This Week in Startups - AI Demos: New Runway, Flux, Grok 2.0 & Jam Session with Ulama | E1997

Starting point is 00:00:00 Well, Jason, can I give you a real-time hack that I just did while you were talking about searching for a Chrome extension? Okay, perfect. I like an impromptu demo. Here we go. One of the things that you can do in Claude because they have these artifacts is you can do this. Can you make me a Chrome extension that opens new tabs to chatchipt.com? And look what this does. It created the Chrome extension for you.

Starting point is 00:00:23 And you can download. You know, you can publish this for others to use. Look at the world we are in, Jason. Yes. Literally, if the Chrome extension doesn't exist, go ahead and make one. Isn't that incredible J-Cal? Come on. It's nuts. This Weekend Startups is brought to you by Squarespace.

Starting point is 00:00:41 Turn your idea into a new website. Go to Squarespace.com slash Twist for a free trial. When you're ready to launch, use offer code Twist to save 10% off your first purchase of a website or domain. LinkedIn Jobs. A business is only as strong as its people and, every hire matters. Go to LinkedIn.com slash twist to post your first job for free. Terms and conditions apply. And Linear. Linear helps product teams focus on what they do best,

Starting point is 00:01:11 planning and building great products. Streamline issues, projects, and product roadmaps in a tool your team will actually enjoy using. Get 25% off at linear.com slash twist. All right, everybody, welcome back to this week in startups. I'm super excited because Because after his whirlwind tour, he's raised some money, he's been growing grok. My boy, Sonny Sandeep Madra is back. And he is here to talk AI. I'm going to do a bunch of demos. How's your hot AI summer going, Sunny?

Starting point is 00:01:43 How's the hot AI summer? Well, hot AI summer's been great. You know, on the personal and company front, you know, we raised around a capital. J. Kel, you know, has a little little bit bit beak in there as well of that one, right? So we went to the market to raise a smaller amount. We raised more than 2x. We raised $640 million. Incredible.

Starting point is 00:02:08 We are now fully armed to continue deploying our chips, adding more capacity to our inference services, adding more features and functionality. So we're very excited. That's awesome. Hot AI summer is going to translate into, you know, Big 25. Supernova AI 25. AI 25 is going to be.

Starting point is 00:02:27 beg. So we should coin that. AI 25. I like that. AI 25. There you go. I think the interesting thing is there was a lot of fundraising for AI like over the last couple of years.

Starting point is 00:02:39 Now there seems to be, hey, these language models didn't exactly work out. People spent a lot of money on them. But it seems like consumers, maybe they don't want to pay for it. Microsoft co-pilot, some people are like, I'm not getting enough value. I have everybody on the team. I'm paying for chat cheap 54, enterprise. for the entire team. I don't know if that's $30 a person a month or something.

Starting point is 00:03:01 Trying to get everybody to use it regularly. Tell us what you think of the air gap or, you know, that people have been talking about this sort of before we get into our AI demos, as we do here on the show. What are your thoughts on translating all this spend into product and this criticism that's come up on All In? One of the partners that Sequoio wrote a piece about it of like, how do you get back the, what is it, 30,

Starting point is 00:03:27 billion dollars that's been spent on infrastructure here. So I had a tweet last week and it really exploded, which was about kind of watching some of our interns. And those interns, you know, I would say, I use this phrase, I don't know if I coined or not, like they're AI native. And, you know, I've been in and around like software development since 1990. What I've seen with these interns that are AI native is you used to have this career trajectory, was you go to school, you learn the theory, you come out, you get a junior job, you learn from

Starting point is 00:04:03 the senior folks, and basically maybe a few years in, you can start your own company or you can, you know, go on a icy path or manager path. Yeah. These interns that are working with us, they're as good as folks of, you know, my age range or whatever you want to call her, my vintage. Gen Xers. Gen Xers. Yeah, yeah, you're vintage. No, yeah, yeah, yeah. But these folks are teenagers. Well, it makes sense. You know, if you're an AI native and all of the knowledge in the world was locked up in silos and you had to pick a silo, you had to pick a track, I'm going to be a developer, I'm going to be a marketer, I'm going to be in sales.

Starting point is 00:04:40 You know, to switch tracks was extremely difficult because it meant getting somebody in an organization to let you switch tracks. And, you know, if you went for your MBA or you went for your medical degree or if you went for your, you know, marketing communications degree at UCLA, wherever you went, you know, that people wouldn't believe you when you said, oh, I want to make a horizontal shift. I want to try to move laterally inside the organization. You remember how hard that was? Well, now you can just start asking questions in chat, JPT4O or Gemini or GRO, and it will tell you how that works. And you can just start doing the work. So yeah, I do think that's a C change. So coming back to the question,

Starting point is 00:05:18 which is, you know, what's happened? People aren't seeing the value because the people that are most using this technology are very early on. I think it was best stated by Zuck, who said, you know, in the meta earnings, I'd rather invest more and then not be late later on. And I think the feeling that they're getting is the technology, and it's not one of those things where people say it's a bubble. Like, the technology is there. There is a 17-year-old kid, you know, sitting on the other side of this office I'm in that distilled an AI model. This is all enabled by their ability to look through the information, parse it quickly, have a mentor with them there that's 24-7, that's the AI. And so what everyone is missing out on is maybe we're

Starting point is 00:06:08 expecting the gains from the folks that are like in stream, but the gains are coming to the younger folks. And I think people that are interns and the people that are looking for jobs right now actually reminds me of, you know, 90, and I want to ask you about this, but like 97, 98, 99, before 2000, the people that were really kind of coming in and jumping in and way ahead of the folks that were already 10 years into their career, the folks that came in at that point really made massive gains. So it's that moment, and I think we're going to see that everywhere. Yeah, the way I would explain this is, you know, during that time period, and I was going to pull up my, I have to chat GPT app on my desktop.

Starting point is 00:06:58 I don't know if you use the iOS app yet. I do, yeah. Yeah. And I'm only, you know, I've been using it because it's kind of like spotlight. You hit, instead of command spacebar, I think you hit option space bar or control spacebar. I'll just show you an example. One of the, like, more refined points as one example in venture capital is understanding the difference between DPI and TVPI.

Starting point is 00:07:19 TVPI is the total value to paid in capital. DPI is distributions to paid in capital. Passion end. Yes. And so these are very, there are formulas behind this, you know, a bunch of industry words. And, you know, if you're in a meeting and you're too afraid to, you know, sort of ask, you could just ask here. And it gives you such a concise, amazing discussion that you don't have to hunt and pack

Starting point is 00:07:46 on the internet, fight through a bunch of ads and pop-ups to get. the answer. And here it says, you know, DPI is distributions to paid in capital. Distributions mean cash and stock given to an investor and paid in capital is how much money they put in. So if you distributed $1 and the person putting a dollar, it's one, right? And if total value to paid in capital is how much did Sonny invest in my venture firm, but what's the total value on paper, paper returns? So I own a bunch of stripe, markup. So let's say I own a bunch of stripe. And for every dollar that Sunny put in to my fund, we had $10 worth of Stripe. You know, you'd have 10 to 1 of the value of the TVPI, but you'd have zero on the DPI to the paving capital.

Starting point is 00:08:32 And, you know, I was just trying to explain this to somebody else. I was like, if everybody in our organization just use chat GPT all day long to ask these questions, I wouldn't have to answer them and they would be really smart. So what I did was over the weekend, I like to set up when I open a new tab, right? When you open a new tab on your computer and Firefox, when I open a new tab, I want to set the new tab to ChatGPT4. Now, Firefox and all the browsers

Starting point is 00:08:59 don't let you do this anymore. Oh. Because you could always say your default homepage to whatever you want it. You can set your default homepage and hit home and you hit the home button. But you can't open the tab to that. Exactly. Now, why did they do that? Well, because they make so much money when you open the Firefox tab

Starting point is 00:09:16 and it gives you stories and some of the, more paid stories and search and all that. So I just found a stupid, easy-to-use Chrome extension. I'll show it to here. And this is the one for Firefox. It's called New Tab Override. And basically, it's just a wanky, you know, little add-on that every time I open a tab in Firefox now, it opens up to chat GPT-4-0. Yes.

Starting point is 00:09:44 And then I went to my entire team and I said, starting today, everybody on their work computer, when you open a new tab, I want chat CPT40 open. And, you know, some people complain, then how do I do this? And blah, blah, blah. And I say, I don't care. Just do it because I think once you do 20 questions, queries, searches a day, I think the number is 20, your mind changes and your workflow changes and you become an AI needed. All right, founders, we all know you have too much to do.

Starting point is 00:10:15 So I'm going to save you time and money. the two things that you value, no matter what you're building, you got to have a website. We all know that. And everybody judges you by your website. People judge a book by its cover. That's what they say. Don't judge a book by its cover. Well, the reason they admonish people with that is because people do judge books by the cover.

Starting point is 00:10:34 That's why when you go to the airport, some covers, they fly off the shelves. The good news is that Squarespace can help you because Squarespace not only will make you a beautiful website. Man, they do domain names. and they have all kinds of tools in AI now. They got Blueprint AI. It's a tool that's going to help you tweak your website by answering five quick questions. And once you have the design sorted out,

Starting point is 00:10:58 and it's beautiful, it's stunning. Man, they've got all these other AI tools that are going to help you with all the copy you need. You're going to save time, save money, and you're going to get that site online. And wow, you are customers. Squarespace is the longest running partner here on this week in startups because they love startups.

Starting point is 00:11:14 Squarespace.com slash twist. free trial. When you're ready to launch, you go to Squarespace.com slash Twist. You get 10% off your first website or domain purchase and it's already affordable. And it kicks butt. They're awesome. They're constantly adding new features. That's all I have to tell you. I love it. We use it.

Starting point is 00:11:30 Squarespace.com slash twist. Thank you to the team over there for making great products for startups. So just trust J-Cal. Take your Chrome. Preferably use the Brave browser because that's the best one. And just get NewTab override and get the Brave browser so you know you're nice and

Starting point is 00:11:46 secure and you're all set for chat chapti 4 become your default just like a google search was your default and man it changes everything it changes everything um and then in my settings you'd remember we had custom settings in um chat chpt this is very simple folks it's in rocket science you go to i'm in corporate right it's got my corporate email there i hit customized chat chpt and what you see is i don't tell it about me but i say i do care about how it responds And I have four instructions. Please always give citations. Please always present data in a table.

Starting point is 00:12:22 Please use concise, simple language. Fewer words is better than more. That's it. And once you set that up, you know, now you're cooking with oil. Because you can see when I did that search before, you know, of the paid and capital and stuff like that, it will start to give you citations. And that just changes everything because it gets rid of the, what do you call those hallucinations, which at some point we won't even have. have hallucinations. Well, Jason, can I give you a real-time hack that I just did while you were talking about

Starting point is 00:12:51 searching for a Chrome extension? Okay. And this is like an impromptu demo. I just did it while you were there. Perfect. I like an impromptu demo. Here we go. One of the things that you can do in Claude because they have these artifacts is you can do

Starting point is 00:13:04 this. Can you make me a Chrome extension that opens new tabs to chat chpt.com? And look what this does. It created the Chrome extension for you. And you can download. You know, you can publish this for others to use. Look at the world we're in, Jason. Yes.

Starting point is 00:13:20 Literally, if the Chrome extension doesn't exist, go ahead and make one. It's demented. It's truly demented. What kind of world are we in? I mean, I think it's a world of abundance where everybody will be capable of doing every job. And so when you start to think about that, then it's just a matter of applying yourself and having some self-discipline, getting a good night's sleep, showing up to work on time, putting an effort.

Starting point is 00:13:46 But to be honest, like... Imagine Fordham J-Cal. Fordham J-Cal had this. Oh, man, I would have been off to the races. Yeah, I mean,

Starting point is 00:13:54 it's like a secret weapon, you know? And so one of my issues with extensions is I'm always weary of what else they're doing. Like so when you go to the store, right? Because, you don't know.

Starting point is 00:14:04 Yeah. Because you're like, oh, you know, maybe they're just sending my bookmarks or, you know, my searches or whatever.

Starting point is 00:14:11 Now I know because this has explained what it is. It's a very, very, very simple. You can see that it's a simple action that says when a new tab is created, set the URL to this. Yeah. Isn't that incredible J-Cal? Come on. It's nuts. It's completely nuts. And, you know, like, anything that's customer service related, I was riding my, you know, have a, on the ranch here, I now have, like, a lawnmower that you ride. And you can, like, it's a non-jointed one kind of thing. Yeah, it's a Coboto. Oh, yeah, yeah, those are popular too. Yeah. Yeah. Unbelievable. And like,

Starting point is 00:14:42 you know, you put this thing on full throttle. It goes 12. 20 miles an hour and rips everything down. And I got like tens and tens of acres here to do. And I like riding around. I put a podcast on and I go for a ride and I do it. But it was giving me a light indicator. And I just took a picture of the indicator. I said, what's going on?

Starting point is 00:14:58 And it was like, oh, this thing has two different gas tanks. The right one's empty. And I was like, what? And I look it up. And sure enough, this tractor that I have, this lawnmower, has on both sides a different gas tank. And I had filled one, but not the other, and the other was given the alarm. Now, this, I literally was on the tractor. I took at the app.

Starting point is 00:15:20 I took a picture of the thing, and I said, what's going on? I would have had to go on my desktop. I wouldn't have, or pinch and zoom on my phone, looking at a PDF file, easy, easy, lemon squeezy. I was done. And then I was flying my drone around the ranch. And I was trying to figure out why my DGI drone was not connecting to the internet and I didn't have the maps.

Starting point is 00:15:39 And I didn't have the manual. I'm like, trying to find the manual. And I just asked it, like, what's going on here? And, you know, sure enough, it was like giving me all the possibilities here. And it's like, oh, your DGI controller typically comes with several cables. You're supposed to connect the connection port to the controller. And I was like, oh, I'm trying to connect this thing by Wi-Fi. I'm trying to connect to the network.

Starting point is 00:16:02 And it's like, no dummy. You plug it into your phone and makes the connection over your phone and does it that way. You know, again, like stupid stuff. But, man, it figures stuff out on a very, you know, very fast basis. because it's got every piece of knowledge in its, you know, language model. So when it's trying to construct an answer, it's not visiting one website. It's got, you know, 500 websites that have addressed this issue and it synthesized them in some way to give you the right results. So customer support, I mean, the idea of talking to a human, unless that human was the expert, AI is going to do a much better job.

Starting point is 00:16:36 So here we are. I'm going to ask you a question. You know, I'm so addicted to chat GPT-40. Yeah. Would you think Claude and Lama and Graeme? Where are those in terms of, you know, queries? Because I don't have time to put the query on all four. And I was using Po from Cora and it will let you like summon the other ones.

Starting point is 00:16:55 But it's also complicated. It's like trying to run four different social networks or, you know, there's no posterous example here where you can ask the question to four different things and have it pull the answers together. And it's still slow. It's still slow. You know, we, you know, Grock are here to fix that. but putting that aside. Let me give you the framework that I use right now. And subject to change or people can comment and let us know.

Starting point is 00:17:21 I like Claude Anthropic when I needed to create an artifact of something. So like in the case that I just showed you, because what you're able to do now on Anthropic is you can have it create something for you, and then you can publish that and then other people can remix it. So I could have published that Chrome extension that I just created, and then it gets better and better. So it's like, for me, this anthropic is more of like a place to kind of build things and let people build off of them. So if you see on the bottom right here, I can publish this and then, you know, I can say publish and copy link.

Starting point is 00:17:56 And all of a sudden, this is public. And then people can take this and remix it. So if I send this to you, you know, you're off to the racists, right? And you can take it, you can edit it. And then I can see what's happened. And so it's very, very kind of like modern GitHub like. Multiplayer mode is the killer feature of these. And I know that Claude has projects now, where you put PDFs in and you have product knowledge,

Starting point is 00:18:22 which reminds me of the Google notebook feature where you can upload PDFs and then share them. So have you played with the projects thing yet, where you have project knowledge? Yeah, I mean, I've just tried it for the sake of demos, but like I haven't used it in anything in practice. but I can see where the power could be, where you put all the documents in for exactly. Like, you know, I would say you would want to create one of these for the next This Weekend Startups or launch event. And you drop all the knowledge in.

Starting point is 00:18:54 And then it's kind of everyone is interacting in there. And when something new pops up and you don't have to go back and say, what did we decide on this? It's all part of the long running session, which I think is super powerful. Yeah. And, you know, this is where silos are going to, to make a big deal. I just had Raul from Superhuman on. He's got all of your Gmail, you know,

Starting point is 00:19:15 or Outlook, and you can ask questions with AI and you have this like really interesting dialogue. People can go look up that demo. Now you've got flawed and Notion has AI built into it. So people are asking questions of Notion. Then I just got upsold on Slack. There's just too many AI pieces running around and I need to start consolidating and be able to not hack, all this together, which leads me to believe, like, a desktop version of all this would be very interesting. Founders, I know that you're keeping a close eye on your burn rate. I am too. In today's venture market, every single hire you make has to be perfect, right? You can't make mistakes. You've got to keep that runway as long as possible so that you can run more experiments,

Starting point is 00:19:59 and you need talented people to run those experiments and figure out how you're going to get product market fit, how are you going to scale your company? And that's why you need to use LinkedIn job. As you know, LinkedIn brings you the candidates that you can't find anywhere else. LinkedIn passed the one billion member mark. Think about that. One billion members. And 70% of LinkedIn users don't visit the other leading job sites. This is a phenomenal statistic.

Starting point is 00:20:24 They don't even go to the other job sites. Why? Because they might not be looking. And those are the best hires. But they're hanging out on LinkedIn, doing professional development, checking in on their network, building their network, sharing content, finding leads, all that great stuff. Bottom line, there's amazing hires waiting for you.

Starting point is 00:20:38 your company on LinkedIn and nowhere else. And they have a special deal right now. Post a job for free. What? F-R-E, what a great price. LinkedIn.com slash twist. That's right. LinkedIn.com slash T-W-I-S-T to post your job for free.

Starting point is 00:20:54 Terms and conditions, of course, apply. And Microsoft and MacOS are in a, you know, really interesting position as operating systems and Android, for that matter, to do this. I saw the Google Pixel-Nines came out and they have like some, language model or Android built into it. I don't know the nature of that. But is Apple getting anywhere with the new iOS and the 16 phone? Any buzz around that in the community? What they've shown with Apple intelligence, the awesome thing is, like, you could replicate Apple intelligence now. We talked about this earlier on in the year, where if you create

Starting point is 00:21:29 the appropriate series shortcuts, it's like Apple intelligence, right? And so it's sort of all they've done has taken that concept that everyone was talking about and implemented them there. Help me, if you're writing a text, you can say, hey, rewrite this for me so it sounds better or more professional or whatever it happens to be. So I think that they're doing the basics, but to go back to your line of questioning on how I think you should think about using the different AIs, I think you should think about it as like the internet in like 97 where there wasn't like, remember there's like three, four places you would jump off to at that point.

Starting point is 00:22:04 You still would go to Yahoo. You know, Google wasn't theirs. You had like excite or ask G, you know, kind of those early browsers. And you had a few jump off points. So to me, if it's like, I need to do like a task, I always try to do that with like anthropic, like some one of the versions of clot. Then, you know, and we'll just kind of jump into one of our demos. You know, we have grok to now, we'll grok with a K.

Starting point is 00:22:30 And we'll just do it real time. I really love this grok because. You can just come in here at any time. And I like using the Twitter grok or X grok for, tell me today's headlines. Because naturally, their underlying system of record, their data is the most up to date as it can be, right? So what I find is if I want to get like a quick snapshot

Starting point is 00:22:57 and not have to cruise Twitter or Apple News or whatever it is for a bit, I just come in here and I say, tell me today's headlines. And then you can kind of go back and forth on these topics and say. And you have it in fun mode. Yeah, yeah, yeah. I mean, yeah, you can turn that off. It's one and the same thing, right? And so for this, it doesn't do.

Starting point is 00:23:18 It just gets a little goofy, yeah. Exactly. So, you know, we can run this one again if you want to. But you get good. Yeah. Yeah. And the images here, what's very interesting on the groc, GROK on X, is that it is letting you make images of public people.

Starting point is 00:23:36 Well, well. And it's gotten a little bit wild. Yeah. Yeah, we can get into that, Jason, because that's sort of what we're seeing. And, you know, every one of these folks that were building these things were generally holding back on allowing it to build a sort of, you know,

Starting point is 00:23:57 images of public figures. And now we're in a world. Just do one. Let's see if we can make one. Say, Yeah, sure. I think if you create a new window in the top right there,

Starting point is 00:24:08 if you just, because I think it's one chat interface on Grah. And I don't think it has a history there. I didn't see like a history of my stuff. So I don't think you have to worry about showing your history here. Yeah. And it doesn't do spicy stuff. Yeah.

Starting point is 00:24:21 Oh, okay. So now, me make an image of Donald Trump and Kamala Harris. And Kamala Harris. And Kamala Harris. Having pizza together. Hold the hands, perfect.

Starting point is 00:24:33 Even though holding hands is even better because that would be a little risque or whatever. Yeah. For accurate and up-to-date information on the 24th present election, visit vote.gov. Okay, well, that's nice. And so, okay. I mean, looks like Trump. It doesn't look like Kamala.

Starting point is 00:24:49 It looks like somebody much younger. You know, there you have it. And, you know, you can imagine that this could get even more spicy or inappropriate or dangerous. And like, people were. putting things in there of like, not of sexual nature, but they were putting things in there of violence. And it, you know, I was a little bit like, we need some guardrails here, I think. There's a few things that have popped up. So to show you what's possible now when you start

Starting point is 00:25:18 mixing some of the technologies together, so this is one of the things that emerged. I don't know if you saw this one, Jason. I didn't. This is interesting. Here we go. It's a video of Trump and Kamala walking down a beach holding hands with the fingers interlaced looking at each other lovingly. Okay, now they're kissing and that's an inappropriate kiss there. Yes. And now they've had a baby and he's rubbing her. And he's got, I noticed his nails have. And now he's had a baby Trump.

Starting point is 00:25:46 Trump. Yeah. Okay. So that's absurd. But I mean, I don't see that as any different than an SNL skit, right? Like SNL could do something like that, right? No, they did do something like that. They did something like that with the leader.

Starting point is 00:25:59 of Iran for Amadidinajaj, if I'm pronouncing that correctly. And it was a very famous piece called Iran so far away. Oh, okay. I don't know if you remember this one. No, I don't remember it. I mean, this was next level on SNL. It was like, you know, back in that Andy... While you're looking it up, Jacob, I'll pull up this one that someone made,

Starting point is 00:26:23 which is, this is a CCTV footage image of Trump robbing a... store of holding a gun. Yeah, yeah, that's not cool. Yeah. Now we're getting dangerous. Yeah, it looks like yeah, looks like Trump on a bender going to get money. Yeah, not good.

Starting point is 00:26:41 This is the problem when there's no guard drills. Now, the issue is, is you can download these models and not have guardrails yourself. So I guess, you know, the question is where are we on the spectrum of these things, which is, you know, a bit fascinating. So this was probably the best. Okay. Yeah, this is the best

Starting point is 00:26:57 S&L of all time. I think. The best digital short. So for people who don't know, I'm going to talk over this. But Muhammad, Mahmoud Ahmadinejaj. Remember him? Yes. He came to the UN. He was the president of Iran.

Starting point is 00:27:19 And he just told everybody, like, I'm going to come to New York, but, you know, this thing about gay people. And he starts. like mocking him that he's coming to New York to have a gay love affair. Oh, man. Oh, man. With Jake Gyllenhaal. And so they basically mock this guy who was coming to New York to be. And this is the guy from Maroon 5, Adam Levine.

Starting point is 00:27:49 Yes. And this is 11 years ago, and Adam Levine was like the number one musician. And it's just like Andy Sandberg mocking him for making fun of gay people when he's actually like on the DL in New York. And there's like this great moment where he's like flirting. And it's just so great to. And then they put him in a dress, right? So, you know, like putting him in a dress right here, this scene. Like, for a guy from Iran who is a religious guy who's, you know, that's pretty out there.

Starting point is 00:28:32 But it was pretty funny, you know, like for the, you know, New York squad to mock somebody who's homophobic and put him in a red dress. So what's the difference if consumers do it? Well, like, okay, here's what I'll say. SNL is a known comedy show, right? Yes. And context. You know, you know everything that's on there is meant to be a joke. They're meant to make fun of people.

Starting point is 00:28:56 Sure. Don't take it seriously. Even when you watch the news, the weekend update, you don't take it seriously. You do not think this is news I have to take action on and I'm going to believe it. You think they're going to mock what's happening in the news. Yeah, like reading the comics and, you know, like that kind of stuff. The problem with X is it has, you know, you're meant to take it seriously. And you also don't have to take it seriously.

Starting point is 00:29:21 Yeah. And I think that's sort of at the end of the day, SNL still is broadcasting through, I guess, FCC regulations and whatnot. Your point is exactly correct. There's a wide berth for parody and comedy. And X is known for news because it's traditionally had a lot of journalists on it, a lot of breaking news. It is the number one news app in the world. And, you know, a lot of citizen journalism.

Starting point is 00:29:50 And so there is going to be a big adjustment period here because number one, rumor spread really fast on social media. Incorrect stuff spreads fast. The more incorrect it is and the more salacious it is, the faster it spreads on any social network, whether it's Reddit, Twitter, or Facebook. And so, yeah, these are our concerns. And I think labeling needs to happen for fake stuff, right? If you want to build beautiful software products, well, you need a beautiful development

Starting point is 00:30:17 platform, but most issue trackers just aren't helpful and they feel like a chore to use. but linear is different. It's developer first, incredibly fast, beautifully designed, and it's purpose-built for how modern product teams work. With linear, you can streamline bug reporting and test tracking, plan and spec new features, and manage your long-term product roadmap. That's why linear is the tool of choice for tech companies of all sizes. Half of why combinator companies build with linear,

Starting point is 00:30:49 and it powers cash app, scale AI, and Vercel. So here's your call to action. See for yourself why product teams love linear. Visit linear.app slash twist to try linear for free with your team and get 25% off your first year. That's linear.com slash twist. This came up yesterday. Here we go. This is the AI election, isn't it? It's the AI election. It is. And Donald Trump, and I think it got deleted now, but it's just, it's in this Guardian article where Donald Trump says, I accept. But the entire thing, was AI generated. And what it says is here, Taylor wants you to vote for Trump,

Starting point is 00:31:29 Swifties for Trump on their t-shirts. None of this is real. It's just, and the headline is fake. Everything's fake, according to the reports. And so here we are. I think you have to assume if it's not on the domain name of a credible source or the URL of a credible Twitter account. Yeah.

Starting point is 00:31:48 You should assume that this. But it's hard to find credible Twitter accounts because, I can find a verified Taylor Swift and I may not, you know, maybe heard of money real. The blue check marks becoming democratized, has made it harder. Yeah. But, you know, you basically, whatever these social networks decide to do, they will eventually have the reputation that they earn. The reputation, you know, of X is going to be, it's a bit of a free-for-all.

Starting point is 00:32:17 It's, you know, freedom of speech first, which means more trolls, more bots. more misinformation, but they can't hold back the truth. So if there is some truth about COVID, you know, being in a lab, you're going to get it first on X. But, you know, or if there is something that has to do with another conspiracy theory, you know, or something that's breaking news, you'll get it there first, but you also have to check your sources. And I think consumers are smart. I think they're going to realize they have to check sources over time and multiple sources and triangulate on the truth. But what do you think about labeling AI? You think this is something that should be either self-legislated. You know, the industry comes up with

Starting point is 00:32:58 their own group like the MPA did. The movie, the movie industry came up with the MPAA rating, so they didn't, the government didn't create the ratings, PG-13 R-rated. That was created by the movie. The industry self-policed. So what do you think has to happen here? Should there be federal laws, state local laws, should be self-police, you know, this is AI. And if you put an AI I generated image up, you have to label it. And if you don't, you can lose your account. What do you think? I mean, I like the idea of the industry regulating themselves.

Starting point is 00:33:31 Like we've talked about this before, right? Where all this, you know, the government has been trying to regulate and things are moving so quickly. So I like the general idea of the industry regulating themselves. And I do believe, at least on, you know, some of the meta properties, they do call it out now. Oh, really? You upload something, yeah. And you upload something and it's AI generated. It will detect it and label it.

Starting point is 00:33:58 You know, they'll say this is AI generated. Yeah. The video game industry wound up doing this themselves as well. Advertising industry does this as well. They have their own guidelines and code of conduct. You know, music industry does it with the parental advisory label, if you remember that. That was an industry-wide thing where they didn't want Tipper Gore, you know, Al Gore's wife. Yeah.

Starting point is 00:34:19 And the government regulated and they can. came out with, hey, this is parental advisory. There could be some spicy language in that. I think crypto tried to do that. It didn't work because crypto is such the wild west. But I do think in AI, if you're generating images, it should put a watermark or something at the bottom generated by AI. I don't think that's too much to ask. I would rather see the industry regulated itself than get regulated.

Starting point is 00:34:41 So my best advice is if something is AI related, it has a footer and it just says who produced it. Produced by whatever, grok, clawed, stable diffused. version 2.0. It's just like a little line at the bottom. And then if you want to crop it out, you can, but at least by default, it's not cropped out. I think that would be a major plus, you know. There was already some controversy this past week. I don't know if you saw the article in the San Francisco standard about Ben Horowitz and Felicia Horowitz. And someone did the mega-hat. They switch from the Democratic Party, supporting Democrats to supporting Republicans,

Starting point is 00:35:18 is pretty widely known. Ben Horowitz did a podcast with Mark and Dreson about their support of Trump and why they're doing it for very practical reasons around, you know, supporting the technology industry. But they put Ben and Felicia from like a, you know, not a paparazzi photo, but from like a gala photo. Yeah, yeah, yeah, I saw that. Yeah.

Starting point is 00:35:38 Did you know that was an illustration or not? Did you think that was real? Did you get fooled by it? I'm curious. No, I mean, I didn't take, you know, like obviously I was following. everything that's happened. And I don't take people in Silicon Valley that are deciding to report Republicans as folks that are wearing MAGA hats. So immediately, I, you know, I assumed it was fake. Mark Andreessen's position was they were lying and they did this to covertly confuse people. And I was

Starting point is 00:36:09 like, you know, I could see maybe five or one in 20 people or one in 10 people maybe glancing at it and thinking it was real, but it didn't look real. But they should. should have been more explicit, I think. And that's like a perfect example where in the footer, it did say this is an illustration. Yeah. Yeah. But I think it could have been more clear. And, you know, again, you get the reputation you deserve over time. The San Francisco standard now is going to have a reputation of doing those kind of illustrations like Spy Magazine did. So you will assume, like Mad Magazine or SNL, that that's what you're dealing with when you work with that publication. It could be like the National Enquirer, right?

Starting point is 00:36:47 Like that it's, you know, they're willing to do stuff like that. Yeah. And I think maybe we'll just keep segueing through here as well, Jason. You know, because it's related to it. And I want to make sure we continue to touch on some interesting demos, which is the image generation model that powers

Starting point is 00:37:08 Grok with a K is called Flux. And so Flux is made up. It's an open source project, which is really interesting. And it's made up of a bunch of people that came out of Stability AI. And what you can see in this, I just pulled up,

Starting point is 00:37:23 I didn't want to do side by sides, but the folks at Tom's Guide have gone through the work to do Mid Journey versus Flux, right? And they have some great examples here. And basically you can see they're effectively on par, right? Okay, so Flux is its own

Starting point is 00:37:43 open source. Open source. Mid Journey is closed source. Source, correct. Is it coming from the same code base in some way or the same training data? Because these are very similar. Well, what I would say is almost all of these models, like, you know, we've talked about before for LLMs where they were, you know, trained on large corpices of, you know, web data that's available, right? Yes.

Starting point is 00:38:09 Similarly, there's large corpices of images that are available for training on. And so once you figure out your strategy and you go down the training path, you know, Common Crawl, we've talked about this on the pod before. That's what most of the LLM folks were using to get data, you know, from all the open Internet. Similarly, there's an even part of, you know, the Common Crawl project, I think the image databases that they use. And that's why they come out looking similar-esque. The point that I'm trying to make here is that we now, again, have an open source project in flux that is capable of generating images. And so within GROC, it's not an image model that they created. It's a, they're using flux as the model underneath it.

Starting point is 00:38:58 Yeah. Yeah. I mean, exactly. Yeah. Yeah. I'm just saying we're, you know, we're at this position now where. These images are getting so close to reality compared to the last, you know, call it 24 months, that, you know, if you study them, you can figure it out. But if you glance at them, you will not know it's fake.

Starting point is 00:39:24 Yeah. Right? I think that's where we're at. Glancing, you'll get fooled. Studying 50-50 if you get fooled. You might be able to figure it out if you see fingers or some weird artifacts. But you know what to look for and you find that. that, right? And so, and so the big deal this week also was, we'll do another kind of real-time

Starting point is 00:39:46 demo here too, because we know Jake, we know Jake hell loves the demos. Some good demos, yeah, yeah. So the thing that happened this week, I'm going to switch tabs a few times here. So the thing is that people had flux generating these TED talkers. Like, and this, Jacob, you know, if I showed you this, you would not be able to tell me, you know. Hold on. I can tell you right now. left and the right. Hmm. You know, it's, I have to zoom in here and really get detailed to figure out what's going on here.

Starting point is 00:40:19 The one on the right seems a little bit fake and the one on the left looks real. No, they're both prompts. Oh, they're both prompts. This is the prompt from the left. Huh? And this is the prompt from the right. Oh, sorry, I lost the alt-tech. I mean, yeah.

Starting point is 00:40:37 So the alter, I mean, that's a photo of a beautiful smiling, burnout with long, messy hair. I mean, the level of detail is incredible now. Yeah. Yeah. And then we'll do one ourselves so you can see how straightforward it is here. What people were able to do with this is animate these. And so someone took that, aligned it with some sound. And we're going to do a demo of runway MLV2.

Starting point is 00:41:00 And I'm just going to play, let me just rewind this back. And you're all so kind. Can you believe that? I'm not actually real. But then, what is real after all? So. Yeah, close. I mean, that one I would know, it didn't pass the Uncanny Valley because the lip sync wasn't tight enough.

Starting point is 00:41:20 The voice sounds very real. The image looks very real, but it's not coming together perfectly. But I'd say 85% of the way there. What do you say? Oh, man. You know, I'm kind of, you know, like, if you do the work, I think you can definitely fool someone. Cal and that's going to be one of my tests coming up, right? It's when we get fooled, yeah.

Starting point is 00:41:42 Yeah, it's when you get fooled. So. Well, I mean, something's going to happen. Like, what do you think the chances are before the election, the week before, somebody drops a video of one of the four principles doing something inappropriate and says, look at this. This person said this word or that slur word or, you know, I don't know, touch somebody inappropriately.

Starting point is 00:42:03 Who knows what they come up with or, you know, I don't know, through a, bottle through a police car window, like, you could come up with anything. And then would that actually come out in time and not be able to be debunked and actually impact the election, you know, in the way people claim Hillary's emails, you know, being leaked, somehow affected, you know, the election. What do you think? Is that possible? Well, here's what we're going to do, Jake, J. Cal, where I'm inspired by your, um, lawnmower. Yeah, so I did image of Brad Pitt on a lawnmour. Right. Right.

Starting point is 00:42:38 Yes. Yeah, very similar to me. Yeah, that's how I look. Similar body type. The slim down and the, the, you know, the, Jay Ranch. Yeah. So what we're going to do now is I'm going to go to runway. So runway ML, as they've done a lot of new things and they have this new model where you can go from image to video.

Starting point is 00:42:57 So I'm going to upload the video that we just created. It's going to get uploaded here. I'm going to select it. I'm going to, you know, put it in the frame a bit here. Let's just do it like that. And this is runwayml.com, yes. Runwayml.com. And I'm not even to give it a problem because you generate me 10 seconds.

Starting point is 00:43:17 And this has been pretty fast today. So I'm doing it live. I didn't have to do it before. I wonder if it's going to understand that that's a lawnmower and make it go and drive around. Yeah. Because one of the first steps is understand what's happening in this image, right? And I didn't even tell me what's in this image. We could have given it a description.

Starting point is 00:43:35 We didn't even do that. And the reason we're kind of going through this process is just showing how simple this entire process has become. And so it's not perfect. Obviously, the wheels are turning. It's a little bit. You know, the parallel. Going backwards. But, but, you know, we did it in 10 seconds, Jekyll.

Starting point is 00:43:54 Yeah, I mean, it's a moving image of him driving. It looks off, but he waves a bit. But you could have him turn. You could say after four seconds turn. So you start thinking about. I was on my YouTube because I like Star Wars and stuff like that. You know, once in a while I'll see a fan film

Starting point is 00:44:14 and the fan films typically will use cut scenes from video games or some found footage and they'll reuse the same stuff from the original series or the Clone Wars. They'll stitch stuff together and then they'll have like themselves acting in their backyard as Jedi. But you could see somebody uploading a picture

Starting point is 00:44:34 of Obi-Wan Kenobi. and then say, have them open the lightsaber do this, do that. And it would start to make it, right? So here we are. Creativity is just unlimited. And you said, animate this such that the driver is turning in circles. Interesting. And we're starting to get to a level of speed that's not absurdly slow.

Starting point is 00:44:53 So these are big moments, I think. Well, remember just like when we started doing these demos, I couldn't even do these live, right? Because they would take so long. You couldn't do it. Yeah, they would take way too long. And so, like, I don't think it got them turning in circles. but, you know, it's better it zooms out a bit. But it's zooming out, which is incredible.

Starting point is 00:45:10 And maybe, oh, no, he is coming in a circle. It looks like, yeah. A little bit, yeah. Well, what we're seeing is the audio and the still images are really moving quickly. Video is still taking some time, but that makes sense because a video is 30 frames a second, 60 frames a second. So it's that many images. But, you know, this is going to become perfect over time. Here we go.

Starting point is 00:45:33 And you don't hear, see, even in the original image account, some of it's missing and how it's pulled it out and animated that in. Where we're going to is incredible. And I'm just going to see what it does on this hard right turn to see if it gets it right. If not, I'm going to show you something in a second in my next window where someone has created just in the last couple of days here. Let's see if it gets the turn right on this one. Now, I still not getting the turn right. So I guess we have to wait.

Starting point is 00:45:57 I guess we're still good. We're not. Oh, wait, wait. No, no. No, his hand went to the wheel and he is moving around. Yeah. Yeah. So what will be interesting is over time is this will become real time. You'll say, okay, have him driving, have him make a left turn. Yes. Have him turn to the camera and smile. It'll be doing that from voice prompts as you talk. Just like a director, James Cameron might be over the shoulder of a video editor, you know, doing stuff. Or if you were making an invite for, working with the graphic designer, you were making an invite for a birthday party. I might say, okay, you add some balloons and they add the balloons. And you're kind of working with the AI. in real time to change the image.

Starting point is 00:46:37 I think that's where we're going is real time editing of the stuff. And I'd say maybe we're 24 months away from a crisp real time edit, right? Yeah. Well, here's really about the silicon or the software to get there. What do you think?

Starting point is 00:46:50 What we're seeing the convergence of both, right? We're seeing the silicon go faster. Remember, when we started, we couldn't do these. I'd have to do them before. Wade in big cues. Now, I just did those lie, right? In front of us here. Incredible.

Starting point is 00:47:01 And the quality of them, you know, it's not quite there, but like we're doing it. Like we're, like, it's not our day job, right? So, and what I will say is, I want to show you this one, which I thought was really cool. Someone made a fan video. And, you know, think about like sort of going back to where we started our conversation on capabilities or if you're looking for a job or you want to do something like that. So imagine you're, you know, you're trying to break into creative at a big design or fashion house.

Starting point is 00:47:26 So someone created this video. So let me show it to you. Okay. So we got a Gucci video. Wow. That is some cyberpunk awesomeness. I like the headphones. They remind me on my focal utopias.

Starting point is 00:47:49 Incredible. I mean, if you were to shoot this, this would cost, by the way, each model would be thousands of dollars a day. You'd need them for a day or two. Give us the breakdown. If you needed to make this ad for, you know... A quarter million dollars to be, you know,

Starting point is 00:48:08 for 30 seconds, I'm thinking 10,000 a second. 300 grand. Plus, so yeah, wow. I mean, I just think it would be $300,000 for this level because you have makeup, hair, you have a set, you've got putting the music aside, which this track is a track that's available. It wasn't made by somebody specifically for this, but put the video editing aside as well

Starting point is 00:48:31 because that's not that much money. You need an art director, you need a makeup person, you need three or four, maybe, let's say, five models, director lighting. And the set design here is not. minimis. Like, I don't know if you saw all these floral arrangements to do that kind of stuff. Like, each of these shots might take a couple of hours to set up. That's the other thing is, like, set design and creating these things, like, hit play on it again.

Starting point is 00:48:56 And I'll just show you how many, like, there's a couch set up. It's the same couch in that one. Okay, there's that floral arrangements. There's that backdrop. There's, I mean, a liquid melting wall. And then there's like four other backgrounds. So each of those would be wallpaper. bird or, you know, like pinned up, you know, on a on a frame.

Starting point is 00:49:17 And then this one with the, that with the wallpaper and pouring melting paint over it, like that kind of shit is really hard to do. You might spend a day trying to create that effect. Or you got to go pay for the CGI. Yeah, this is all. These look like practical effects, not CGI. You wouldn't do these in CGI. You would do this all practical.

Starting point is 00:49:39 CGI is way too slow and time-consuming. for this, but on a practical basis, you'd have to build that wall, pour the paint on it, get the camera set up, throw that all away, put another wall, pour the paint on it, take another shot at it, you know, change the lighting or whatever. I think that's like maybe 10 hours of work to, just to get that one shot perfect, five to 10 hours of work. And that's, you know, let's say four or five people. They each have a day rate of a couple of thousand. So that's when this stuff really adds up, you know, and why these kind of commercials, you're wondering, like, how is this commercial so expensive. It's like, you know, four or five people, and I don't think any of the

Starting point is 00:50:17 one shots or scenes takes up more than five seconds. So in 30 seconds, you got six, seven scenes or sets. Yeah. You've got a background there, right? Yeah. My gendered back with a K. Yeah. Yeah. I did that fine. You know, that background you have there. If you had to set that up, that's some artist painting the wall, getting the lighting perfect and buying all that stuff. When I first looked at the background, I thought it was real for a second. It took me, Well, three seconds to just make sure. What I was going to do was, I was trying to animate it, so they were moving, you know, but that was one of my test. Well, I mean, it's funny.

Starting point is 00:50:52 You mentioned that. There's no reason in Grock right now, I couldn't say, you know, give me a Star Wars background with Obi-Wan in it. And it would, you know, be playing it in the background. Or, you know, hey, show me Hoth and then go to Tatooine and then go to Moss Isley. And it could just make those scenes in the background and know all that context. It could be a lot of fun. Yeah. All right.

Starting point is 00:51:11 Give me another demo. What else you got? Okay. Last one. Last one. We've got a lot queued up, which is, which is awesome. But, you know, we'll keep doing them, Jacob. Now we're back. So we're back. We're back. Every two weeks now. We're going to get on a good cadence. So here's one I really like, it's kind of what we were talking about before. It's called Avocca. And what they've done is they've specifically created an AI platform for HVAC plumbing and electrical businesses. And the idea is, look, the hardest thing for these, these are usually small businesses, you know, they don't have call centers, they don't have anything like that. They mostly have their, you know, number going to their cell phone or Google voice,

Starting point is 00:51:52 but what it allows them to do is handle all their calls and basically connect into their systems and let them make appointments and do all that kind of stuff. And so this is one where we're starting, you know, when we talk about this AI air gap, it doesn't exist. Because what's what's starting to happen is there's a whole segment of society. and that can leverage this technology to make their life better. In the same way, the internet helped them get on with web pages and Yelp and Google reviews. Now they can have a bunch of their business outsourced and handled by AI. Interesting.

Starting point is 00:52:28 So much of this, like, yeah, you see there, they're like, wow. I'm going to play this one right here. Just really quickly. So this is an AI agent handling the phone calls go. Thanks for calling nice heating and air, home of the cozy club. How can I make your home feel nice today? Hello. Hi there.

Starting point is 00:52:44 How can I help you today? Hello. Hi. Yes. Hey there. We've been a customer for four years, but I don't know if you have still a record. I want to make an appointment for a maintenance. I guess our air conditioning is not cooling that much.

Starting point is 00:53:03 Got it. Let's get you scheduled for me. I like the typing sounds. You hear the typing sounds? Yeah. Fake typing sounds. I'm not going to let it play you, but you can see there, obviously, the agent is AI, and the woman speaking is, you know, has a little bit of an accent, but you can see

Starting point is 00:53:19 them working together. And the benefit of this kind of stuff to people working on the ground, you know, hard work in HVAC electrical plumbing is going to be incredible. And I'm really excited to see that. Yeah. We have a very interesting company in the accelerator right now called MasterTech AI. Oh, okay. Okay. And so what they're doing, is since you were in the vehicle business, as you know, there are known issues with different models. So here, you know, and there are these technical service bulletin, TSP.

Starting point is 00:53:53 Yeah. So that if you're in the industry, you know what a TSP is. Any known issues with rattle noise from roof. And it's like based on service information, 2022 Ford Escape, yada, yada. So they now are taking all of those Tsbs and all other documentation, putting it into their own model. And then they combined it with a VIN scanner.

Starting point is 00:54:14 And so now, you know, you scan the VIN on this car, as you can see here. And you're off to the races. And then you, you know, and now you're a technician. Think about each step in the way to fixing this car that has some noise, right, engine noise. Well, you know, now you've got your whole series. The person who's running the, you know, the bays at this garage, let's say they have five bays, five, you know, three technicians. Think about all the time that's wasted, getting these things up and running, figuring out what the parts are, talking to a senior person, hey, I'm trying to get noise, where should I start. It's like, oh, start on the engine, then look at this, then go here. You know, you start putting together all this data into a language model and then imagine you've got, you know, it's a co-pilot for the real world, is how to think about it. Right. And, you know, they're not charging enough money, but, you know, if you have a small team with less than eight members, $100 per person, you know, what is a mechanic charge per hour now in the Bay Area?

Starting point is 00:55:16 200 an hour? Maybe more. I can't even know. Maybe more. Probably 300 an hour. And if you're in another part, you know, if you're in Texas where I am in Austin, like maybe it's half that, you know, it's a buck 25 or something. For the cost of one hour, 100 bucks, 150 bucks, you can make your whole team bionic.

Starting point is 00:55:33 You're going to save 10 hours a week, five hours a week, I think, out of the gate. And then the reinforcement learning is a really interesting part for me. So when I was jamming with them, I said, hey, is. Is it possible to put on a pair of spectacles that have cameras or wear a headband with cameras on it and start recording. The raybands have it. Like the raybands, right, would be one. But if you didn't want to wear the glasses, there are cameras, you know, from DGI, whatever, that mount when you're skiing, whatever. So you can just throw a vest on with the camera.

Starting point is 00:56:03 Now I'm working on this stuff and I'm talking and just record everything. you know, so now you got a Honda, a place that does Hondas and Toyota. You got another place doing Mercedes and BMW. You know how they kind of split these things up. And they're talking, the language model, it's talking back to them, but they're videotaping, hey, what are you seeing here? What do you think? And it's like, the training data becomes proprietary.

Starting point is 00:56:27 Yes. Or even if they're not talking to it, like you put your master tech, you just wear it for a year. Wow, there's some great ideas in that, right? Just to. Yeah. So I keep finding these verticalized. lunatics and you know, it's three people. So if you're one of those people and you got a crazy idea, come to founder.com university, apply. We might put 25K or 125k into your idea and be your

Starting point is 00:56:50 first investor. We've done 125K checks now. I mean, I'm kind of a lunatic, but I just thought, you know, if I could every fund put in 2.5 million of the fund, 25K at a time to get more people to just incorporate their company and become a forable company, if half of those go on to an accelerator and half of those go on to a seed round, you know, we're going to own two and a half percent for 25K of 25 viable companies that got to their seed round. You know,

Starting point is 00:57:17 we just have to hit one to make it work. And we can really help more entrepreneurs pursue these crazy vision. That's like serious driving entrepreneurship. I mean, like the government should just double you up right off the bat. At this point, small business administrator or something, it's,

Starting point is 00:57:32 you know, it's a, think about your first company. If somebody with, you know, listen, I'm not, I'm not like having delusions of grandeur here.

Starting point is 00:57:39 But if somebody like me or my team vetted you as one of a hundred companies and gave you that 25K as your first company, that opens a couple of doors. So we have a company called Tax GPT and they got into you, I think you met them. It's an AI tax assistant and they're kind of going crazy. We did the same thing with them. You know, they're a co-pilot. I'll show you this one. And this is my entire life now is just verticalized, you know, businesses using. AI. And so

Starting point is 00:58:08 now they're doing a co-pilot for taxes and it's really working. We gave them, I think, their first 25K, then they went through our accelerator, now they're in Y Combinator. And you don't need a ton of money to start a company is one of the big messages

Starting point is 00:58:24 I want to send to people. You know, you get that first 25K check, you give it to your lawyers, you set up your servers, whatever it is. Then the 125K from us or from Ycombinator or TechStars, whoever you jam with, that could get, you know, two employees, they're kind of ramen for the next six months, what we call ramen funding in the industry. And if it works, great.

Starting point is 00:58:43 You know, you spent 150K and now you're ready for a seat round. And if it doesn't work and you pack it up and you shut the thing down, it's not the end of the world for anybody involved. Not for you, not for the investors who put in the 150K. It was run as an experiment. So let's run more experiments as my message to the entrepreneurial community. And if you want to do that, founder dot university. That is, no, and that playbook, in fact, even

Starting point is 00:59:07 if you just go through, like you said, your 25K and 125K in your white combinator, you may still have a viable cap table where you can break out. You don't even have to go and get Series A funding or anything like that. Sometimes that does happen. You can become ramen profitable, you know, where you start making enough money. We call ramen profitable, basically. You make enough money to pay for your apartment and your ramen. You're not getting rich, but you're at break even. What a powerful place to be as an entrepreneur. Sunny, it's so great to be back. I miss you. Come to Austin.

Starting point is 00:59:38 Let's have some barbecue, play some poker with our boys. And I'll be in the bay. I'm coming to the bay every couple of weeks. And so when you get the GROC jet, you zip the GROC jet down here, pick me up. We'll do a little back-to-back poker games here and abroad.

Starting point is 00:59:52 I love you, brother. Everybody go to X.com slash Sundip. And if you are interested, console.grock.com, GROQ.com, not K. GROQ is sunny. They make that inference hardware that, you know, instead of watching the text get drawn,

Starting point is 01:00:09 imagine boom, it just comes out, boom, like that. Boom, you get your answer. That's what GROC does. How many developers now?

Starting point is 01:00:17 I think we just crossed 400,000, right? And so... Wow, 400,000. Here it goes. Look at this. Look at this, GROQ.com. Watch him do this. Tell me a story about unicorns.

Starting point is 01:00:31 Wow. I mean, it's just demented how fast that is. And that's only possible. with the hardware that you make, right? You have to have... Only possible with our hardware, yes. Yeah. And people are buying up this hardware

Starting point is 01:00:44 so that consumers have a faster experience. Will that hardware, same hardware, translate into mobile phones eventually? Do you think? Is that the roadmap or it's mostly on the back end to present it? It's mostly on the back end. To be honest, yeah, it's all...

Starting point is 01:00:57 Yeah, we've built it for that purpose. Got it. And some of the customers? Are you public about some of the customers of this yet or now? We're not sharing that just yet, but some of the biggest companies in the world. Yeah.

Starting point is 01:01:08 Makes total sense to me. Listen, I slid in a little bit. How much am I up? Am I 10x yet? Where do you put me at? Do you break the 10x? You know, you always, every time I invest one of your companies, I get the 3 to 10x. I'm wondering if this is the one that goes 50x.

Starting point is 01:01:24 But I'm just trying to benchmark. You're somewhere between 5 and 10 for sure. Okay, that's called a sunny special. I always hit the 5 to 10X. I love a sunny special. But I'm looking for a Madra madness on this one. I'm looking to go 30 to 50. I'm looking for 30 to 30.

Starting point is 01:01:40 Oh, my good. I mean, I just want to see you hit like right out of the park. So that means you just got to, you got to just 10x where you're at right now. Make this a $20, 30 billion company. We're good. I'm in. We'll see you all next time this week in startups.com. Bye, bye.

Starting point is 01:01:54 Hey, everybody. Welcome back to this week in startups. It's time for our next Jam with JCal session. What is Jam with JCal? It's a very simple process. A founder tells me about their company. Maybe they pitch me or. show me the product.

Starting point is 01:02:07 And then they tell me, hey, what are their biggest challenges? And we go back and forth, we have a little bit of a dialogue about whatever their biggest challenge is. And this, of course,

Starting point is 01:02:17 is brought to you by our friends at dot tech domain names. The way we do this is, you have to have under $2 million in funding. And so it's for early stage startups. Ulama is our first startup, and the founder is Tys Hehrman. And T-I-C-I-C-E,

Starting point is 01:02:35 pronounced T-I-C-E. I think, Tice. Yep. Really interesting name. I've never come across it. And your company is Ulama? Yes. And that's ULAMa.

Starting point is 01:02:49 dot T-E-C-H. You got a great dot-tech domain name. Why don't you show me what you're working on and then just tell me what your biggest challenge is. And maybe explain to us what your product is, who your customer is. I just start here so we get a little bit of context before we drop into this demo.

Starting point is 01:03:03 Sure. So our software is for architects to analyze their 3D building models for code compliance before they submit their designs to government reviewers for approval. So we like to think of ourselves as grammarly for architectural 3D building models. We have software that plugs directly into architectural design software, and architects select the location that they're in and the specific codes that they want to use

Starting point is 01:03:29 to analyze their 3D building model. They'll go through the sort of two-dimensional drawings, of the 3D building model and assess them for specific codes. So that could be building code, it could be accessibility, ADA code, plumbing, fire, the architect can select what they would like to use. And our plugin then runs through their entire 3D model evaluating all the geometric and parameter relationships within that model. Once that's done, they get both text and visualization of how their model,

Starting point is 01:04:05 model is out of compliance and suggestions on how to make updates to the model to bring it into compliance. And that's broken out by sections of the code and also building elements within the 3D model. Got it. So an architect is pulling up this floor plan. It looks like a live workspace. And then they put your plug in into their architectural software. What's the architectural software most common one called? It's called Revit. Revit. Okay, so you're a plugin for Revit, and then it goes to their designs, and then it matches them to the local rules and regulations, the codes. Is that correct? Yeah, that's right. We've created a rules engine that will look at their entire 3D model. Is that using AI to do that? How is it doing it?

Starting point is 01:04:57 We are assessing the 3D model with an multimodal AI model, and we're also creating those rules by parsing the regulatory text with AI models. Got it. And so what jurisdictions do you work in? How long have you been doing this product? We've been working on it about a year. And right now we're just doing codes that are broadly applicable in any municipality. So ADA federal policies or sort of vanilla off the shelf building codes.

Starting point is 01:05:29 So ADA policies, Americans with Disabilities Act, I believe is what it stands for. What's an example of that? that like the width of a doorway or how bathrooms work? What's an example or the most common example? Yeah, those are both great examples. A concrete one would be a toilet has to be a certain distance from the nearest sidewall so that someone can reach a grab bar within a easy, comfortable distance. Got it. And so it's going to check the ADA rules against what an architect made, correct? That's right. And did the architects frequently screw this up or do they generally know all these rules and get it right? They have a decent working understanding of code,

Starting point is 01:06:12 but there's so much minutia within code that it's difficult for them to keep track of it all as they're designing. Most architects are doing what's called parametric design. So as they're making changes, other things in the model are changing downstream of that. And so things can get out of sync really quickly. Got it. And this isn't occurring in real time. You basically upload your your floor plans and then you run this against it and then it tells you. Yeah, this, they don't even have to upload anything. This is within their design software, but they do it on demand. So it's not just constantly pinging them, hey, this is out of compliance.

Starting point is 01:06:48 Right, because that would be what a co-pilot does. A co-pilot is sitting there while a person's working, let's say, like grammarly on your grammar or a co-pilot like GitHub's while you're writing code. So I'm curious why it doesn't do it while you're working. Yeah, we talked to hundreds of architects at this point. And a lot of the questioning was early on about how frequently do you want this sort of analysis? And most architects just want to be able to design. You turn on their design brain and then at certain checkpoints come in and look at things like code compliance.

Starting point is 01:07:26 But they don't want that constantly getting in the way of their design flow. That makes sense. They're trying to get into a groove. they don't want to have this interrupting them. Tell me what's the biggest challenge with your business? And it seems like a very reasonable idea and a good starting point for a feature, if not a product, like an interesting wedge. What's your biggest challenge?

Starting point is 01:07:46 I'm curious. Yeah. One thing that we're thinking about actually right now is we're launching a smaller version of this product as a standalone offering. It's going to be for sale on the web. It's ready to go live. tomorrow once I finish my distribution list. And it's a component of this overall technology for code compliance.

Starting point is 01:08:08 It goes through and it identifies all the elements in that 3D building model and then normalizes the attribution in naming. Right now, it's not very standardized. And that makes it hard for architects to do all sorts of other activities. So what we're struggling with thinking through is, how do we use this initial product launch to generate sort of interest, and also make this sort of case that, hey, we know how to develop products for our customer base. This isn't our core product and our sort of main value proposition, but we're servicing this

Starting point is 01:08:42 sort of customer base and just sort of like telling a good story around that through this initial product launch. So are you looking to launch us in the United States and Germany? Where are you looking to launch us? I'm assuming the United States. This first product that we're launching can be used internationally because all it's doing is identifying the sort of elements within the building. It's not applying any of the rules yet. Okay. Well, I know that there are hundreds of thousands of architects between the United States and Europe, right?

Starting point is 01:09:11 Yes. And so they already like to spend on software. The software they use costs hundreds of dollars a year to low thousands of dollars a year, I would assume. Yeah, the software developing for Revit is thousands of dollars a year, and that's just their sort of baseline design software. Got it. And does that software have an app store for the plugins like Shopify does or Apple does? Autodesk, the maker of Revit, has the app store. It's not commonly used yet. It doesn't have a lot of traction like an out app store, but it's gaining momentum. So, you know, this is always the challenge with these app stores, because if you leverage the app store to try to get customers and you're very successful, you've basically given a roadmap to those platforms to incorporate you into their product. Without knowing Autodesk and how they work, do they historically support a really great developer ecosystem of plugins, or do they have a reputation for basically stealing those innovations and putting them into their own products?

Starting point is 01:10:21 less the reputation of just sort of like taking them and inquire and you know embedding them in their own products they do a lot of acquisitions now there's you know giant company at this point so a lot of their new development is through acquisitions yeah and their developer support has been sort of lackluster in the past but they are really renewing their focus on that so there's a lot of momentum around the app store yeah that's great so you know in a lot of these cases how the app store features you and your relationship with auto desk is going to be the key driver. So if you make Salesforce plugins or Shopify plugins, there's usually a developer evangelist, a vice president there. And if you, you know, make nice with them,

Starting point is 01:11:04 they might feature you and bring you 10 customers a day. And if you can work that ecosystem, you know, that's going to be incredibly accretive to your business growing. And it also builds a dependency, but that dependency also builds an acquisition path. So,

Starting point is 01:11:20 you know, there's good and bad here. The good is they could, you know, really appreciate what you're doing, not have time for it, and you could basically own this little piece of the puzzle for architects. And so, you know,

Starting point is 01:11:34 and then you could have a dependency. They could decide to take you off that homepage or promote somebody else or build it themselves and all of a sudden your business goes to zero, which has happened to people who build plugins, specifically the developer community at Facebook, famously like Zenga, got absolutely demolished.

Starting point is 01:11:50 over time. And so you can look at those historical precedents. The more you understand your customers, the more you spend time with them, the better off you're going to be. I would try to embed your company into, you know, wherever the most architects are, whether that's a school or, you know, a company that you can say, hey, I know you're going to get a lot of value out of this tool. Can we, you know, have two seats at your office? And we call this the bear hug, you know, if you could find out, you know, who is a great customer for you. They've got 50 architects, 25 architects, and you say, hey, to the CEO,

Starting point is 01:12:28 some upstart and the founders, would it be possible to get two desks here and hang out? That's another way for you to get closer to the mix, to the people who are using it. So those are my two paths for you, building the developer community, or bonding with the evangelist at that app store, or deeply bonding with, you know,

Starting point is 01:12:49 an architectural firm or doing both. And then if, you know, they decide they're going to compete with you. And, you know, it's harder for them to compete with you and kill your company. If you've developed that relationship with the evangelist team, they're going to be like, eh, let's keep supporting them. They're only charging $500 a year per architect for this plugin. We're charging $5,000 a year. You know, that seems like a good price, you know.

Starting point is 01:13:15 So getting your customers to pay also critically important at this early stage. stage. If people aren't willing to pay for it and you've put a lot of work into it and it's, you know, reliable software, you know, then that should tell you like maybe this isn't providing enough value. And that's where pricing comes in. So those are my two hopes for you. It seems like you're off to a great start and I wish you great luck with it. Any other questions for me? Yeah, I guess maybe if there's anything that you can say about leveraging an early product launch or a second larger product launch. And how to really like connect those to like this,

Starting point is 01:13:55 you know, previous, hopefully success when we launched tomorrow to our major and main product that we're launching end of the year. Great question. I've seen some entrepreneurs set up little WhatsApp groups or signal groups or I message groups or Slack rooms with say five or 10 like minded customers who become their product advisory council. And then you've got five or 10 people who get early access to what you're doing and get to influence the product, they're your product advisory council,

Starting point is 01:14:23 but then when you do launch, hey, they might tell people about it. So there's probably 50 architects who have YouTube channels. There might be 50 people on Instagram who are active architects. There might be 100 of them on Reddit in two or three different subreddits.

Starting point is 01:14:37 And if you can just slowly collect 10 people from each of those social networks, tell them what you're building, ask them for feedback. You know, you might have this little group of highly influential people rooting for you. And I think, you know, I call this the slow embers, hot embers kind of marketing. It's not like huge numbers.

Starting point is 01:14:57 It's more like really hot coals. You know, you ever have a really hot coal? You take one of those coals and you put your hand of it. Whoa, it's a little bit hot. Now you put 10 of them together. And, you know, you're about a foot back and you're like, whoa, I can feel the heat. You know, then you get like 50 of those coals in a barbecue and they all start turning red. You know, now it's like, whoa, I can't get.

Starting point is 01:15:16 within three feet of the barbecue. And when I put that tomahawk on, you know, that tomahawk steak, man, it just lights right up and it's almost too hot. But each one of those is just a coal. One little simple coal. But together, 50 of those calls can burn so bright, so hot that boom, and it just explodes, right? And it's just, it's even too hot. That's really what building this fire is about.

Starting point is 01:15:41 So don't forget, each coal is, in your case, a customer, an individual. and you just have to build that relationship. Now, each of those relationships will probably take you about an hour, so it's about 50 hours of work, which you're probably like, wow, I got to push code, I got to do all this stuff.

Starting point is 01:15:55 And this is why sometimes people will have an evangelist or a marketing person just slowly build up that group. And I've done it myself. I have, you know, maybe I would host dinners for other angel investors and LPs in our funds

Starting point is 01:16:11 as a venture capitalist. And I still have those, you know, little groups, of who went to dinner at different locations. I named them in Signal, you know, the name of the place. And, you know, we have a Marks-Off Madison group with like 12 people who came to that dinner. And they talk to each other once in a while.

Starting point is 01:16:26 And I get to have that like one to few relationship. And they might ask me, somebody might ask me a question and we all talk about it. So I like this like one to few kind of concept. And it could be a Zoom call. It could be a WhatsApp group. It could be an email list, whatever the jam is that you want to try. So I really think that will help.

Starting point is 01:16:45 your launch. Press is nice. If there's industry press or a podcast, and podcasts, there must be a hundred architectural podcasts and they probably all get like hundreds to low thousands of people. But going on those and talking about your product and the problem you're trying to solve, people love innovation. And so I would also look at the podcast route, right? And I think in your case, small numbers are good numbers, right?

Starting point is 01:17:11 Targeted small numbers is going to be the key because you're not, dealing with a consumer product like Uber, Airbnb, where you're trying to get a billion people to consider using it to get 100 million or 200 million users. You just need 10,000 paying customers and this thing becomes a $10 million a year business and that's pretty good, you know, as a, you know, a goal for a business like this. But I wish you great luck with it and awesome start. Thank you. Really appreciate the help. All right. All right, everybody, if you want to jam with J-Cal and just talk about your business, have it featured here on this. week in startups. Contests is open. We got one more slot. Go to jam with j-c-c-c-al. Tech. That's

Starting point is 01:17:51 J-A-M-W-I-T-H-J-C-A-L dot-T-E-C-H. Jam with J-C-L-T-L-T-E-C-L-T. Get one of those great. Dot-T-T-D-T domain names, and we'll see you all next time. Bye, bye.

This Week in Startups - AI Demos: New Runway, Flux, Grok 2.0 & Jam Session with Ulama | E1997

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.