This Week in Startups - AI Demos: Sunny’s Back with Luma Labs, Kling, Claude Sonnet & Getting AI Native | E1976

Starting point is 00:00:00 There is a lot of innovation happening now, and it's faster than ever. And it's coming in the way of new types of models, open source models, agentic reasoning, in voice. And so in the second half of this year, J-Calc, we are going to have some surprises that no one was anticipating. Okay, there you go. So that's inside information. That's the inside line.

Starting point is 00:00:19 Or a prediction. Generally, I'm telling you, we're going to be more surprised in the back half of this year than we were when opening high first came out. Oh, it's a big claim. It's a big claim. I don't know how we phrase it as a bet. But what you're saying is everything that happened up to now is the epilogue and that the story's coming. The real story is coming now.

Starting point is 00:00:37 This week in startups is brought to you by NetSuite, the number one cloud financial system bringing accounting, financial management, inventory, and HR into one platform, giving you one source of truth. By popular demand, NetSuite has extended its one-of-a-kind, flexible financing program for a few more weeks. head to NetSuite.com slash twist. Open phone. Create business phone numbers for you and your team that work through an app on your smartphone or desktop. Twist listeners can get an extra 20% off any plan for your first six months at openphone.com slash twist.

Starting point is 00:01:17 And DevSquod. DevSquod helps startups design better products. If you need UI and UX expertise and don't want to hire an entire design team, head to DevSquod dot com slash startups and book a call. Mention that you are coming from Twist to get 10% off. All right, everybody. Welcome back to this week in startups. Madra Mondays are back.

Starting point is 00:01:41 Here is my guy, Sandeep Mandra. I don't know if we're releasing this on a Monday. We're taping it on a Monday. Taping it on a Monday. Welcome back to this week in startups. People are like, yeah, where you've been? Where you've been? Everybody's asking where's Sunny?

Starting point is 00:01:51 Where's Sundee? Where's Sundee? Business called fundraising. And you know how it is when you're doing that. The calendar got prioritized for. I get it investors and you, you know, you've had to do it too. I don't know, Jacob, actually, I'd like some advice from you. Like, you're able to keep the shows going and you have multiple shows and fundraise. But for me, you know, fundraising, you know, it's kind of like took over my

Starting point is 00:02:14 calendar. It's like, we're meeting with this and you can't really move them around. So how did you manage that? I broke myself as how I managed it. I mean, if I'm being honest, I didn't manage it. It literally broke my brain. Like, we are living in the toughest fundraising. environment since probably the dot-com era, possibly, you know, right ahead of the Great Recession, I would say. And so, you know, with an AI company, that's a notable exception, obviously, but even still, because there haven't been a lot of returns, if you look at, you know, IPOs, they've been few and far between and they've been muted, right, Instacart coming out at $8 billion, private valuations being $30 or $40 billion for that company, which means LPs and

Starting point is 00:02:58 VCs are less apt to put money into products. So we've had a little bit of indigestion. The gears are a little gunked up. We talked about it on this pod a couple times. M&A is really anemic. We used to have a very vibrant mid-market for M&A since the Biden administration hired Lena Khan. That has destroyed the startup space. And I think putting aside any politics, how you feel about different presidential candidates, this is an area that really matters to America's viability. If there's no mid-market for selling Figma to Adobe or Instacart to DoorDash, whatever it is. Or the little teams, the little acquisitions. And the tuckins, like you're referring to, 10 million to 100 million tuckins, that mutes investment. So now you've got LPs and VCs, either not investing in as many

Starting point is 00:03:51 companies, not investing as many dollars in companies, companies not being able to raise funding, which then means less job creation, which means a less competitive America. You have to have two different concepts for M&A. M&A for Facebook, meta, Google, Amazon, Microsoft should behave one way. In other words, companies over a trillion dollars. They should have one rulebook for M&A. For the mid-market, there should be a different rulebook. So if Uber and Airbnb, Coinbase, you know, call at the 30 billion to 250 billion crowd. If they want to do interesting things, I say let them. You can review them.

Starting point is 00:04:28 And then I would say any acquisition under, let's pick a number, one billion dollars. Just you can do whatever you want. Under a billion dollars, you can do everyone. And if you've watched, we've now had two AI acquisitions in the space where they were aqua hires. With leaving the shell company and somehow money's flowing to the investors. they're getting some of their money back somehow through like backdoor contracts. It's all a giant scam, a hack for route around Lena Kong.

Starting point is 00:04:58 So if you tighten your grip, more deals will slip your finger. So Lena Khan's a complete disaster and binds a complete disaster when it comes to this M&A situation. It also ties into like incentives drive behaviors, right? Like these big companies can also put offers in front of people which look like acquisition. Right, which, you know. Say more. Explain. Well, look, like, you know, if you're Microsoft and you want to staff up a new AI, you know, division, like, as they did, right?

Starting point is 00:05:30 Or you're Amazon and you want to step up an AI division like they did with the, you know, adept thing that just happened last week. You offer the founders $10 plus million dollars a year or over two years. You know. I may bow. If you stay there, let's just say you get $10 million a year and you stay there for $5. years, right? That'd be probably the equivalent of you selling a startup for $500 million and you owning like 10% of it after a couple years. Yeah, you and your co-founder are 10%? Exactly. So now what is all this, you know, crazy M&A philosophy of future competition done?

Starting point is 00:06:06 They just routed around you, Luna Con. Yeah. Yeah. They routed around you and you accomplish nothing. And then what happened was the big companies get stronger because now they don't have the competitors in the smaller company. So you're actually having the opposite effect. And the investors kind of get weary where the investors are like, hey, I'm not sure I want to do this because what if the team gets scooped away or decides to go away or walk away? Yeah. So there's unintended consequence. Show me an incentive. I'll show you an outcome. And here's what happened. So you've got to be very careful. If you want to put your thumb on the scale, Nina Khan, you could just tilt the scale and everything goes flying off the scale. And you don't catch any of it. You don't do any of the things you intended. So it's just a terrible approach. It's been executed very poor. But let's get to demos. Everybody wants to see demos. I've been playing with Chet GPT40 and the new Claude Sonnet,

Starting point is 00:06:56 and I pay for both of those, obviously, for my companies. And it's getting pretty damn impressive. Well, yeah, Claude Sonnet is where I wanted to start today. You know, as we get into it, I just want to give credit here to this really incredible thread that Min Choy put together, and then I'll show a demo from it, but I will give this credit real quick.

Starting point is 00:07:15 He really went and went across sort of a bunch of different ideas and came and aggregated them. So here's build an iPhone app prototype from scratch. So really the thing that we're going to talk about when we do, Cloud, is they have this mode now where there's like a, on the right hand side, you have that enabled, where it can do sort of an output for you that's different than the text output.

Starting point is 00:07:37 And so here, someone made an app. I'm going to demo this next, which is make an interactive dashboard from an earnings report. You can make games. You can go through this, right? You can get insights on business, one-click SEO tool and not. So we'll put a link to this, but what I did here was I basically dropped in, you know, Tesla's Q1, 2024 earnings report.

Starting point is 00:07:59 A PDF. Yeah, a PDF. And I said, come up with a detailed analysis and summary with stock recommendation by carefully reading this report, put together a nice and interactive but highly detailed report. And so here's sort of the summary, which we were all used to getting. What I got on the right hand side is an interactive report. Wow. So this is the presentation layer. So instead of just seeing text that you would copy and paste and start building a deck,

Starting point is 00:08:25 it knows that you want an interactive report. And that means it's making something beautiful. Now, I don't know what format that is. It looks like HTML, but well done. You have to enable this thing called artifacts. If you turn that on, right, in the features preview, you know, I'll just copy this prompt over and I'll add in a, you know, here's Nvidia's last 10Q, right? And, you know, you're sort of limited to the upload side,

Starting point is 00:08:50 which I think is like 30 megabytes or something like that. More than enough for any kind of text. And I like how you tell it to be like thoughtful or something or to be thought. The AI needs to know that. Just in case it doesn't know. Just in case it doesn't want to do its best work. You've given an encouragement to do its best work. We have to be mindful to our future overlords.

Starting point is 00:09:13 Absolutely. And so what you see here, it's obviously doing the standard analysis. And what it's going to start doing once it's done this, it's going to basically open a tab to the right-hand side. Wow. And you see here is where it's starting to create the interactive dashboard. So it's basically, you know, writing all the code. The code for it, right? So this is being done in HTML.

Starting point is 00:09:35 This looks like JavaScript or something, right? And so that it can be interactive. And so at the end of that, you'll get this. And so now. So it says vehicles delivered, operating margin, the production, and then even a chart. Yes. They're deliverables, revenue, net income. Exactly.

Starting point is 00:09:52 And then a quick summary. And you can download this, you can copy it out. So think about how powerful this is. Like if you have a bunch of data from, say, an event you just ran, like the liquidity summit, if you have all that data about, you know, how people interacted and how many people went to different sessions or whatever it was, you throw that in there, ask it to do the work. And basically, it comes out with a really, really nice report for you. Whereas before all we were getting were these kind of text updates, and you had to take these and go and copy and paste them into PowerPoint or do something,

Starting point is 00:10:24 now it's creating the dashboard for us. This is a huge jump forward. The less your business spends on operations, multiple systems, and on delivering your product and service, the more margin you have, the more money you keep, right? You want to get fit in this new era here in Silicon Valley and in tech broadly. but with higher expenses on materials, employees, distribution, and borrowing, of course, everything is costing more.

Starting point is 00:10:49 So to reduce costs and headaches, smart businesses are graduating to NetSuite by Oracle. NetSuite is the number one cloud financial system, bringing accounting, financial management, inventory, and HR into one platform, giving you one source of truth. With NetSuite, you're going to reduce your IT costs because NetSuite lives in the cloud, and it can be accessed from anywhere. This means you cut the cost of maintaining multiple systems. with one unified business management suite, and you improve efficiency by bringing all your major business processes

Starting point is 00:11:19 into one platform, slashing manual tasks and time-consuming errors. Over 37,000 companies have already made the move, so do the math to see how your profit with NetSuite. So here's your call to action. By popular demand, NetSuite has extended its one-of-a-kind flexible finance program for a few more weeks. Head to netseweet.com slash twist.

Starting point is 00:11:41 That's netsuite.com. So I don't know if you've tried this with your teams yet, J-Cal, but give me your thought. It's absolutely extraordinary. And we have a lot of information that I would like to put through this. So one of the things we've been doing since we're an at-scale venture firm doing 100 a hundred investments a year, think like, you know, 25% of what Ycombinator does a year, maybe 20%, they do 500 companies, we do 100. We get a lot of information from those companies back.

Starting point is 00:12:06 We put them in a database. We're currently using Notion as our database. And to be able to pull that data from them. and then say, hey, make us a report, which of our companies are in group one, two, or three. And we have like a very basic, I like creating simple systems. We put our startups into three buckets. One is like high growth, consistent growth, you know. In other words, they're probably series A, series B.

Starting point is 00:12:29 They figured something out. They have product market fit. Bucket two, figuring out product market fit. And then number three is they've run out of money. I don't know. The co-founders broke up. In other words, they're still active in some way. but, you know, there are something's, you know, in all likelihood going to wind down and they're

Starting point is 00:12:47 trying to find a buyer. They're in the middle of M&A. And, you know, we have humans figuring that out, analysts and researchers. But eventually, AI should be able to figure that out. And we should be able to then do some analysis over time of companies and say, hey, which companies look like they might be the next to move from group three to group two, from group two to to group one. Yeah.

Starting point is 00:13:07 And look, you know, this is the one that we just did. So it's slightly different. So you can really tune this and it comes up with incredible. Incredible. I've been using this quite a bit because I think if you're ever any task to do, even if you just want to brainstorm for yourself just to get this visualization. So I find, like I said, like this type of prompt, which definitely came from one of the threads that Minn shared.

Starting point is 00:13:29 So I want to give him kudos for it. But it's a really useful prompt and it's powerful. And it's a great way to analyze a bunch of things. Yeah, these things are definitely getting smarter. I give this. this is an A plus for me. I'll be honest. I kind of finally feel like these LLMs are not doing random moments of interesting potential.

Starting point is 00:13:51 I feel like they're nailing it. And I'll give you one example. Maybe you could just fire up chat GPT40 and says you got the screen there. We're opening up an accelerator and a studio in Austin, Texas. So you heard it here first. And so I'm spending a little more time here. and imagine you were looking for, I don't know, a personal assistant, right?

Starting point is 00:14:11 I don't know how much a personal assistant. I know what that cost in the Bay Area. I don't know what the question. So here, just write the sentence. Please tell me how much a personal assistant costs on an hourly basis in Austin, Texas. Please cite five sources and please put the high, low, and medium salary in a table.

Starting point is 00:14:27 And I did this in 4-0, and the result to me was extraordinary. Okay, great. Okay, here it goes. Okay. So when I did this, it was unbelievable to me. So here you go.

Starting point is 00:14:39 A personal assistant in Austin, low salary, median salary, high salary. And I just said, hey, can you link to it? And it gave me a similar result here. And so it's really, and you remember when it used to go out to Bing last year, you know, we'd do the search and it never worked. Yeah. And it crashed. And like we'd ask it to put stuff in a table.

Starting point is 00:14:57 And sometimes it'd put it in the table right. Sometimes it wouldn't. But here it is. It's now quoted salary.com. And now just give it a follow up and say, give me five different. sources of data and average those sources. And so when I did this, and I said, give me five different sources, it gave me Glassdoor, Indeed, Salary.com.

Starting point is 00:15:15 And I was just blown away because when I got, here you go. ZipRecruiter, ZepRecruiter, Zephia, I don't know that one, Glassdoor, Indeed, salary. And now you start to see a range of them and you see the sources. This is what a college-educated researcher would have spent in our companies in Silicon Valley, How many hours? Two, three, four? Half a day.

Starting point is 00:15:39 This is like, you know, the way I think about it, you'd come in in the morning, you'd tell someone, and maybe in the afternoon you'd get an email back with this. Right. And so this is wasted work for a human to do now in the age of AI. So when you see that companies aren't adding a lot of positions and you're wondering, why is that happening? How is Uber and Google and Facebook growing, whatever they're going? 15%, 30% year over year, adding billions of dollars, but they didn't add any people.

Starting point is 00:16:08 This is a perfect example of this. I would normally ask somebody on my team to do this in operation. The amount of time for me to explain to them and get the result is greater than the amount of time it takes me to ask you to be longer to write the email because you'd have to be a little bit more formal and detailed. Correct. And I'd have to tell them why I'm doing it and all that stuff. So this now is really becoming, as I said, extraordinary.

Starting point is 00:16:32 So I just want to say to the team over at OpenAI and Chatty-F24, incredible job on giving citations here, which, you know, we're sitting here a year or two ago. We would have been complaining about citations. But this idea that the work is actually good enough to put into actual production is something we saw with developers checking the work. Now we're seeing it in operations. We're seeing it in data analysis. This is heating up, right? I'm trying the same thing just in clots on it, right? Yeah, and Klausana did a very good job at this.

Starting point is 00:17:05 Yeah. So here you go. I mean, boom. And you could actually create a visualization. Yeah. Now, what you have to worry about here, I think, is this goes back to our discussion about, is this stealing or not? Yeah.

Starting point is 00:17:19 And so why would I ever click through to ZipRecruiter Salary or Indeed.com? There is no reason for me to ever go to those websites again. 100% of my attention and money is going to clawed and to opening eye. I'm paying 20 bucks each for these services, maybe 30, I can't remember. Yeah. And so not only is the human being taken out of this that would have worked for me previously. When I say previously, last year, last year I had humans on my team working on this. I no longer have those humans working on our team.

Starting point is 00:17:51 This is definitely a real life problem for you, where you were like, hey, I need to get an assistant. Please help me. So I know, yeah. Now, the next phase of this would be write a job description. Okay, it's going to do that very easy. So now I just say, write me a job description for this job for somebody with five years of experience and give me 10 bullet points to choose from for the skill set. That would be something.

Starting point is 00:18:13 Again, I would ask, a human being working in Silicon Valley or working remote, let's just call it $50,000 to $100,000 a year salary, depending on if they were right out of school or if they had five or 10 years. And so here we go, boom, about the role, key responsibilities. To write up a job description would have been a ton of work as well. So now the job description is done. So if you're an HR professional who's been in the field for 10 years and you work with

Starting point is 00:18:43 management to tell them in an interactive discussion, how much it costs to hire people in a city and, you know, to write a job description, it's done. Now, the next phase of this would be post this to five services, use my corporate card. And then the next phase of this is sort through the candidates, ask them five questions on email, and then rank them for me and put them in a table and schedule an appointment with them. You want an agent that's designed specifically to work on this for you, both in real time and offline, right? Where it starts and then once it's going, it's like, hey, look, post it, come back, and even set up interviews, maybe even do a screen of these folks. Well, yeah, that's kind of the next part

Starting point is 00:19:23 of it. So if you're an HR startup, these kind of things are easy. The next hard part is, of course, the agent doing stuff. And now when we talked last year about the Maestro concept I had where I'm in myestro running a company, if you're a startup, you're not hiring an HR department, obviously, and you might have hired an HR consultant or you might have hired the proverbial jack of all trades, right? The utility player on your startup. It was a place for the utility player on the startup. I don't think there is anymore because you can just do this. If you're the you know, sales executive, you can just do that. So I think there's a large group of administrators, operators who are on the cusp of just being 10 to 1. So if you had 10 of these people who are

Starting point is 00:20:07 operators in a 100-person company, which is what I would say, like 10% of people are in operations, I think they could just go down to one, maybe two if you wanted redundancy and people to answer requests over the week. Yeah. Don't call. Okay, juggling multiple devices and apps to run your business is a mess. We all know that. Open phone is here to make that simple. I have an open phone number. I use it to communicate with founders. And it really works for me. I have a desktop app. Boom. I can go in there. I can do voiceover IP in this beautiful elegant app right on my phone or on my desktop. I use it a lot on my desktop. I'm being totally honest because I have my headset on. And sometimes founders want to do a call. They don't want to pop open a video conference. My sales team loves it. Why? Because they get to keep their

Starting point is 00:20:44 private phone number for their private phone number. And then they have all of their business stuff track in one location. So if they need to make a phone call to somebody they talked to last week, they can see their call history, click on it, send a text message and start a phone call with that person. And the ops team uses it. When we have people calling, they have questions about their investments, LPs, etc. We have a round robin phone number. So it will forward the call to two or three people on our team because we like to have really good customer support. And all this is easy to do with open phone. It's super affordable at just $13 a month. But Twist listeners get an extra 20% off of any plan for the first six months at openphone.com slash twist. What if you have an existing phone number with another service?

Starting point is 00:21:22 No problem. Easy peasy. Lemon squeeasy. Open phone is going to port them over at no extra cost. So head over to openphone.com slash twist. Start your free trial. Get 20% off. You're going to love this product. It is so affordable and it's so elegant. Just to the product team at open phone. Great job. I look at products all day long and yours is elegant and simple and powerful. Well done. One really good AI native. You know how we used to say like internet native and then cloud Mobile native, cloud native. Mobile natives. AI native, yeah.

Starting point is 00:21:51 AI native. That just thinks the first thing they do is go here and try to, you know, get it done. I just want to give clouds on it in A plus. Yeah. A lot. Great. And I want to give 4-0 an A-plus officially because we've been off for a month. You've been busy fundraising.

Starting point is 00:22:05 Congrats on getting that done, hopefully. And so I just want to get both these A-pluses. These are ready for prime time. And if you are listening to this and you're wondering why you're not getting a job, you were previously considered a very valid. valuable contributor, and now you're wondering why you got laid off and why you're not finding a job, it's because the stakes have gone higher. The stakes have just gotten higher. You have to provide more value. And what is the more value? Well, what can't this do? I describe what it can't do.

Starting point is 00:22:31 It can't post it to Indeed, and it can't sort through the resumes, et cetera. And so every company right now is just figuring out, you know, I have 10 people in ops. I need four. I have four people in ops. I need two. I have two. I need one. Whatever it is. Or twice as much productivity and I may have to switch some people up. Yeah, and so this is just what's happening at startups and big companies. And big companies, I think, are not AI natives, but they will be soon. And so if you work at a big company and you see me do something like this and your HR department's the same size, you probably need, and listen, I don't want anybody to lose their jobs, but those people need to,

Starting point is 00:23:05 you know, half the people need to be cut loose and reassigned to something more productive in your org. Maybe it's sales, maybe it's marketing. There might be some other place. But I suppose this level of efficiency is coming to everything. And in sales, unlike operations, being more effective just means you increase your sales per person. And can I add one thing, Jake, on this. Maybe just so, yeah, I think we both give them A pluses. I'm on A plus on both as well.

Starting point is 00:23:27 I think really well done, more and more powerful by the day. You know, the other thing that's happening is the model rollouts are just happening on a continuous basis now, right? And so you just have to start using the tools and maybe use more than one of them, and you'll see the advantages that come from it. The other thing that, you know, is related to, you know, so you have to change your behavior. you have to become an AI native. If you just did sort of what J-Cal was asking for and dropped it in Google,

Starting point is 00:23:51 look, and I have generative search. And able, look at how weak this result set is now. Like, what is going on? Like, you know, in fact, I feel like Google's going backwards here. I mean, it's giving you the lengths that Sonnet and Foro ingested very quickly. I mean, the speed is also the really, you know, a very important Google, I said, you know. If you make things faster, consumption goes up and usage goes up.

Starting point is 00:24:18 Here we go. These things are going really fast. I do think ZipRecruiter and Glassdoor in these places have to block the crawl. Have to block chat, GPD. Unless there's a relationship. Unless they pay per citation, per query, for whatever. Because these things are going out to the web. And there is no reason to click through to ZipRecruiter or Indeed or Glassdoor

Starting point is 00:24:42 and go to their website and get bombarded with. ads, pop-ups, whatever it happens to be, calls to action, et cetera. Just since we're here, Jacob, this wasn't even part of the plan, but I'm going to drop it in Gemini Advanced as well. Okay. Gemini Advance. Yeah. So Google's Gemini Advanced.

Starting point is 00:25:00 Okay. It didn't do the table. Oh, no, it did do the table. It did down here. Yeah. Oh, and let you export to sheets, which is, I remember that feature. Such a great feature. Okay.

Starting point is 00:25:07 I mean, it looks comparable. I mean, they're all comparable. I started also doing this with product searches. So I was looking for a new portable speaker. I was just curious, like, what the highest rate wants? And I asked that specifically, tell me what's on wirecutter. Oh, really? Yeah.

Starting point is 00:25:23 Oh, okay. So go ahead and do this. What are the best portable speakers? Please cite wirecutter, put it in a table with a link to Amazon. Give me up to 10. Now, what's interesting about this is the way wirecutter and other people who do the review sites make money is they get an affiliate link. And so here, it didn't put the links in.

Starting point is 00:25:45 But if you do that, if you cut and paste this and you put the same one into chat GPT40 or Omni, it gave me an Amazon link, obviously without the wirecutters thing. Now what you're going to see is like the wire cutters behind a payroll. So how does it know? Not always. Not always. Sometimes it's not. Yeah. Because I do this all the time.

Starting point is 00:26:05 Oh, here you go. And there's the Amazon link. Okay. And I think you could put it in here. give me, when you do the follow-up, just say, give me the same table with a quote from wirecutter about each one. Yeah, modify the table with a quote from wire cutter for each product. I hope somebody from the New York Times is watching this because here's the wire cutter quote. And by the way, this is a much better presentation. I don't know, but it's really good.

Starting point is 00:26:35 I mean, I don't know. I've not done this, but I'm going to start doing it because I'm, I always do whatever, my, my, my, go to is this search, but like on, you know, on Google. And then I basically end up on wirecutter, right? Like, that's it. So I use Google as navigation in those cases. So, you know, if you have other rating sites, like I also like PC magazine, they have a lab or something like that. So you can start to say put a quote from PC magazine, put a quote from here, put a quote

Starting point is 00:27:00 from here. I put a personal assistant does. So I did this. And I'm like, well, why am I getting a personal assistant again? And that it's for real world stuff, right? So. If it does this, then we know it's a loosening. though. You've got to be careful.

Starting point is 00:27:12 But let's see. And they had a quote from PCMET for each product. Yeah, there you go. It says I can't do it. There you go. It says restrictions. Oh, so, wow, whoa, that's the first I'm seeing it. It seems that I am unable to retrieve the PC MAG reviews directly due to

Starting point is 00:27:25 restrictions on their website. However, I can't provide general speakers based on available reviews and wear it that's interesting. Oh, wait. General PCMag insights, it did it anyway. No, I know. But I think that's like it's hallucinating a little bit. Like, yeah.

Starting point is 00:27:38 Interesting. Well, we'd have to cross-reference this. But again, you know, you start thinking about where you're going to start your journey. What do we use Google for? I use Google to find restaurants. I use it to buy products. I use it to hire people, you know, and this is a better experience.

Starting point is 00:27:53 The results here are better. And they give me a shout out, but I do think if you're PC magazine or wirecutter, you need to say, if you use any of our data, you need to pay us. And if you are, in fact, you know, in the lawsuit from the New York Times and Open AI, wirecutters, why I pay for New York Times. It's like wirecutter, then New York Times in order. I am no longer visiting the wirecutter website. So if you want a piece of evidence in your lawsuit, clip this and submit it in your briefs.

Starting point is 00:28:23 That is super wild. Hey, startups, does your product interface lift? Well, a lot of companies get up and running quickly, and that's part of the idea here in Silicon Valley and in startup land. But, hey, the design suffers. Maybe you got flawed UI, the U.S. decisions were made in a rush. And that can be the difference between product market fit and not getting product market fit or weak product market fit, medium product market fit, or strong product market fit. Well, here's the good news. Dev Squad's asynchronous design squad is here to help you. Our long-term partner, Dev Squad, built this service specifically for startups like ours.

Starting point is 00:29:00 That means you don't have to worry about winging it yourself and you don't have to buy a subscription to an unlimited design service and have them under-delivered. No, thanks to Def Squad, you can buy hours from expert designers who will handle everything for you, from UX audits to very sophisticated UI design. And the result is going to be an improved user experience and ultimately happier customers in a more successful business. So if you need to map your user flows, maybe you need to craft some interactive prototypes, or maybe you need to conduct a rigorous usability test, DefSquod has you covered. So if you are ready to transform your startup's UI without the painful overhead of having to hire an in-house design team, just head to devsquad.com slash startups and book a call. That's right. DevSquad.com slash startups. Mention you heard about the service here on this week in startups, you get 10% off.

Starting point is 00:29:55 I mean, I was not expecting that. We use these things all the time. So I really like this. I got to say, that is a way to. And this is what being AI. native is about. I think AI Native is, you know, the name of this episode. Yeah, you've become AI Native. When I'm in the car with my daughters now, they ask me questions and I told you I put my quick key on my iPhone to be the chat GPT4 dialogue where you talk to it, the talk interface, and that's actually getting better and better. Yeah, getting better and better. So you can do these similar type things. Tell me the best three sushi restaurants near me and what their top dishes are. And that will work, you know. Tell me the top three sushi.

Starting point is 00:30:36 restaurants in San Francisco and what their top two signature dishes are. You're the top three sushi restaurants in San Francisco and their top two signature dishes. One, Waco, squid with Shiso and Lemon. This dish features a delicate combination of squid, shiso leaf, lemon, offering a refreshing and balanced flavor. You know, if you're driving in a car and you can't look at your phone, this is pretty amazing. And then when this has a hookup to make the reservation amazing, it really is game over.

Starting point is 00:31:06 Sometimes with these changes in technology, we say, there's like a tipping point, as we've talked about in the world. And when a tipping point gets hit, I think people are going to collectively realize, wait a second, I'm not an AI native. If I am an AI native, I'm just going to get a lot more done, a lot faster, and then you're going super competitive. And I just look at what my startups are doing when they are planning an event or retreat, a offside, you know, a new product, a press release. It's just going faster and faster and faster. So velocity is going up. Yeah, definitely at liquidity. I could tell this year, Jason.

Starting point is 00:31:38 And obviously, your teams are always getting better because you're pushing them. But, like, it felt like there was more AI behind the scenes for organizing. Oh, yeah. A thousand percent. AI is starting to work a lot better for planning events. 100%.

Starting point is 00:31:51 Any key insights from that? I think if you were to just ask, like when I was doing the Allens Summit and we moved it to L.A. last year, you can pull up, uh, chat to B4O. I started asking you questions like, tell me arenas. and theater spaces that could have over,

Starting point is 00:32:08 you could have between 2,000 and 4,000 people and seats. And it was clujy, slow, and it would find me some, you had, you know, half the information was wrong, but it did find some places that I may not have thought of. So I would just ask it, what theaters have between 2,000 and 5,000 seats in the Los Angeles area for events?

Starting point is 00:32:33 And put it in a table and put the address. And you do some sort of interesting search like that. It didn't work, you know, 18 months ago when we were planning that one. It didn't work comprehensively. So I worked with a producer. And then I told them, hey, you didn't have these three places. And they're like, oh, yeah, I didn't think of that.

Starting point is 00:32:53 And like one of them was the YouTube theater. It turns out there's like a YouTube theater. And there's what used to be called the Nokia theater, then the Microsoft theater. And here, you know, you're starting to see links to the theaters for, more information, their address. And now if you ask, give me 20 theaters and put it in a table, it's going to give you 20. And it's going to find the YouTube theater and the other ones. And that was the comprehensiveness is moving up and the speed is moving up. So here you see it's searching the web. And it found 10 different websites, dude. Ten different websites at once and started

Starting point is 00:33:26 doing this. You know, it's got the same ones so far. But there's the Adobe Theater I mentioned. There's the Hollywood Playtium. There's a shrine. And there's the Greek theater. Disney Concert Hall, El Capiton, the Regent Theater. Man, I mean, this is incredible. This is more than a morning. You'd give someone this task and they'd come to you a couple days later. Yeah. Yeah, this is a week project.

Starting point is 00:33:48 Now, if you said, add a column with the price to rent the theater. Now, that's where, you know, sometimes they have the information on the website. Sometimes it's third party websites that have it. This is where you start getting into the job of, you know, an event producer. And so, you know, we're going to add a price here to rent this. that would be the next thing your client would ask if you were. And, you know, it couldn't come up with anything. And here, we'll see if these are, you know, if these are actually real.

Starting point is 00:34:14 You have a good sense, right? I think the Dolby Theater might be more expensive than $5,000 an hour. But anyways. Yeah. I mean, these prices are probably wrong. So, you know, and that's because a lot of times these prices aren't available. So it's probably estimating here. So what you'd want to say is cite the source.

Starting point is 00:34:29 If you can cite a source of the cost per day to rent this, let me know. Or you could say, get me the email of, the contact person or find me a person on LinkedIn, right? Remember, we're trying to do LinkedIn searches? Like, these are all the next steps. Yeah. And so I don't know if there are somebody at OpenAI looking at the search stream and saying, hey, solve this next.

Starting point is 00:34:53 Or the AI is just watching people do it and trying to figure it out or it's ingesting more information. But brave new world here, right? Well, I think to answer your question, I, I think what the focus really is, and we're going to see more of this on the back half of this year, is we're going to see a greater focus on agentic type use cases, where I think the first aha, when this came out,

Starting point is 00:35:20 was like, oh, my God, I can just type something comes out. Now it's like, hey, go do a bunch of stuff for me and come back. I think that's where we're going to see the biggest improvements. And when we think of the next generation of these coming out, and it's like we did here, all of this didn't, you know, we're doing what's called multi-shot, right? We're working through our problem, but we're going to see the agents just kind of work through it on their own.

Starting point is 00:35:41 And you saw that in the very first iteration. People were trying that. Remember with auto-GPTs? We'd be like, yeah, come up with like a work plan and then try to go fill it out. I think what we're going to see now is it's going to be able to do that just for a question you ask. And it will be able to sort of figure out, hey, I should probably put this in a table and figure out prices and do that. Because you can literally farm the task off to a bunch of different agents and come back and, you know, aggregate the results.

Starting point is 00:36:05 Yeah. So the ability to do the same search against five different things, five different language models and have it then build one table from five different models, some master one do it, or say, you know, what other questions should I know about renting a space? And it's like, well, you need an AV company and, you know, you're going to need sound and you're going to need food and, you know, um, these are the basics. Okay, let's go through it. You have more.

Starting point is 00:36:31 I do. And I know there were a lot of image stuff that came out. So I'm very interested in the image stuff. I don't know if you saw those trending and if you have it on your docket, but somebody was making, they were taking famous memes and then bringing them to life on Twitter. Oh, yeah, yeah. I mean, I have that one in Twitter. I didn't demo it, but like I can show you who did it.

Starting point is 00:36:51 Let's pull it up. That was a fun one. I have a demo slated with these guys. I didn't do the meme real thing. I have them pulled up for my next demo, which is awesome. It's a good call-up. You always do that. You have this super power.

Starting point is 00:37:02 Is it like Phil Helmuth? He knows how to read your cards and he'll be like, oh, you had this. But you want to range of hands. Yeah. So here it is. This video is created using Luma AI. Yeah. So we have them queued up for the next demo.

Starting point is 00:37:14 But we'll just start here. So it's just, this is the start of the video. Yeah. Okay. There's the guy with the beard, the bear guy. There's Tiano. Eating a bagel or something. Oh, this is good.

Starting point is 00:37:29 The girlfriend and the whistle at the, yeah. Oh, the. there's the girl with the fire behind her there's the crying guy from some TV show on the night at Dawson's Great here's Charlie in the chocolate factory I know that one yeah oh here's the guy thinking guy

Starting point is 00:37:48 yeah I love that way that's his temple okay here's space balls blue Skywalker meme I didn't know that was a meme Rick are you just looking at things in the office and saying that you love them Describe what Marcellus Wallace There you go

Starting point is 00:38:07 Pulpiction Incredible Oh you were the chosen one Was it Or her Yeah Oh yeah that girl The little girl

Starting point is 00:38:19 With the crazy look And actually putting on glasses I mean it's nuts And this is Like All right we get it Oh yeah This Putin

Starting point is 00:38:29 Kim Jong-un Driving together Oh wow I didn't watch it all the way through I didn't watch it all over there. I didn't know they did modern ones. Oh, and they're having them do the drop-off. I wonder if they told it to make that story about memes.

Starting point is 00:38:43 Oh, they Rick-roll this at the end. So there's the, like, a guy at the Celtics game who's just... Up there's a dog. Shima, Enu, whatever's called. And then finishes with him. Back to the kid. Wow. Wow. I mean, pretty impressive.

Starting point is 00:39:02 Yeah. Gonna have to go through our bet list. I think I'm going to owe you a lot. lot of money. You always took the short end of these. I'm a technologist, right? So I always think, you know, slow than fast. So that was Luma Labs that did those.

Starting point is 00:39:19 I did this one earlier because it was taking some time to do them. And I said a panda riding a bicycle through New York City. And these things are just incredible now, right, in terms of their capabilities and what they're able to do. It's just mind-blowing. I mean, I- It hasn't crossed the uncanny valley for me. It still looks like it's AI generated.

Starting point is 00:39:44 I would not have been fooled by either of these. I would have said, oh, that's cool. AI is getting better. I think they would have to put these into post-production to fool me. And I think so, too. And I think the framework we should use is just enhancement, right? What I would say, J-Col is, you know, let's say you're trying to produce something and you need some screens or some storyboard.

Starting point is 00:40:03 Storyboards is what they're called, yeah. We're in the storyboard era. And I think it's not that far away. I think when we see the really, really highly produced stuff, well, basically, Luma was one and we'll just rate two back to back. I had another one queued up here as well. This is a Chinese company, and they really blew some people's minds away over the last couple of weeks.

Starting point is 00:40:25 And it's called Kling. Did you see this one? No. Show me. K-L-I-N-G. Yeah, the video that these guys had. You could just see some of these. Wow.

Starting point is 00:40:35 Yeah. Look those flowers. Here's a panda playing a guitar. Yeah. I mean, so close to the Uncanny Valley. This one of the Parrot, it did cross the Uncanny Valley for me. I wouldn't have known that was fake. Yeah.

Starting point is 00:40:49 Obviously a rabbit drinking coffee. Even this one. I think if you were making a product. Yeah, that's a flat white in a quartado glass, you know, like the glass glass. There's a cortado. And that's one of my favorite beverages of coffee beverages. And that looks like it's from a stock image. Now, what company says, cling is the name of it?

Starting point is 00:41:06 Now, you can't use it because you have to download an app and you have to be in the Chinese app store to do it. So I can only demo the screen here. But I think the point on these, maybe the next episode, we'll review the bets is I think some of these have really crossed. Now, I do think these require a lot of tuning. So it's like being in mid-journey and creating things. And so in order to get there, you really have to do,

Starting point is 00:41:31 you have to kind of do many, many iterations of it. Yeah. But I do think it's definitely crossed sort of the point of convincing us that, you know, these are real. I think we've crossed the Uncanny Valley, but. Fantastic. What do you give those two? I give those B pluses. Okay. Yeah. And what do they need to be better? I'm grading them on crossing the Uncanny Valley. So I don't know it's AI created. So if you were saying, judge it on it being a storyboard tool, like a creative tool to give you ideas. I give it an A. I give it a solid A. Now, if you said, you know, to actually make production videos, not storyboards, I give

Starting point is 00:42:12 it a B, B, B plus. I think they're six months away from crossing the Uncanny Valley. Like, even the coffee one, if you said, look at it and tell me AI or not AI, I probably would have got AI from just some of the artifacts. The parrot, maybe not. So I felt like out of the maybe five or six, we saw, it's not going to get me the majority already at the time. Okay. Yeah, I'm going to give them an A minus because they're going to help you win bets against you so they can get to A plus. Okay, great. Awesome. So you're just, yeah,

Starting point is 00:42:41 the fix is in. Got it. You're trying to motivate them. Yeah, exactly. Guys, get an A plus. Let's, let's win some money here. Whatever bet you're saying, I'm taking the under. Whatever you're proposing going forward, I'm going to get chelot here. No, no, no. What I was going to suggest is we do like a little bakeoff where we do it, real image and AI image. And you have to pick. We'll do that. How about next episode we do that? Yes. We'll just do it for a $100 a pop. It would be just flips. Just flips.

Starting point is 00:43:07 Just flips. AI flips. Deal. Next episode, we'll do in two weeks, AI flips. Get two full weeks to do it. Oh, that is so fun. I think it would be good. Producers get on it.

Starting point is 00:43:17 Make us one quartado from a stock image library, get one of these ones. You know, take the watermarks off and let's see. Let's see if we can do flips. Now, the last one I want to do here is just actually that there's two things I want to do really quickly. Just one I want to let me talk about open source. So we'll do a quick lightning round and a few things that have also happened since Sunny's been out. So a couple of models have come.

Starting point is 00:43:42 And this Quintu, it's this team out of Alibaba. People that are not tracking, this is probably right now the best open source model. And so it's come out. It's really powerful. It's fully open source. They have a lot of different sizes ranging from 500 million all the way to 72 billion, train on a bunch of different languages and a really big context, like

Starting point is 00:44:05 128K tokens. You can put a lot in. So this is available. You can try it out at Hugging Face if you want to. They have a version here that's running. Okay. So this is open source. Open source from the Alibaba team. And this is really interesting

Starting point is 00:44:21 because what we're seeing is these teams in China are pushing for these open source models and they're coming to the top of the leaderboard. Interesting. Yeah. There's a great tradition of not respecting IP and just wholesale copying stuff. I'm curious in these codebases for new projects from China, are these built from scratch?

Starting point is 00:44:43 Or do you think these are, you know, inspired by common LLMs? Do you think these are, you know, backdoor hacks into proprietary stuff? I'm just curious. Well, I'm going to actually share something slightly different. I don't want to comment, you know, a lot about where they're getting the data from. because it's just not known. But I know one thing that's happening because I definitely heard

Starting point is 00:45:06 some of the real professionals in the space talk about it is they're using the current large language models that are available to generate synthetic data so that they can make their model better. So think about what makes a model really good, not so much is just like open wild data on the internet, but it's matched paired information. So it's like if we just took the stuff that we were doing,

Starting point is 00:45:31 which is like a question and an answer and then mapped it all. And you gave that as training data. It's really, really high quality data. Interesting. So that search we did on, you know, salaries or speakers. Then you say, get me some more data similar to this and then you feed it in. So it's kind of meta. Yeah, and there's also the chance in China that, you know, they would just rip the entire New York Times archive, the entire magazine archive from somewhere where other people in the United States now would respect that IP. And we just saw the lawsuits against two of the music companies that you and I played with last year. Yep. Yep.

Starting point is 00:46:09 Like those companies are gone, by the way. I'll just tell you right now. I don't think there's- Are you calling it? I'm calling it right now. Both of those companies are going to- Napster. You're calling to Napster.

Starting point is 00:46:17 I'm just saying, it's straight up Napster. There's two of them, I think, that got sued. I think they're both going to be Napster Roadkill, or they will have a $50 million fine against them, and they will have to work it off. and then I don't know who's paying for that $50 million fine because they're going to need a perpetual license to the music. So they'll be fined $50 million, $25 million for what they've done already. Then they're going to get charged on top of that, right, for a future use.

Starting point is 00:46:43 And they don't have enough revenue to make that work straight up. So I think there are anybody who invested in those companies, and I don't know if you saw, they said in their training data, got the name of the company. There's Suno. There was two. There's Suno. It was the one. And I think Suno is saying they can't tell them.

Starting point is 00:46:59 how they train their music. It's proprietary in discovery, which means they're screwed. But in China, they'll just take every song ever written and they'll make a better open source LLM and then what is the music industry going to do? They're going to have to do an injunction. This is what I predict will happen. This is as crazy as it's going to be. They're going to do an injunction, the music industry because they are the most powerful

Starting point is 00:47:22 and organized, correct. More than anyone because they have the RI-Double A, right, reporting industry. Association of America. And then you have the songwriters, and then you have the live stuff, you have the labels. I mean, you're going to get it from all angles. And you're like just trying to understand how many different ways you're going to get sued requires a major legal team. Because they're going to come at you from all ends. So what will happen is a Chinese or whatever firm that, you know, is operating in an area that doesn't have restrictions.

Starting point is 00:47:53 Or even Israel or India or Pakistan with LinkedIn data, right, with sprays data. You're going to have an LLM come. out of LinkedIn and other, you know, Facebook data, Instagram data. And you're going to be able to use that L.M to search and say, hey, I'm looking for CTOs at this thing and it's going to just spit it out, like things you can't do now. And that's going to then wind up on Hugging Face as an open source project.

Starting point is 00:48:15 And the weights and everything and the data. And then they'll do it for the music industry. They'll do it with every movie ever made. And then those language models are going to get sued. And GitHub, Hugging Face are going to get sued if they publish them. So just think about that. If you host them, they're going to say you're contributing. You know that this is built off stolen data.

Starting point is 00:48:34 They're going to get sued. Now, I don't know if that lawsuit works or not or hugging phase hosting an open source project, but I could see that happening. That's how much is that state. You paint a very doomed picture for a wonderful place. We've got to find a better compromise, so, J-Cal. We can't just shut innovation down either. Yeah, it's a very simple one.

Starting point is 00:48:51 Pay the man. Pay the man. Pay the man his money. Pay the man. Pay the man his money. Yes, exactly. It's like we did to OG Ananoi at the Knicks. We paid them and his money.

Starting point is 00:49:02 Yeah. All right. Last one. I'm really excited by this. So, you know, there's been a lot of energy and notion around like large language models. But there's been, there's a lot of innovation happening in the space of voice. And Cartesian AI, what they're doing is, so they do something called state space models. And specifically space models.

Starting point is 00:49:22 Okay. Yes. And so this is like in English. It's basically a model that. uses a different area for the search base, which is not necessarily language, right? Okay. And I'll do a more detailed explanation of it

Starting point is 00:49:34 because I have to get up to speed. I don't think I'm fully there 100%. Well, cut it. And so, yeah. It shows the product. Yeah, okay, of course. And so what they've done here is they've created these voices using this model.

Starting point is 00:49:47 And these voices are incredible. So let me just make sure the audio comes. So this is a 1920s radio man. And I know you look to do this voice. Ladies and gentlemen. Ladies and gentlemen. Gather around the wire. as we delve into a world of speakeasy.

Starting point is 00:49:59 The tragedy of the Hindenbird. Roaring excitement. Yeah. I mean, it's literally the classic. And I know you'd love this, right? Hey there. Have a seat and relax. You're in for the best haircut and some great conversation.

Starting point is 00:50:11 Oh, ASMR barbershop guy. Yeah, barbershop guy. I feel like he's about to slip my throat. Yeah, like, you know, here's the Indian man. Every journey starts with a single step. I'm not doing my Indian accent. I'm not getting canceled. You let's try, sonny.

Starting point is 00:50:24 I will not do my Indian accent. Yeah. But they've done an incredible jar. You can do yours. You're in Dan. You're allowed to. No, no, no. I'm not going to do it. I'm not going to do it. I don't do a good one. But the wizard. How about this is the last one? With a flick of my wand and a whisper of incantations. And the stras and revel and magic fills the air. All we have to do is decide to do at the time that's been given to us, subteep. Yes. It's Ian McCallon. Yes, it is. Exactly. Exactly. But do they have one for her and for, Scarlet Jim Vincent in here.

Starting point is 00:51:00 I don't, they don't. I don't see one here. But no, what's really happening here is like what, you know, there's so much innovation happening now, J-Cal, and maybe in one of the next episodes we'll just do a, that's what, you know, we talked about it a little bit in liquidity, but, or at liquidity, there is a lot of innovation happening now.

Starting point is 00:51:18 And it's faster than ever. And it's coming in the, in the way of new types of models, open source models, agentic reasoning, in voice. And so one thing that I want to go into is in the second half of this year, J-Cal, we are going to have some surprises that no one was anticipated. Okay, there you go. That's just a, that's something that you know running developer.brock.com or that's something,

Starting point is 00:51:43 so that's inside information. That's the inside line. No, I don't know. It's not an indication. I'm just saying like being in the thick of it, you know, seeing how everyone is innovating, I have a very, very strong. And I don't know exactly what it'll be. I mean, I have insights in a few.

Starting point is 00:51:57 places that I can't share, but. Okay. But beyond, like, just sort of generally, I'm telling you, we're going to be more surprised in the back half of this year than we were when opening high first came out. Oh, it's a big claim. It's a big claim. I don't know how we phrase it as a bet, but what you're saying is everything that happened up to now is the epilogue and that we're going to really see the story's coming.

Starting point is 00:52:17 The story is coming now. Okay. There you have it, folks. Stick with us. X.com slash Sundeep. X.com slash Jason. this is this week in startups, rate, subscribe, write a review,

Starting point is 00:52:30 and then make sure you tell Sonny how much to love him and that you missed him for a month so that he is no longer MIA on the pod. We need him here doing these demos. See you all next time. Bye bye.

This Week in Startups - AI Demos: Sunny’s Back with Luma Labs, Kling, Claude Sonnet & Getting AI Native | E1976

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