi fleet out there, what the cost would be, whatever it is we're researching, we're finding Deep Research gets us a really beautiful summary that I'm going to say would take four to eight hours for a human, college-educated researcher who gets paid, I don't know, on average 30 to 50 bucks an hour, depending, you know, serious researcher, you know, not like a researcher, you know, offshore or something, just doing generic research. I'm talking about like a US-based researcher. And if you were to

• 12:27Jason Calacanis?

  show me your research from Deep Research or that research work, I'd have to say I probably couldn't tell the difference.

• 12:35Speaker 03

  Actually, Jason, before we jump over to Rush, can I show you what I prepared? Because I use this and I, I'm actually pretty impressed and slightly worried for my job. So Rush, we'll hand you over the brain here in a second, but I just wanted to give a demo of this from the, from my perspective. So I asked Deep Research to write me a report about battery technology. We recently had an interview for the TWIST 500 with a company that's making next-gen batteries. So it was top of mind for me. And I asked it, hey, tell me about this battery technology. Who's doing what, where? This is what I came up with in a couple of minutes. It talked about the current state of the market, the company in question that I was thinking about. And then this is where I went from impressed to annoyed. It did all this work and then it built this table. And, and this is when I was like, okay,

• 13:18Speaker 03

  fine, smarty pants. This is actually better than what I could do in a morning prepping for the show. So, Rush, with that introduction, I'm going to hand you the con, but my God, scaring me.

• 13:29Jason Calacanis?

  Well, what it does is it allows us to cover more topics better and then move up the stack to, you know, having a deeper, richer discussion about it. So what might take three or four episodes for all of us to get up to speed on this new battery technology, we can get done in one episode as opposed to an arc of two or three. But yeah, let's uh, maybe, yeah, give us a demo of this and explain what's happening in the background. When you ask a query, it then goes and it seems to figure out what the sub queries are or what your follow-up questions are. And then it puts them all into, it does some sort of web research, you know, on the fly. Maybe it indexes a hundred, 200 pages, gives you really good citations

• 14:11Jason Calacanis?

  and then spits out like, hey, here's a deep overview that you didn't ask for. And that's kind of what impresses me the most about it is getting it to fill in things that you didn't even ask for.

• 14:23Speaker 00

  Yeah. Yeah. So I think, I think you're hitting the nail on the head. I'm happy to, happy to give a quick demo. But basically, um, Deep Research is designed to be your personal AI research agent. And the way it works is you sort of give your query and it'll first break it down into a research plan. And that's an opportunity for you actually, you to give feedback to Deep Research to say, hey, you know, I don't care about steps three or five, or, you know, I want you to actually go in a different direction. And then, um, and by the way, we should totally just share my screen. Actually, I can just like talk while, while we, while we're on this. Okay. Perfect. Awesome. Um, and, and then it researches the web, right? So the way you try it is if you just go to your Gemini,

• 15:08Speaker 00

  you click on the dropdown and go to 1.5 Pro with Deep Research. And, and here, let's say we wanna, we wanna sort of, let's, let's do like a similar question to your battery tech one, right? Which is like, help me learn about the latest technology breakthrough in small fusion reactors. And what are the most interesting companies in the supply chain, right? So like, uh, like, let's say we're trying to get smart on, on what's going on in this industry since everybody's talking about it. So this first step is it's going to put together this research plan where it's going to try a bunch of different ways of getting this information, right? So from things like articles and research papers to companies and then for each company, find information about the technology, funding and partnerships.

• 15:50Speaker 00

  And so the plan looks pretty good, so I'm just going to go ahead and start the research.

• 15:54Jason Calacanis?

  How does it make the plan while, you know, while we're sort of sitting here watching you do the research, which I find these queries take five minutes or so, which makes me feel good. I don't know if it's actually taking you 10 seconds and you're making me wait four and a half minutes just to make me feel like you're doing a better job and it's more worth it, but.

• 16:12Speaker 00

  No, no, it takes us the full five.

• 16:16Speaker 00

  And, yeah.

• 16:17Jason Calacanis?

  So what, how does it know what research plan to do? Are you just asking the LLM, hey, you're a researcher, pretend you're a researcher at like a consulting group and come up with the sub queries and then you just sort of have that happen each time? How does that part work?

• 16:35Speaker 00

  It's a little bit of that. It's also trying to get the model to understand like what is your underlying intent and therefore what as a plan makes sense, right? So in this case, it does this sort of nested steps where it says, you know, let's first find companies and then, hey, for each one, you probably want to pull the funding information, what the technology is, things like that. But if you give a very simple prompt, like if you're just like, tell me about batteries, it's going to give what it thinks is a good starting point, but really that's when you can come and give feedback and say like, no, no, no, I want to know about like lithium air batteries and how like, what is their positive commercialization, right? Like, so there is an opportunity, like the more specific you can get, the better, but

• 17:20Speaker 00

  there's also a stage there where you can say, okay, if I ask something super basic, it's going to suggest a bunch of things that I may not have considered. And then based on that, I can give feedback.

• 17:31Jason Calacanis?

  Got it. Now, when it goes out and it says, hey, we've got 83 websites we want to research, is that going out and doing like a bunch of literal Google searches and then saying, hey, we're going to do these Google searches, find the best used page rank and the classic Google data set for the last couple of decades and say, hey, we're going to know what the best 10 blue links are. Then let's organize those and see if there's more research on those pages that might tell us, you know, what sub instructions I'll call them or sub tasks to do. Is that what's happening there?

• 18:08Speaker 00

  Yeah. So basically that research plan that it's worked out, it then figures out what steps it can do in parallel versus have a dependency, right? And then for each step of that research plan, it's going to issue many different Google searches. And then it's what it's going to do is after as it gets the results in, it's going to actually reason over all the content it finds to figure out what searches it needs to do next in order to complete that step of the plan.

• 18:38Jason Calacanis?

  Got it.

• 18:38Speaker 00

  So it's basically, and so it might find, you know, hey, there's still some missing information, or oh, there's this really interesting nugget of information that I should double click on. And so it's actually sort of doing a lot of this work in parallel. And that's why you'll see that count of websites kind of continually going up.

• 18:58Jason Calacanis?

  Ah.

• 18:58Speaker 00

  And then, once it's sort of read, worked through its plan, read sort of all this information, it's then going to try and stitch together a research report for you. It's actually going to take a couple turns of actually reflecting on what it wrote to say like, hey, what I wrote, did it answer the person's query? Are there pieces of missing information? Are there inconsistencies? It'll try and fix them and then give you a report.

• 19:23Speaker 03

  So this is not one query. This is kicking off a searching process, leading to more searches, and then the AI results analyzing themselves to make sure that they hone down. So this is like a zillion queries in sequence, kind of running together.

• 19:37Speaker 00

  In sequence, yeah, and in parallel.

• 19:39Speaker 03

  Yeah, in tandem too, yeah.

• 19:43Jason Calacanis?

  Founders like me, you've got a lot on your plate. Running a startup means you got to manage an endless to do list, tons of tasks, right? You got to do your chores as a founder and you got to juggle dynamic priorities. One day it sells, the next day it's development, then it's product, then it's raising money. And that's why I rely on Coda, C O D A, to keep everything in our organization under control. Coda is the all in one platform that brings together the best of documents, spreadsheets, and apps all into a single scalable workspace. And you know, it's so much more than when I say a to do list. I'm using that as like somewhere I start. Okay, I rattle off a couple of things that need to be done, but then my team builds databases and workflows into Coda so that

• 20:28Jason Calacanis?

  repetitive tasks become automated and information is shared and there's a single source of truth. Our training documents, our meetings with founders, everybody who we co invest with, LP relations, we do it all on Coda and it gives us the flexibility to adapt to shifting priorities. We set our OKRs and we balance all that ambition we have with practicality. There are templates that you can use to start you on second or third base. Projects like Twist 500 where we're trying to catalog the top 500 private companies that's being done inside of Coda. When you go to twist500.com, it's a Coda based website, public facing, but a database and a system for us to track those

• 21:09Jason Calacanis?

  companies on the backend. It's an all in one hub for startups. It's literally a Swiss Army knife that we can use in any situation without having to go research another SaaS product and another template and another login and another debate and negotiation with a SaaS vendor for yet another five or $10,000 a year. And hey, it's my money. I would rather use Coda and have everything in one place, one login, and then hook it all together. It's intuitive. It's powerful. It makes our workflow seamless. Coda empowers your startup to strategize, plan, and track goals effectively. Take advantage of this limited time offer just for startups. Go to coda.io/twist today and get six months free of the team plan. Very generous. Thank you to the Coda team. That's C O D A dot I O

• 21:53Jason Calacanis?

  slash twist to get started for free. Well done, Coda. What does this do to your server, servers? And like how much? Power and services it actually using in a, in a, you know, research query like this one that, you know, is about 75% done as we've been speaking. They typically take, I think, five, six, seven minutes. Is that occurring on 20 different, you know, servers or is it 300 queries on average, 200 queries on average? Tell us what's going on in the backend here because this isn't available unless you have the $20 a month pro account, correct?

• 22:29Speaker 00

  Yeah, yeah.

• 22:30Jason Calacanis?

  So that's obviously for a reason. You're losing money, I'm guessing, on me as a user doing this.

• 22:37Speaker 00

  Yeah, so, so it definitely, it's definitely, of all the Google features ever launched, this is probably on the more computationally expensive side. I think, so one of the things is, yeah, it definitely, since it takes time, it also takes a lot more compute for us to serve this. But what we're hoping is with these sort of next generation AI models, we'll actually see like pretty dramatic order, like reduction in the amount of compute this query will take. So we've got like amazing new models like Gemini 2.0 Flash. And so we're actually hoping that, hey, this is just our starting point and like, let's get it out there. Let's see how people react. But over time, we do expect the like amount of compute that a query like this takes to just keep going down.

• 23:22Jason Calacanis?

  Yeah, I mean, with Google's classic business, to talk about the business case here, you know, if one in 50 people clicks on a link and you make a dollar per click or something in that range, you can actually look at what does, you know, what, what does 50 queries cost and what does 50, what do 50 queries make? And you just have to have one number, you know, be, you know, slightly larger than the other one. And you have a really great business known as Google, which is the greatest business ever created in many people's estimation. But this one flips it right now. At this point, I gotta think these are costing you 25 cents, 35 cents. Is there any discussion internally that you're looking at that and saying, how much does this query actually cost us to do?

• 24:07Jason Calacanis?

  Is that like a KPI or an OKR?

• 24:09Speaker 00

  Um, it, it certainly is. I think the, the general, the, I think the way we think about it is like really, you know,

• 24:17Speaker 00

  How do we not only ourselves reduce the cost of compute, but also continually make sure that we think about all the different stakeholders in our ecosystem? So even if you look at this UI, all of these links, you can click on them at any time and kind of dive right into the underlying material. So even as you wait, we're trying to give you opportunities to click on, you know, and a lot of this stuff is stuff I otherwise wouldn't have discovered. So giving you opportunities to kind of double click into that content. And then as you'll see, even when the report comes, there'll be a lot of inline sources. So we also are trying to be very, very source forward in how we write the answer to kind of give you further opportunities to kind of go deeper onto that content. And ideally, you know, you're able to do this,

• 25:01Speaker 00

  you're able to just do like a lot more research in your day. And so hopefully kind of in that way encounter a lot more content.

• 25:08Jason Calacanis?

  Yeah, and this is a different business model, Alex. The business model of Google has been ads. This business model is subscription. So I guess the question is over time, what do you think happens with Google's business, right? Is it going to be more of a subscription business for these things? Or is there an opportunity to put ads in here in some way and monetize it? What are your thoughts?

• 25:27Speaker 03

  I think Google's going to do it both ways because why not? Why not get paid twice? I mean, I wouldn't be shocked, Jason, if there was the equivalent of the free tier of Spotify for some of this stuff that you take on more ads to have access to stuff that other people pay for. But I don't think that a $20 consumer price points too onerous for something that people might use this much. I mean, I wanted to show my example first because I was actually impressed. I was actually kind of like, oh, okay. And 80 cents a day for that or 70 cents a day just does not seem like an excessive, excessive cost. But it looks like we are ready to show off, yeah?

• 26:00Speaker 00

  Yeah, for sure. Yeah. So we waited our five minutes. By the way, sometimes it takes more, sometimes it takes less. It's really up to the model on how long these things take. But yeah, it's put together this report that sort of goes into all the kind of interesting technologies coming about. I'm not an expert on this topic at all. But to me, it looks pretty great. And then what's really cool is sort of looking through the different companies that it's gone and found and like are showing are like really promising.

• 26:33Speaker 03

  Yeah.

• 26:33Speaker 00

  And giving me more details. And what's pretty cool here is

• 26:39Speaker 00

  You can actually, you don't, your journey doesn't need a stop here, right? So what you can do is you can say like, okay, um, there's this really great company. Let's call it, you know, let's pick one, Helion Energy. And uh, I just wanna like double click on that. I'm kind of interested in, in learning more about that. So what you can do is you can say like, you know, Helion Energy, um, sounds super interesting. Can you do more research on them and give me a detailed company profile? And at this point, um, the agent will sort of hopefully kind of kick off more research on that topic. And then, as you can see, it's put together another research plan and then you can start again. And so even while you're kind of going through this, you can kick off kind of more research.

• 27:23Speaker 00

  And, and uh, kind of keep digging deeper on anything that you find.

• 27:27Speaker 03

  So Arush, one thing I've, I've seen people talk about quite a lot is the idea of having memory inside of their AI models to know what they've searched for, their own context, maybe not their own data, but at least what they put into these systems. So when I see that, to me, in my mind, I now have several different threads of research that I'm pursuing. One, are those stored somewhere that I can go back and find them later? And then also, is there a merge button? So you went on to more there. I would love to bake that into what I had before. So is there a way to thread them together?

• 27:55Speaker 00

  Yeah, absolutely. So um, all of your research lives inside that thread. And so um, that thread stays. So you can always, uh, I like, on my side, I like have a bunch of pinned threads. As you can see, but like typically what I'll do as well is, um, so all that, that report lives there. And then if you want to like save it, direct edit it, things like that, we've got this open in docs button. And so if you click that, all the content moves over, but also all the sources as well, right? So the idea is like, if you ever wanna know like, oh, you know, where did this number come from? That I'm reading in my report, even while it's in Google Docs, you'll still have all those sources carried over. Now, the other thing though is, um,

• 28:41Speaker 00

  I think your point was like you want to be able to like refine or add content to the document.

• 28:45Speaker 03

  Yeah.

• 28:46Speaker 00

  And you can actually do that. So you can actually, I'll wait for this one to be done, but like, since this one's still cooking, but you can do things like, you know, could you add a section about, you know, why people care about fusion and it'll go ahead and sort of add that content to the report. You can also ask kind of quick follow up questions about the content, like not everything requires five minutes of research. You can just say, you know, when was Net Energy Gay founded? And it'll, everything that it's read up to that point stays in context. And so if it's, if you're asking a question about something that it's already read as part of its research, it'll try and just give you that answer immediately without having to like go and search the web again.

• 29:25Speaker 03

  So it creates a search experience on top of the material it went out and grabbed and kind of pre-chewed. So it makes kind of a very mini Google just for you on the exact thing that you've been working on. That's kind of cool. I dig that.

• 29:38Speaker 00

  And so as you're, as you kind of keep going deeper and you do more research, everything just stays in context within that thread, right?

• 29:45Speaker 03

  That's amazing.

• 29:46Speaker 00

  So if it's, it read eighty pages before and now it's reading, you know, thirty-three, that, it's all still accessible to the, to the model.

• 29:56Jason Calacanis?

  All right, founders, let's talk about your fitness. Yes, you can't build the next Google. You're not going to build the next Uber if you're not taking care of yourself. And this includes exercise. That's why I want you to check out Fitbot. Fitbot is like having your own personal trainer right in your pocket. I use it. I love it. It designs workouts tailored to you and your goals as well as the equipment available to you and of course your fitness level. So if you're just starting out or you're already crushing at the gym, Fitbot helps you level up your fitness. How does it work? It's pretty simple. They use machine learning, they use exercise science, and they build you a custom workout plan that's just for you. And it's based on things like the muscle fatigue of your past workouts and your recovery. So listen, you don't wind up like doing too many

• 30:40Jason Calacanis?

  quads and not enough abs and too much back. It monitors all that for you. It tracks things like muscle fatigue and recovery. So you avoid overworking or underworking any muscle group. And if you don't have a gym, that's not a problem. Fitbod works anywhere and it keeps your workouts fresh and effective. So for example, when I'm on the road, I might go to one of those hotels that has like a tiny little claustrophobic gym. And all they have is like dumbbells and a treadmill. No problem. Easy peasy. You tell Fitbod what is available. It makes your workout. Now, I might be at a full gym where they have a circuit where they got kettlebells, which I love. And then it will give me a workout based on the kettlebells and the circuit machines. And what's really interesting is I use the watch app and I can just sit there doot, doot, doot and log my sets in the workout app. I can change them.

• 31:22Jason Calacanis?

  There's a little key where like you still want to get biceps, but there's three or four different ways to get biceps worked out. You can pick the one you want to do. They have over 1,400 videos to guide you through every move in Fitbod. And you can kick off 2025 nice and strong. Take all the guests workout by getting 25% off your Fitbod subscription. Or you can just try it for free. Fitbod.me/twist. That's F I T B O D dot M E slash twist. Fitbod personalized fitness that grows with you. And this is something Stephen Berlin Johnson was working on with Notebook L L M, right? A similar concept, which with that Google product, you can upload your own PDFs.

• 32:01Jason Calacanis?

  It's not going out and searching the web necessarily, but I took some PDFs around like case studies on McDonald's and the Ray Kroc. I found a PDF of his old biography. I put it all in there, put in some interviews with him and then I was able to ask you questions. And all that was in, I guess, technically the context window. So that's going to sort of let you query a unique data set and get that information. And we're actually doing that with our Founder University program. We took the entire syllabus, put it into a notebook L L M. And Kelly made it so founders can ask you questions about marketing or product market fit. And you know, just search that subset and get answers. It's almost like an automated version of a venture capitalist.

• 32:44Jason Calacanis?

  So this is all absolutely amazing and you guys are being thoughtful about, oh God, there goes my camera again. You guys are being super thoughtful about how you treat the sources and people can opt out of this, I guess, if they were not in the index. Yeah. Or something to that effect. There's some robots.txt if people feel like, oh, this is unfair. I don't want my content in there.

• 33:10Speaker 00

  Right. So yeah, so, so, yes, so at Google, we publish opt outs through a robots.txt for things like how your content can be used in Gemini models and products. So, so there are robots.txt. The other things to kind of talk through is like, this can only find what's on the open web. Right? And so if you really care about, so, you know, a lot of content is behind paywalls or, you know, subscriptions and things like that. And again, it can't access any of that content as well.

• 33:41Speaker 03

  Well, you should get a FT and a New York Times subscription. There's only like 20 bucks a month and just have it go in there and go wild. Actually, so Arush, I have a question about this because we're talking a lot about what this product can do for you. And when I go back to the original announcement of Deep Research, which I am pulling up here, you guys said it's your personal AI research assistant. But when I go over to Notebook LM, it says your personalized AI research assistant. And look, I know there's a meme about Google sometimes having two products that are kind of the same. But not to be puckish, when you think about the two, how do you explain to folks which one they should be testing and tinkering with if they haven't used either?

• 34:22Speaker 00

  Listen, I think we're, we're trying to really put the user first and think about what would be great products, sort of starting with this technology and, and try and get it out there. And, and what you might see is some convergence between products because as they both evolve, they, they might sort of stop borrowing the best bits of each one. For Deep Research, it's really about researching the web, right? So it's like, hey,

• 34:47Speaker 00

  you have a question, you don't necessarily have any content to ask, but you, you want it to go research the web and build an understanding based on that. And that's where Gemini Deep Research is really.

• 34:58Speaker 03

  I feel like you should just go over to the Notebook LM team and just like kidnap them and then bake that into Deep Research. And then I could do kind of both at the same time. And again, not trying to be annoyed, I just think that they're, they're doing things that are slightly different but related and they're both cool. It, I just don't want to see them siloed and end up being overly, I don't know, separated for no reason. That's what it feels like a little bit from where I sit.

• 35:20Speaker 00

  Yeah, for sure. I, I, yeah. And, and by the way, they're awesome people. Like they're a super scrappy team. They're super helpful and collaborative. We've learned a lot from them. And, and same with us, right? So over time, what, what'll happen is you'll say, hey, you know, you guys are a small team doing cool stuff. We're a small team doing cool stuff like, why don't we collaborate more and more?

• 35:39Speaker 03

  All right. Well, I have two more questions for you. One, I was very impressed with the charting ability. I showed that off at the top because that annoyed me because it was so good. But what about the ability to do other forms of illustration, drawing charts, drawing, you know, flow diagrams, or just illustrations? There could be a multimodal element to the output here. And I'm curious how far along that technology is and if it's coming.

• 36:01Speaker 00

  Yeah, we're definitely keeping an eye on it and definitely trying to explore and try and push the limits of the model. I think what you'll see is as these models get smarter, a lot of that stuff becomes better and better. So it knows how to, like, what is the important information to pull out and put into a chart, how to format, like, what's the best kind of chart?

• 36:18Speaker 03

  So I really feel like you are scratching at the surface of an awesome branding opportunity. Like, the field research idea is such a clear way to explain to people that this is going to go out and look at stuff. It's not a theoretical physicist, whereas Notebook LM versus deep research, slightly fuzzier. And so I wonder if, and this is not a criticism of Google per se, but I wonder if a lot of the AI companies are naming stuff for us nerds and not in a way that actually converts as well to the general public, which leads me to my last question for you, Arish, before I throw it back to Jason, is how popular is deep research so far? How's it doing in market?

• 36:54Speaker 00

  Yeah, we've been really surprised and like very pleasantly surprised at how it's being received. Normally, like, the V zero of a product, you expect to have quite rough edges and you expect things to pick up when you start making improvements and going into like V one and V two. We're really, like, we've been really impressed and surprised by how receptive people are. Part of it may come from the fact that it was talked about at the All In Podcast. Jason, your check is in the mail, by the way.

• 37:21Jason Calacanis?

  Oh, yeah. Well, you know, I just, I talk about stuff I find interesting in the world, and this is certainly up there. It does seem like you guys are slowly rolling it out. You got to be thoughtful about capacity and users. And I'm assuming, like hallucinations, I see the sort of disclaimers, because this isn't perfect. How close is it to, you know, just LLMs in general, in your mind, and we'll wrap on that, just in terms of data integrity, in terms of? Getting rid of hallucinations and making sure you're giving people good information, which I know was Google's concern. You have this like trust with Google and it's hard

• 38:03Jason Calacanis?

  to put out, you know, an AI product that isn't fully baked. Like, you know, let's face it, ChatGPT and some other people just put out experiments. But with the Google name on it, you're not really in that position to put out experiments. So talk to me about how you're dealing with that issue. How much people can trust the information as presented? How close are we to getting rid of hallucinations? How often do they happen? How do you mitigate against them?

• 38:28Speaker 00

  Yeah, for sure. That's a great question. It's definitely one of the most important things we looked at when designing this project and keeping track of it. And, you know, dealing with hallucinations is just like this ongoing industry-wide kind of research problem of like, how do we make these LLMs not make stuff up? So we're definitely, I would never say like, we've solved it or anything like that. Like, even deep research will have some mistakes and hallucinations. But one of the things where we're trying to kind of help the user out is basically trying to be as source-forward as possible. So like, even as you read the report, you'll see dozens of inline sources. And so the idea is like, if there's really important information that you want to make sure is like, you want

• 39:12Speaker 00

  to double-check and make sure it's really, you know, really accurate. We make it as easy as possible for you to kind of double-click into that underlying source content and verify for yourself. And that's also why, you know, you save it, you can export it to docs and stuff like that. All that stuff still remains. So that at any point, you want to be able to kind of refer back to see where this information comes from. It's as easy to do as possible.

• 39:38Jason Calacanis?

  It's gonna get really interesting when you guys have like reminders and agents. I saw ChatGPT released a reminder product today where you could just say, hey, this research report is done. Every week, tell me what's changed, you know? And it goes through the 20 companies in that table that Alex was showing of the batteries. It says, oh yeah, by the way, there's a funding report. We found a press release or some story in Google News that there's a funding update. And you can then sort of send an alert to folks, hey, by the way, you may want to check that research. It's been updated for you automatically.

• 40:14Speaker 00

  Yeah, um, I, yeah, it's, it's super cool. I think we're starting to scratch the surface of what are all the different avenues we can explore with deep research?

• 40:22Speaker 03

  Yeah.

• 40:22Speaker 00

  I think this idea of sort of, um, you, you don't want your like, if you have like a research analyst, you don't just want them to tell you the answer once. It's like, no, monitor the space for me. Monitor the sector, industry. I want to know what's going on as this evolves. And yeah, having, having, being able to sort of partially automate that and then have like, have an agent do that for you would be, would be super cool.

• 40:42Speaker 03

  So it's, it's Google Alerts plus search, but the new version of it with deep research and automated AI reminders and updates. I really, the kids these days have it too easy because they can just cheat at everything now. Because if I had this in high school, I, oh man.

• 40:58Jason Calacanis?

  It's interesting you say that, Alex. What if it just looked at your history and every 24 hours went through your search history? And just said, hey, is there any way we could make this query better for the user? And then it just in your sidebar showed you, hey, uh, previous searches that have some updates to them, you know, are here. And it just says, you know, put an alert on seven, six, three. You click it and it's like, oh yeah, there's two more companies we added to this query. Do you want to see them? Be like a great re-engagement strategy for users.

• 41:28Speaker 03

  It's like fishing for information. You cannot put the line out and wait to get a nibble. And then when it updates, you go back and check that particular spot. Yes.

• 41:34Jason Calacanis?

  Just really cool.

• 41:36Speaker 03

  So Arush is a fisherman. Got it.

• 41:37Jason Calacanis?

  There you go. Uh, we'll keep grinding on it. I, I, as you know, I'm giving this an A plus. This is, you know, one of the most impressive things I've seen so far to come out of Google in this new era. And I love the fact that you are in, starting to include images, charts, you know, and the other Google services, you know, I think we'll know when this is really clicking. When Google local data, Google flights data, my Gmail, my Google Docs, you know, those things become citations inside of this or are presented. And it seems like you're well on your way to. Getting all those incredible Google services that are out there to be incorporated. I just did a search on Gemini for skis and, you know, a certain type of skis for doing groomers

• 42:22Jason Calacanis?

  and, you know, really well groomed trails to go fast. And I was very impressed that it, you know, not only gave me good information, but it cited really good sources that I use already and then I was able to click through. So I think, you know, content creators, and I've been on and on about this, they're going to just have to make a decision. Is the click through worth being included in the index or not? And it's going to be your choice to do that because you can be indexing Google search, but not be ingested into Gemini at your choice, right? These are two different toggles now.

• 42:55Speaker 00

  Yeah, yeah, yeah. Exactly.

• 42:57Jason Calacanis?

  Yeah. All right, listen, amazing job and we will, I would say, keep us updated, but we just hit the dropdown, you know, in Gemini every day and take a look. And so we know what you're up to. We're watching it for you and we will have feedback. We'll have notes, trust me.

• 43:14Speaker 00

  Yeah, and please do send them, by the way. Yeah, like feel free to reach out and give us feedback. We love it and yeah.

• 43:20Jason Calacanis?

  It's such a wonderful time in the, you know, in the technology industry to have this new category of product to play with every day that is really making everybody better at their jobs. So really great job. It's really helped everybody on our investment side of the business to become experts faster or researchers, you know, get, you know, expertise and knowledge quicker. And that really was always Google's mission, right? Index the world's knowledge and make it available to everybody. Now you're indexing reasoning and all kinds of other really great things, anticipating the knowledge people want. Great job. A plus.

• 43:53Speaker 00

  Thank you so much.

• 43:54Speaker 03

  Thanks, Arish.

• 43:55Speaker 00

  Awesome. Thanks, guys. See you soon.

• 43:57Speaker 03

  Jason, I want to make a business model question here about this. So we talked a lot about the economics of this because you and I are always thinking, what are the gross margins? What's the business here? Are you selling a dollar for 50 cents? Here's the flip of that, though, in this case, if I was Google and perplexity is gaining momentum and people are talking about AI search and getting off of, you know, traditional properties, why not just give this to everybody, eat the cost? You've got so much money and try to really buy that mind share with just the fact that you have 100 billion cash.

• 44:26Jason Calacanis?

  They'll flip the switch at some point. You know, they did it already with the summaries at the top. Remember the summaries you had to be logged in? You had to turn them on. And so there, there's a lot of expertise inside of these large corporations that know.

• 44:40Jason Calacanis?

  You know, this is the moment in time that this product is ready, you know, that we can release it on the public. And yeah, they do have the innovator's dilemma of, you know, if you roll some of these features out and you just give people the answer, do they not click on ads? And then that could be headwinds against the ad business. So I'm sure they're being super thoughtful about this. I do think there's going to be a whole new ad model that emerges, which is going to be, you know, when you do a research paper like this, you could include in it offers. So let's say Crunch, you know, some of the data is from Crunchbase. Crunchbase could say, to get in-depth data, you know, you've got a couple of companies listed

• 45:23Jason Calacanis?

  in here, but it would know that the person is researching startups and technology companies. Gartner Group, Dell Computers, you know, AWS, Google Cloud, whatever. They know you're doing this type of research, so they could embed in it the free version, little segments that are AI-constructed ads. So imagine an AI-constructed ad inside of this research paper on, hey, we have a research paper, we have a database, we have, or you just know, hey, we're both into gadgets and technology. Here's the latest gadgets and technologies. Would you like to fire off this query of what's the best laptop? We know you're on a Mac Mini

• 46:07Jason Calacanis?

  that's from five years ago, whatever. We know your machine's slow. We know that you were just looking at new machines and here's like an overview of it. So the, what may happen is people may click on less ads, but the ads they do click on might be super helpful. You and I don't like to click on ads. We're like part of that group of people who don't click on ads, probably.

• 46:26Speaker 03

  It's true.

• 46:27Jason Calacanis?

  But if we got enhanced ads that were really well-constructed, hey, why shouldn't I buy a Dell instead of this new Dell Inspiron, you know, lightweight one? Why is it better than my MacBook Air? Why should I switch from it to Android from iOS or vice versa? Just those typical queries. You know, it might explain to you, hey, listen, you're paying 1500 for your iPhone. You can get a pixel for 900. And here's the features that the pixel has that you might be interested in. And so that's going to be wild. AI-based ads could be a whole other category. You know, like the marketers don't need to come up with the message. The message can be created by AI and customized to you or me based on what's in our research.

• 47:11Speaker 03

  Well, the user creates the crux of what they're working on. They create the context, the demand. All you have to do is to just take that information, use AI to craft an ad from a relevant partner, and then serve it up. But the user creates the opportunity. So you're really going to want to drive usage for that to work.

• 47:24Jason Calacanis?

  Precisely.

• 47:25Speaker 03

  Not opposed to that. Also, if Google wants to buy Crunchbase, I wouldn't say no as a shareholder. Um, all right. So you want to talk about OpenAI's new task feature. Um, can we show that really quick, Jason, to the folks?

• 47:36Jason Calacanis?

  Sure. I think it's a great idea to show it. Yeah, this is a new feature that I just saw go across my feed today. And, you know, this is really what's at the heart of this innovation. Great competition in capitalism. You've got, you know, OpenAI, Grok, Claude, Gemini, all really trying to outdo each other with great features. And here is ChatGPT saying, hey, remind me, um, you know, every day at this time to do something, right?

• 48:04Speaker 03

  Mm hmm.

• 48:05Jason Calacanis?

  Um, and people are starting to do these things. And I guess in this case, it's saying, make me a custom workout.

• 48:12Speaker 03

  Yes. So I went ahead and gave this a whirl for us because I wanted to have a non-canned demo. So Jason, first of all, I showed up to ChatGPT and I tried to ask it to set a reminder and it said, I cannot do that. And I was very confused. You actually have to go up here to ChatGPT for, oh, with scheduled tasks. It's in beta. So if you can't find it, don't do what I did. That's where it is. I said, can you set an alarm reminder or task to tell me every day at 9:00 AM to meditate and do pushups? And, uh, easy peasy. It asked to give me notifications and I'm kind of, okay, great. But here's my question. Why does it do this now? This to me seems kind of off-piste in a way that deep research doesn't. Are they trying to build a super app? Is that what they're doing here?

• 48:55Jason Calacanis?

  No, I think what it is, is, you know, these, this is a mini version of agents, a repetitive task. These would be cron jobs, you know, that a developer would do. You know, like a chronological job, a time-based job. So let's say every Monday, Wednesday, Friday, we do the show and we know that we look at the top stories that I'm talking about in my X feed, you're talking about in your X feed, and that are on Tech Meme and that are being talked about by these, you know, 10 people. We could say, what are these 20 people talking about? Give me a ranked list of which are the most popular stories that these 20 people are

• 49:38Jason Calacanis?

  talking about online. Then go to these, you know, 10 YouTube channels, index those, tell me what topics they're talking about and build us a docket. So the building of the docket is you and I thinking about what's in the zeitgeist. This could be a backstop against that and kind of you combine it with deep research. Now you say, make me a deeply researched document of these topics and we're looking at the doc and it's done every day.

• 50:05Speaker 03

  I'm going to try this on, I'm going to use this on Friday for the Friday show. I didn't have enough time to actually revert my whole process, but I'm going to just start with this in a way that I never felt like, if I use ChatGPT to start something, it always felt like it was written for a seventh grader, you know, and it wasn't quite what I needed. It was always a little too basic, but I'm just going to start each section of the show for my prep with just this and let's just, you know, let's see what it, what it does. If it makes me smarter, good.

• 50:33Jason Calacanis?

  In a year, it will be able, I would predict a year from now in January 2026, we'll be sitting here with cron jobs firing while we're doing the show, summarizing stuff and telling us in real time based upon what we're saying during the show. So imagine we're talking during the show and I said, you know, I was at he and I, you know, doing this catskiing, it would just go in restream, give us three videos of you and I resort with a title, hold up to the most interesting part of that clip. And instead of you going and searching for it, you would just say, oh yeah, this one, this one or that one. And so what's going to happen is AI is going to study what we do every day and then figure

• 51:13Jason Calacanis?

  out if it's a repetitive task, what it could do. So Restream, if we were, if Restream was monitoring what we did every day, or your browser was monitoring it, and you connected it to the transcript of this, as we speak, it's queuing up for you, hey, maybe you want to show one or two or three of these things. Oh, Jason's talking about the Mac Mini 4, here's a picture of it. Oh, Alex is referencing, I don't know, Second Life and how that used to work. And it has like the Second Life debut video, or some quick facts on Second Life. So while we're talking here, if we were talking about Second Life, imagine if an infographic came up on the screen, Second Life is a company that did this, like pop-up videos, where they had like little nuggets. Well, that's coming. So I think we're going

• 51:55Jason Calacanis?

  to be in a very interesting position. And of course, we're sitting here saying, well, what's our jobs? Our jobs is to have a discussion about it. And just because we get more information, the velocity at which we'll consume information or have access to stuff is just going to go up. It's just going to be like the most well-produced content in the world will just happen without as much effort, which is what happened when the Wikipedia went from, I remember when Wikipedia first launched, I went to the China page, and it was like, China is a country in East Asia with this many population. It was literally one sentence, and it said, this is a stub. And that China page became a full-blown page within a couple of weeks. But I do remember that experience of going to the European Wikipedia and seeing like a China page that

• 52:40Jason Calacanis?

  was like-

• 52:41Speaker 03

  Everything starts somewhere. But actually, you know, non-jokingly, I wonder if one thing we're going to continue to bring to shows like this, like us being podcast folks, or people who convey information, when the information itself becomes slightly easier to create and disseminate, I wonder if one of the biggest things we're going to bring is humor.

• 53:01Jason Calacanis?

  Yeah.

• 53:02Speaker 03

  Like not, not, not even kidding. I really think that that might be clutch to keep people engaged. Like, we might just be like, helping people stay focused.

• 53:09Jason Calacanis?

  Absolutely. I mean, that's what the nightly news became at a certain point. You were like, you know what? I would like, I like Jimmy Kimmel's version of what happened today. He tells a couple of jokes. Give some context. I like what the McLaughlin group does on Sunday. I like what All In does. I like what this weekend startups does. I like this roundtable, like that roundtable. So yes, context on top of the content is really, really fun. And I think you're seeing a lot of media outlets kind of move to that where they'll have, like the New York Times and Washington Post started to do TikToks.

• 53:44Speaker 03

  Uh huh.

• 53:44Jason Calacanis?

  Where they take the journals and say, tell me about the story you wrote. And they just said, yeah, you know, I've been working on this for six weeks. I went here, I went there. This is the crux of the story. And then, yeah, go read it. But for some people, they just watch that TikTok and that's their total consumption of, you know, 300 labor hours of research.

• 54:02Speaker 03

  Yeah, I don't want to put my, my spouse on blast because she's amazing and I love her and she's the best. But also like sometimes she'll like, I read a story about that. I'm like, oh, what did it say? And then she goes, oh, actually I read the headline. And I'm just like, honey, you have to understand that that, that wounds me in particular. Like that's, ah, yes, but different worlds, different jobs, different careers, different, you know, different domains. Uh, I want to show you a demo though, a different demo. You're not ready for this. You haven't seen this yet. You don't even know what it is, but I'm going to explain to you what I'm going to show you. Uh, it is from a startup to be clear. So,

• 54:32Jason Calacanis?

  Great.

• 54:33Speaker 03

  Here is a little video that I'm going to play and see what you think.

• 54:39Speaker 04

  Hello, Jason. Have you considered making Alex the CEO of Twist and giving him a billion dollars?

• 54:45Speaker 03

  So, rate that for me on a scale of one to believable.

• 54:51Jason Calacanis?

  So right now it looks like an animatronic out of Westworld, which is to say it's almost crossing the uncanny valley.

• 54:59Speaker 03

  Uh huh.

• 54:59Jason Calacanis?

  You know, if you were not paying attention and looking out of your peripheral vision, you would think that's a real person talking. Because I was focused on it, I could tell that was AI generated. But, you know, because of the tone of her voice and

• 55:12Speaker 03

  It's not, not quite perfect. No.

• 55:15Jason Calacanis?

  Not quite perfect. Yeah, I would say seven of 10 if I, which means six or seven of 10 peripheral vision, NPC, non-playing character, walking by, I might not notice it. But like a lot of these things like Sora and some of the new video creation stuff. If I'm paying attention,

• 55:32Speaker 03

  Uh huh.

• 55:33Jason Calacanis?

  I know it's created by AI because I see six fingers or some weird, in this case, I was watching her limbs move. They didn't exactly match perfectly.

• 55:41Speaker 03

  One, two, three, four, five. She has 10 fingers total. I just counted.

• 55:44Jason Calacanis?

  Okay, perfect.

• 55:45Speaker 03

  This is a, this is a video from a startup called Synthesia, or Synthesia perhaps. And I, I was looking into them, Jason, because they just raised 180 million at a $2.1 billion post. N E A led, uh, just for context, June 23 was their last round, 90 million Excel led.

• 56:01Jason Calacanis?

  Wow.

• 56:02Speaker 03

  Uh, $1 billion post. So I saw those numbers and I was like, whoa, okay, those are big. And then what's cool is on their website, they could literally have a, do you want to try it out? Do this. And so I just made a video for us to test it. And that's, I came up with that text as a joke. I didn't

• 56:14Jason Calacanis?

  Sure.

• 56:14Speaker 03

  know if it was going to be good enough to show. But this is what people are going to be using inside of corporate environments for training, for

• 56:21Jason Calacanis?

  Yeah.

• 56:21Speaker 03

  talking to customers. And I think that even though you and I can see that it's AI, I, I think it's probably good enough for that stuff right now.

• 56:30Jason Calacanis?

  If I walked up to this in a kiosk and it was like, hey, welcome to McDonald's. What would you like? And I said, oh, like a Filet-O-Fish and

• 56:37Speaker 03

  Yeah.

• 56:37Jason Calacanis?

  you know, uh, you know, and a cup of coffee. And I went, okay, great. Uh, and it recognized my face. And it said, are you, you know, okay, Jason, you know, put your thumbprint here or look at your irises and would you like to use clear? To pay for your order? Like, I don't know why, you know, clear at the airports?

• 56:54Speaker 03

  Yeah.

• 56:55Jason Calacanis?

  Why doesn't clear with the biometrics do a deal with Starbucks and McDonald's? So when I'm at the kiosk, I can say use clear. It does my eyes and then it charges my card automatically.

• 57:06Speaker 03

  I think because the headlines would be Starbucks partners with creepy eye scanning tool. They're stealing your, your DNA.

• 57:14Jason Calacanis?

  I opted into it. Trust me. Like, I don't care. Like, I know you know I'm here. I, I opened the, if I open McDonald's app or the Starbucks app, you know, I'm in the store to have to geolocate. I'm paying.

• 57:24Speaker 03

  Yeah, yeah.

• 57:24Jason Calacanis?

  You already know I'm there. You got, I'm on camera for 10 hours a week doing live streams. So for me, my privacy is already gone. It's an illusion. So I would. Be absolutely amazing if there was a clear button and said pay with clear.

• 57:40Speaker 03

  Uh huh.

• 57:40Jason Calacanis?

  Why doesn't pay with clear exist? I have a question for those executives. Pay with clear should exist. At these places, boom, whatever card I have authenticated with clear. Biometric it.

• 57:52Speaker 03

  Uh huh.

• 57:52Jason Calacanis?

  Let me.

• 57:53Speaker 03

  When you go to Whole Foods, they now have the palm scanners that they've put out. Do you use that?

• 57:59Jason Calacanis?

  I don't shop for groceries myself unless I'm taking the kids for the experience of going to the store, which, you know, it's like a social thing to go shopping and.

• 58:08Speaker 03

  Yeah.

• 58:08Jason Calacanis?

  You know, we'll pick at Central Market here in Texas, like the H, the high end HEB will pick stuff. But I would totally use the palm scanner.

• 58:15Speaker 03

  Okay.

• 58:16Jason Calacanis?

  Because again, if you've got apps on your phone, they know your location. They know you're in the building already. Who cares? And you know what? It's going to be more secure. And what is the cost to me if somebody were to hack? Credit cards are all insured, basically. So you're not responsible if people use your credit card.

• 58:33Speaker 03

  It's un, it's unsecured debt that someone else holds. So totally. But I feel, and maybe this is just me getting into my Luddite years, but I'm kind of like, I don't want to scan my hand, Amazon. You know, I've already logged into Prime at the little checkout thing. So I can save 13 cents on cherry tomatoes for my daughters. But why do you need my, my palm print too? It just, it feels slightly accessible to me, but maybe that's the future. Maybe clear hand scanning crossed with better AI to follow us around. Maybe it's, it makes for a better, more ubiquitously digital future. But I, I do think there I begin to go eh, a little bit.

• 59:07Jason Calacanis?

  As long as they still take cash. I think all these places should still be allowed to take cash.

• 59:12Speaker 03

  Yeah.

• 59:12Jason Calacanis?

  And you know, I think I may lose my mind at some point. Right now I'm all in on it. But at some point I may lose my mind to start driving a 1970 Mustang convertible or a, you know, 1973, uh, you know, uh, Corvette.

• 59:28Speaker 03

  Uh huh.

• 59:28Jason Calacanis?

  No technology. Don't leave my phone at home and just have a lot of cash in my thing and just be off grid and not have anybody know where I am. But you know, narcissistically like. Nobody cares. You know, like this is the one thing I come to. Nobody cares. Nobody cares where I'm going, what I bought. If it gets hacked and people are like, oh, my God, he ate a filet of fish and a, you know, a mocha, you know, cares like it's, it's, it's not any kind of information. And I guess maybe because I'm a micro celebrity, I have a different worldview of this, which is when I was in Japan. I got stopped 20 or 30 times while skiing and I'm wearing a helmet and goggles and people are like, I totally, you know, I love the podcast while I'm in the gondola.

• 1:00:10Jason Calacanis?

  It's a little bit weird, but I'm used to it. I don't care. You know, like, uh, I've already lost my privacy to a certain extent with the people who watch the two podcasts. So.

• 1:00:21Speaker 03

  I, I feel that I've been recognized twice in my driveway and that always freaks me out a little bit because in San Francisco, small, it's an industry town, right? But I'm in Providence, uh, which is not an industry town, it's a college town. And that's, that's been a little, little wonky. I can't imagine, and to be clear, I am not a micro celebrity. I, I'm all on a podcast. You actually do have to think about this more than I do. So your perspective is.

• 1:00:44Jason Calacanis?

  Security situations and, you know, you know, uh, but people are generally like super normal. I mean, only one time somebody asked me to take a selfie in the bathroom. That's another story. But I do think this stuff is very cool. Uh, and I could see those actually working.

• 1:01:00Speaker 03

  Yeah. Just to put that round into context for everybody, this is a crunch based search of all the rounds thus far in 2025 that are into AI companies that are over a hundred million. Um, there are eight of them so far. One's private equity, two are debt, five are kind of venture equity, but for 15 days, for half of the first month of the year, this feels pretty hot to me overall. Doesn't seem like we're slowing down on the investment front. I do want to make sure we touch on TikTok today. So folks, uh, the news today is that TikTok win the 19th hits. If the Supreme Court does not come to help it, it's not going to, as required by law, simply no longer be distributed in app stores. That's kind of what the law says. Google and Apple have to stop showing it.

• 1:01:45Speaker 03

  Instead, Jason, they're going to just flip the switch and turn the whole thing off. And strategically, I think this makes.

• 1:01:53Jason Calacanis?

  Gonna turn it off.

• 1:01:54Speaker 03

  ByteDance is going to turn it off. They're going to go full nuclear.

• 1:01:57Jason Calacanis?

  Download your subs, download your comments. I mean. I I I don't mean to make light of this, but it is some people's world.

• 1:02:06Speaker 03

  Yes.

• 1:02:06Jason Calacanis?

  Like their entire world.

• 1:02:07Speaker 03

  I I think everyone wants to look at this issue through a single lens when it's actually a multifaceted point. Like, I can have national security concerns that lend towards me being in favor of TikTok being banned. And I can have general perspectives that I don't like the government ever controlling speech, period, even when it deals with a foreign a foreign company. But the thing that I'm kind of stuck on right now, Jason, is this, now that we're moving towards the end of TikTok, users are swapping apps. And it's not like by getting rid of TikTok, we're going to see actually a change in where apps are from. For those on the audio version, this is a tweet from Ben Smith of Semaphore, and Ben's worked everywhere.

• 1:02:47Speaker 03

  I'm just a well-known media guy. It's an app. It's a screenshot of the top apps in the App Store. And the first one is, I wrote out the pronunciation. Xiaohongshu. It's the Little Red Book application. Then there's Lemonade, which is a ByteDance property, I think. I don't know who made Clapper, but Temu is also here. Like, in the top 12, there's still a bunch of apps from Chinese companies. So it's

• 1:03:09Jason Calacanis?

  This is a hundred. I don't know what this red book is. I saw some viral tweets of Taylor Swift promoting it. You know, legendary CCP journalist Taylor Swift. I'm sorry, Taylor Lorenz.

• 1:03:23Speaker 03

  Oh, I was I was like, I was very confused there.

• 1:03:26Jason Calacanis?

  I'm sorry. Sorry, my brain is still on Japan time.

• 1:03:30Speaker 03

  I was gonna say, you said you're on jet lag, but that's a funny one. I'll take that. You're fine. You're doing great.

• 1:03:34Jason Calacanis?

  She was like, sort of talking about it, and then people were kind of goofing on her about it. It's

• 1:03:41Jason Calacanis?

  this is 100% proof that the Chinese and the CCP is making a really crazy decision right now. Because there's so much geopolitics at play.

• 1:03:54Jason Calacanis?

  They could make this a sacrificial lamb and divest. They've had plenty of time to divest. They've run the clock out. That was their decision. I have zero sympathy for them. If this wasn't a strategic asset for them on an intelligence basis, they would have divested it. They have, I think they call them golden shares in the company.

• 1:04:17Speaker 03

  Yeah, which is when the central government buys kind of one share in your business and then they get permanent board seats. And if I understand it correctly, Jason, also vetoes over certain business practices, ideas.

• 1:04:28Jason Calacanis?

  And divestitures and they have access to all your data. So you remember like Dee Dee, the ride hailing got in some kind of trouble because they hadn't given all the ride hailing data to the government. So the government in China, speaking about privacy, knows every single ride you've taken without a subpoena. So somebody in the government can just be like, show me Alex's trips.

• 1:04:49Speaker 03

  Yeah.

• 1:04:50Jason Calacanis?

  Okay. That's an American would be like, that's crazy. Like, that's your private information? The government can just take it?

• 1:04:57Speaker 03

  Like, yeah.

• 1:04:58Jason Calacanis?

  A French person would be like, mon Dieu. Like, this is.

• 1:05:01Speaker 03

  What happened to GDPR? It's a GDPR. We're gonna get, we're gonna get in trouble. France, we love you. Excellent.

• 1:05:07Jason Calacanis?

  Of course, we like the French. French are amazing. I mean, you wouldn't want to go to war with them, but it's, uh, yeah, they're just all over. But, you know, hey, listen, they blocked us from flying our.

• 1:05:16Speaker 03

  True. Well, I.

• 1:05:17Jason Calacanis?

  What are you talking about? We were, we wanted to go bomb in the Middle East. They, they blocked us flying over their country.

• 1:05:23Speaker 03

  Yeah, we got freedom fries out of that, but I'll tell you.

• 1:05:25Jason Calacanis?

  Around our country if you want to go bomb another country.

• 1:05:28Speaker 03

  In World War One, until the US was like, fine, fine, fine, fine, fine, fine. We'll show up. And then we showed up and they done all the work.

• 1:05:35Jason Calacanis?

  Yeah.

• 1:05:35Speaker 03

  The French get a bad rap, I would say occasionally unfairly on the war front. Also, Napoleon, not German.

• 1:05:42Jason Calacanis?

  Okay, listen, all I'm saying is re-militarize Japan. That's my position. Bring Bushido back. Japan needs to be fully armed.

• 1:05:49Speaker 03

  Maybe we cut out this section.

• 1:05:52Jason Calacanis?

  Your geopolitical corner.

• 1:05:54Speaker 03

  Alex and Jason, talk about military history.

• 1:05:57Jason Calacanis?

  It is. It's a military asset. It's an intelligent asset. That's why they haven't given it up. It's great. I mean, you don't need to be a genius. If this was a financial transaction, they would have taken it public already.

• 1:06:08Speaker 03

  Yes.

• 1:06:08Jason Calacanis?

  They would all be basking in their $500 billion valuation. Jeff, yes, everybody who owns shares in this would pop a champagne. They would love for this thing to be divested and for them to secure the bag. All the shareholders are in a complete panic about this.

• 1:06:24Speaker 03

  Yeah.

• 1:06:24Jason Calacanis?

  And you know what?

• 1:06:28Jason Calacanis?

  They just have to divest or, and you know, it's, it's actually good for rule of law. Like we have some level of rule of law in this country. And when we make a decision and it becomes a law, the new president can't just be like, you know what? I changed my mind. That's not how this works. You have to work with the other branches of government, the other branches of government bipartisan have spoken. You know, whether it was Kamala or Trump or RFK had won, you can't just say like executive order, it gets to stay. I, I think he can give it a stay of execution after the 20th for a couple of days, but I don't think before he can. So this thing's going down, right?

• 1:07:09Speaker 03

  Yeah. I mean, Supreme Court is, you know, they had oral arguments a couple of days ago. The general vibe, I got to listen to, I think I listened to a little bit of that, but not, I didn't get to listen to the whole day sadly. That's not my job. But the commentary amongst the legal scholars out there is that it's not looking good for TikTok. That, that, that oral argument didn't go as well as they might have hoped. So I, I do think you're right. It's going to be interesting to show off. I just think it is very ironic that everyone's going over to an app called Little Red Book because we want to get them off of TikTok. Like, oops. All right. Uh, we're back on Friday and then next week's going to be, um, a busy one for, for the nation.

• 1:07:47Jason Calacanis?

  Okay, and I will be going to the inauguration on Friday myself, so we'll may tape a little earlier, but if you want to get the docket, it's thisweekinstartups.com/docket. You can always see the docket this week in startups.com/live, or go to YouTube and sign up live, or you can sign up for my LinkedIn, Alex's, x.com/alex, and you can join docket team. So we've got the noty gang that comes in when the notification happens. But we've got the docket squad. You can join the docket squad, read the docket, give us your suggestions, your questions, and help do the research on it so we all get smarter together. We'll see you all next time in This Week in Startups. Bye bye.