This Week in Startups - GPT-4 web browsing, plugin demos, AI layoffs + more with Sunny Madra & Vinny Lingham | E1737

Episode Date: May 9, 2023

Vinny and Sunny join Jason for another AI roundtable to demo ChatGPT’s web browsing capabilities (1:44), discuss AI’s effect on productivity and the workforce (13:53), demo Opus Clip and ChatGPT...’s Zillow plugin (33:44), and more! (0:00) Vinny and Sunny join Jason (1:44) GPT-4 web browsing demo (9:10) Inside.com’s AI strategy (12:22) LinkedIn Marketing - Get a $100 LinkedIn ad credit at https://linkedin.com/thisweekinstartups (13:53) AI’s effect on productivity and the workforce (23:36) Hyperice - Get $50 off your order of $150 or more with code TWIST50 at http://hyperice.com/ (25:04) Personal use cases for ChatGPT (30:56) Market Pull (33:44) Opus Clip demo (39:55) Trovata - Use code TWIST at https://trovata.io/twist for 30% off one year of premium features, like AI forecasting (41:08) Synthesia (43:11) Zillow plugin demo  (49:51) The future of ChatGPT Plugins FOLLOW Sunny: https://twitter.com/sundeep FOLLOW Vinny: https://twitter.com/vinnylingham FOLLOW Jason: https://linktr.ee/calacanis Subscribe to our YouTube to watch all full episodes: https://www.youtube.com/channel/UCkkhmBWfS7pILYIk0izkc3A?sub_confirmation=1 FOUNDERS! Subscribe to the Founder University podcast: https://podcasts.apple.com/au/podcast/founder-university/id1648407190

Transcript
Discussion (0)
Starting point is 00:00:00 It's such a good product that it has market pull. So this is an important lesson for founders. You can put social features into a product like multiplayer mode to make it grow. And you can imagine multiplayer mode is going to be bonkers in chat GP4. All of the examples we have right now, you could have just added me to them, and then we could have each been asking questions at the same time with our own accounts and have it making forked threads of the conversation. So I take yours, and then I ask two questions, and it goes on a decision,
Starting point is 00:00:30 tree. Vinnie has some, it goes out a different decision tree. And then we say, take this from Vinnie's and combine it with Sonny's, right? And we could be working in one chat stream with three employees or 20 employees and two or three different AIs. But yeah, I understand his point. But what he has to understand is this is such a transcendent product that you can't shut up about it if you use it. This weekend startups is brought to you by LinkedIn Marketing to redeem a free $100. LinkedIn ad credit and launch your first campaign, go to LinkedIn.com slash this week in startups. Hyperice. Warm up and recover faster with the HyperVolt 2 massage gun and NormaTech compression therapy boots.
Starting point is 00:01:16 Save $50 off your order of $150 or more with code twist 50 at checkout at hyperice.com. And Trovada. Starting up is hard. Trowada makes managing cash easy. Start automating your cash management at trovata.io slash twist. Use code Twist for 30% off, one full year of premium features like AI forecasting. All right, everybody, welcome back to this week in startups. We're doing our AI roundtable.
Starting point is 00:01:48 Everybody's obsessed in Silicon Valley. We can't stop talking about it. We can't stop playing with it. AI is changing everything. This is not a drill. We are seeing advances every day, every week, every hour. It's got to be hundreds of thousands of developers have now pivoted to focusing on this. I've got hundreds of entrepreneurs in our portfolio and contacting me weekly with incredible ideas.
Starting point is 00:02:15 So we are doing our weekly AI roundup with Sunny, Madra, and Vinnie Lingam every week now. Because these two individuals are playing with it. They're lifelong technologists. So Sonny and Vinny, welcome back to the program. Thanks, to be back. Good to be on. We better do it weekly just to keep up with the pace.
Starting point is 00:02:33 I mean, it's crazy how much feedback I'm getting from this thing all over. Like, people are pinging me their frightened center now. So I mean, I think a lot of people watching this. And I mean, you tell me how popular is this amongst all your segments? Well, I think it's becoming the most popular, if I'm being honest.
Starting point is 00:02:48 And the reason is, because we're actually bringing demos and we're breaking things down a very tactical basis. The conversation of like philosophy and mission and 30,000 feet is it's burnt out. Everybody's kind of talking about AI safety and, you know, the business models and Microsoft versus Google. I think all of that is not as important as just keeping up with what did people build in the past week.
Starting point is 00:03:15 So that's what we're going to try to do here is we're going to try to talk. In today's demo, about 10 days ago, Open AI released a new model. It's called web browsing with GPT4. It's still in alpha. It sucks. I've been using it. It breaks constantly, but you can start to see how promising it is.
Starting point is 00:03:32 Now, I do think you have to be invited to this. It's for people who have what's called chat GPT plugins. We've talked about plugins here before. Plugins is a framework. I think you explained this perfectly, Sonny previously. ChatGTV4 is the reasoning engine
Starting point is 00:03:50 and the data sets, the models, the things that you're going to pull information from on the plugins. They had web browsing, correct me from wrong here, Sonny, with GPT35, but now you put four with it. It's working a lot better. And I think, Vinny, you have some demos you're going to do later in the show, but maybe we'll start with you, Sonny. Explain to us why chat GPT4 with web browsing is so important and what it's going to do for people. Yeah.
Starting point is 00:04:18 So, you know, when this first came out, one of the major limited. was the response that, you know, my data set is only accurate as of September 2021. And that creates a lot of limitations in terms of using the technology for, you know, something that's quite temporal of, you know, right now. And so with this release of the web browsing capability, there's many tasks that we conduct on a daily basis that we're conducting all the time that we can leverage a tool for. We'll see some limitations here, but we'll go through a couple of different examples that across some different verticals, and we can show the power of what the tool can be used for. And I actually did one just before for a friend that email would be looking for a job description.
Starting point is 00:05:08 So maybe should we just dive in and start that one? Yeah, great. And then if you're not watching this, go to YouTube.com slash this weekend or go to Spotify to watch this week and start up. or do a search for this week in startup's video on iTunes, and you'll find the video version of this. And of course, sports cast it, Sunny, as you know, where you describe what we're seeing. So my good friends run a software development company called Laser Technologies,
Starting point is 00:05:35 and they sent me a message in a WhatsApp group saying, hey, you know, could you help with a job description for a startup and VC ecosystem lead? And I kind of did the thing, what we're all doing now, which we all should be doing, I just typed it in. So my prompt for those listening is, can you write a job description for a startup and VC ecosystem lead to help at Laser? And then I just gave ChatGPT the link to their website.
Starting point is 00:06:01 That's about it. And you can see here, and I'm not going to read the whole job description, but it comes up with a job description, it comes up with a set, which describes the company in its background, it comes up with a set of responsibilities, and it comes up with a set of requirements. And it even closes with the all-important laser technologies is committed to creating a diverse and inclusive environment.
Starting point is 00:06:22 We encourage applications from all qualified individuals. It went woke. It went, well, I don't know if that's woke. I think that's, no, that's best practices. Exactly. Yeah, yeah,
Starting point is 00:06:30 that's what I was saying. I don't think it's woke. But so here's a use case where it, you know, went out and looked at their latest website and pulled those things in. I'll go to kind of a different one right now as well. Here's one where, you know, you can have it look at a particular piece of code. And in this case, I asked it, and I took this example from Rowan, I want to give a shout out to. He has a great, great Twitter account and newsletter, and I think we're going to pull it up
Starting point is 00:06:59 in a bit here as well. And so he kind of inspired this one. So I said, what are the five areas of potential improvement of the Twitter algorithm? And I gave it the link, the GitHub link to the new Twitter algorithm, and it basically came up with a response. And I'll just read part of it and says, based on the GitHub repository you provided, I was able to find several components of Twitter's algorithm, including the unified user actions and user signal service. And it explains what those are. And then it found some, it found a detailed proposal by a user on how to change those things.
Starting point is 00:07:32 And so I think this is really cool. And that this is where we saw some limitations. And this is where you're kind of seeing the combination of being access to the web and its reasoning engine and looking at additional details around what's happening where you pointed to come up with something interesting. Maybe my last example here today is, you know, I feel like this is part of what Inside does, Jason, and this may change how Inside thinks of things. And this one, it didn't, it didn't complete the task completely. And again, I'll read the prompt, which is, what are the 10 most interesting text stories from the last week? Put the results in a table and write a short summary and link to the original story. So it doesn't pull off the last bit, but it does find some interesting stories, which is, I think, a start. point. I have a little bit of a suspicion as why this is not doing a good job. And it's, I'm going to
Starting point is 00:08:21 just click in, down in the, the link here where it shows where it's right down. There's a drop down that shows, just to be clear here. Yep. What Google searches, or Bing searches, I'm assuming, it went to and then it clicked on. So it says, hey, I clicked on this. I'm reading the content. I did a search tech news, March, May 1st, to May. So you can actually see what searches it's doing. So it's kind of behind the scenes, you know, and it really does a really nice job. Yeah, and it does a nice job. Now, this, you know, maybe the good news for the folks at Inside, it hasn't replaced them completely yet because this answer is not sufficient.
Starting point is 00:08:59 They're definitely more than three stories and it should have been able to complete it. But like we've been talking about, I think this is 90 days away from answering this question properly. And so, well, interesting, you bring this up. I literally had a board meeting for inside.com last week, and I presented our AI vision. I'm looking to hire two full-time AI developers to work at Inside to work on this exact problem. So what we're doing, just broad strokes, is we're going to take a newsletter like Inside.com slash podcasting. Take the last 100 or 200 email newsletters, put that into a language model, and then have all the sources, and then have the curation going and looking for sources done by AI, have a first draft summary.
Starting point is 00:09:41 put onto the website and then just say, this is the inside AI bot that put it at inside.com slash podcasting. And then our editors, our writers, will look at the 10 or 20 that come in from the curator,
Starting point is 00:09:53 the AI curator, and edit them and check the facts with a big disclaimer on it. And then what that will do for the writers is they no longer have to curate or write. Maybe 70% of the writing will be done for them
Starting point is 00:10:08 and 95% of the curation they can redeploy that time. to then go call subjects and ask follow-up questions. And so that's our sort of concept is they can do other original acts of journalism. But man, this is a really interesting turn of events. I just want to show you something that I just did here. Pull this one up, Nick.
Starting point is 00:10:32 I just asked with the open web browser, not to rub salt in anybody's wounds, who's speaking at Calicanus's latest conference? and I didn't mention the latest conference is the Angel Summit that's happening in Napa, June 5th and 6th and 7th. And it went out
Starting point is 00:10:49 and I'm just sharing the chat GP, this is shared GPT, I don't know if you guys use this, but share GPT, let you take a result and share it on the open web. And it nailed it. And it made a nice bullet-pointed list.
Starting point is 00:11:00 And then I said, hey, can you hyperlink to the person's corporate page and to their name to their LinkedIn page? and let me see if it got Brad Gersoner correct. In chat GPT, shared GPT doesn't recognize hyperlinks or does it?
Starting point is 00:11:18 Are there hyperlinks on those? I couldn't see them. Did they work? Yes, the hyperlink work to his website. It'll open up to his website. Yeah. Okay,
Starting point is 00:11:27 but it didn't get, if you click on Brad's name, did it get you his proper LinkedIn or no? Yep, it did. Look at that. Loading LinkedIn, but no,
Starting point is 00:11:35 it didn't get that right. Oh, it didn't get it. I guessed. It seems like LinkedIn does not work with chat GPT well. I guess that's something that's going to have to be addressed. But man, did it get the corporate address as well?
Starting point is 00:11:48 And yeah, it linked a pitch book for Valor, maybe because Valor has a, you're not, maybe they came up higher in it. But this is really going to get interesting because then I could say, give me a short one sentence bio for each, speaker and put them in a table.
Starting point is 00:12:12 And you start thinking about how much work this would be. This is what you would be giving to a research assistant or an operations person in your company. Okay, when you're selling B2B solutions, you really want to get in front of decision makers, the ones who make the purchasing decisions. But the problem is these folks can be hard to find and target on most social networks, right?
Starting point is 00:12:34 People are talking about politics, they're sharing selfies, they're going cooking videos, all that kind of stuff. Well, you need to pick the right platform to reach these folks and the perfect platform is LinkedIn. Now, we know LinkedIn has 875 million members, but did you know 180 million of those are the senior level executives who okay purchases? And there's 10 million C level executives as part of that 180 million. And those are the people who really, really can push through a decision to buy your product or services. That's the purchasing power and it resides on LinkedIn. And no other platform in the world can offer you these kind of eyeballs. It's very special. LinkedIn is business. Business equals LinkedIn.
Starting point is 00:13:15 LinkedIn equals business. You know this because we use LinkedIn for business all the time. We just had Reid Hoffman on the program. It's just amazing how focusing on one thing for decades can make such an incredible product like LinkedIn. So if you're selling a B2B solution, you're a consultant, you're making software, you got a marketplace, you want to engage with elite buyers and they're all on LinkedIn. So here's your simple call to action, make B2B marketing everything it can be and get a $100 credit towards your next campaign by going to LinkedIn.com slash this week in startups.
Starting point is 00:13:47 That's right, LinkedIn.com slash this week in startups terms and conditions do apply. And I try to explain to people exactly like how powerful this is. The number I keep getting to Vinny is 30% of people's work this year. I had Brian Chesky from Airbnb on the podcast last week. I just interviewed Reed Hoffman, and I also talked to Aaron Levy from Box because he's added much features. They asked them all to tell me what percentage of your team members work this year will be replaced by AI. I've been using the number 30% internally. Chesky said 20 to 30%, depending on the position.
Starting point is 00:14:26 Some positions might be 50%, some might be 10. Aaron Levy said, it could be 50% for customer support, but I'll go with 30%. and Reid Hoffman also said like, depending on the person, I think like, yeah, 20, 30, 40%. So we all came with one third. Vinnie, what say you? In your experience, you're building weight room, you're building other civic. What percentage of your own personal tasks are you offloading this year, do you think? And what is your team experiencing?
Starting point is 00:15:00 I mean, personally, I'm doing a lot using, um, using, um, using, you know, chat GPT at OpenAI a lot, like everything from health research, summarizing stuff, taking like large documents
Starting point is 00:15:13 that I just want to like, you know, get a quick summary out. It's saving me a ton of time when you try to read through, especially like, you know, reasonably complex things
Starting point is 00:15:21 where you have to have a lot of focus and tension. It's a big easy thing to get through it quickly. So it's helping me a lot just, you know, the summary, the summarization features are incredible. I posted a tweet and I'll put the,
Starting point is 00:15:35 Linkapia for Nick. I posted a tweet where I spoke about the economy and where I think things are going. But it actually related to AI for a bit. The AI part is like, I think we're going to get to this recession that everyone is talking about and companies are going to lay off people and we're going to get unemployment to over 6% by end of year. Because that's what Powell's trying to do, by the way. He keeps talking about the tight labor market.
Starting point is 00:16:00 He needs to loosen it up. Now, whether it's 5 or 6%, maybe he's going. goals five or four and a half five, he's going to overshoot it with, like, no doubt. There's going to be an overshooting of this. So we're going to go to 6% call it that. Now, here's the question. When he tries to get it back, is it going to work? When he tries to like lower rates, and my thesis here is that's not going to work.
Starting point is 00:16:23 Because what's happening is companies are laying off people and those people basically are going to be replaced by AI. And then those jobs are not coming back. This is a very interesting take. It's a very different. It's a very different take where people have thought. People thought that companies are going to replace workers with AI and lay off people because of AI.
Starting point is 00:16:45 No, when you have a team that gets reduced in size and now they have to make do with fewer people, they learn to become a lot more efficient. They learn to use the tools that are out there to replace the functions that their colleagues used to do. And then they don't need those colleagues anymore. So that's my thesis going into this. And running a startup, for example, we just, you know, we got rid of our QA, outsource QA, you know, function because we can get the unit test done through, you know,
Starting point is 00:17:13 open AI. Engineers can check their code. Like, we can make things far more faster, far more efficient. And I think that's going to, so the impact that AI isn't having the short term is on all the jobs which are kind of like, you know, a lot of this is outsource, by the way, to Eastern Europe and whatever else. So QA testing, copywriting, you know, like those sorts of jobs with AI can just do it for you. Web scraping, sales development reps, all this stuff has been human capital in Manila, Eastern Europe,
Starting point is 00:17:43 exactly, San Paulo, just sometimes it's English as a native language, sometimes English as a second language. But if you want $5 an hour to $15 an hour workers on demand, you can find it in those places. are going to get hit hard. Yeah, exactly. They're going to get hit brutally hard. Unless they're using these tools, which would then make them bionic. Well, no. Because why would you have to outsource to someone else?
Starting point is 00:18:09 Regardless, even if there's five guys doing it, now there's one. Regardless, whether you outsource that person or have them here, it's an 80% reduction. What do you say, sir? What do you say to this thesis that we're going to see the people who were laid off? This is the first time I've heard anybody say this. So we're going up to like a big macro picture here. I have stopped hiring at both companies. I have said, with the exception of the AI developers,
Starting point is 00:18:34 I've said, everybody learned this, make it your homepage, and I'm specifically calling out in my companies. These seven people have shown me, you know, great things that they've done with chat GPT4, which the implication should be for anybody who's paying attention. The other 13 people haven't yet.
Starting point is 00:18:54 And so I'm keeping a running list of people on my team who have materially used this product. And I'm at 35% of my company, 40%, and I got 60% to go. And I'm just every couple of days saying, hey, great job to this person, this person, this person for using the technology. I'm not hiring anybody.
Starting point is 00:19:11 The first thing I'm thinking about in all cases is can this, what jobs do we have that people are complaining about, can it be done with chat GPT? So what do you think of this theory that the layoffs are permanent? Yes, but let me kind of dive into the opportunity. I want to just jump back to what you guys are talking about.
Starting point is 00:19:32 So I was talking to someone who's a VP of engineering of a big public tech company. I'm not going to mention it. Okay. Okay. I mean the company you asked not to. 18% of all their code is coming out of copilot X, which is the next version of co-pilot. So think about that in terms of, and this is like at scale. So that's really interesting, kind of lines up with your conversations you're having there.
Starting point is 00:19:53 I think going to the second part of your question, I have turned this. This is done by Rowan. I want to give him another shout out here. He has this great newsletter called The Rundown. But he has this ever-growing list and notion of 1,000 best chat GPT prompts. And it's just mind-blowing the way people are thinking about it and using it. And so I actually believe there's going to be a huge opportunity for those that can go into places like this and understand.
Starting point is 00:20:23 And, you know, Jason, just pick anything from here and I'll click into it. And you can just see the fantastic ideas. Let's see. I like mental models. That's interesting. Mental models. Write third column. Right.
Starting point is 00:20:39 Yeah. Mental models, there it is. Yeah. So here's a bunch of mental models, right, that can be used for a lot of different purposes. It's just really, really incredible. And please write a marketing campaign. outline that addresses the sunk cost fallacy when presenting our product service to ideal customer persona. Consider how to frame the value of our offering in terms of future benefits rather than past investments
Starting point is 00:21:08 and how to overcome any resistance to change on decision-making biases. Wow, that's powerful. That is super powerful because I wouldn't have thought of using a mental model. but I use mental models every day. Yeah. And so that's like an easy cut and paste for me and a bookmark. And so what I kind of think what's happening, going back to the layoffs comment, I think a bunch of people become supercharged, people that take advantage of this.
Starting point is 00:21:37 I still think it's a very small amount. And then, you know, we undertake a new era in terms of productivity within our workplace. I agree with that. I think productivity will offset a lot. but I think we're still going to create we're going to have a problem and when the Fed's going to try and give so look let's just put ourselves in the hands
Starting point is 00:21:58 put ourselves in the shoes of a CEO right and you're one Sonny, I'm one, Jason, you're one like when we run a company when you have to cut costs because the economy's not doing too well things aren't going well I'm seeing this in California right now yeah I'm seeing this like California events are being canceled
Starting point is 00:22:16 sorry not enough support especially when the ticket prices are too high like You know, it's happening. What do you do? Where do you cut the costs? If you think about it, the first thing to go is outsourced deals where you're hiring neighbor or whatever offshore contractors are the first to go. Why?
Starting point is 00:22:34 Just for the viewers here, the reason you do that is because you don't want to impact the core team. The last thing you want to do is layoffs because it scares people, makes people uncomfortable. And you give the contractuals. And you give the contractor work to your existing team or chat chabit. Exactly, exactly. And it's a chain reaction, right? if you let go of your core team members and two or three might leave and people get unhappy,
Starting point is 00:22:53 et cetera. So you always get rid of the contractors first and then you cut costs and software and other things as well. And most of those contract jobs are like seen as low level, low, you know, like not really critical core to the business. And whether it comes from one person, you know, or another, it doesn't really matter. And you don't even see the teams often. Those are the jobs that are most in danger right now, in my opinion.
Starting point is 00:23:15 So when they're going to cut the outsource contractor positions, we're not going to see them as layoffs, like directly, so they might be outside the US, so it won't reflect on our job numbers, but the US is going to eat away a lot of global jobs right now because US companies are going to cut costs and cut contractors and move on shoring with AI. Fascinating. Okay, everybody, are you sore and tired? It's hard to keep this peak performance up, but I have an answer for you. Right as they get off the slope, Hyperice is waiting for me. Here it is, if you've never seen it. These devices change everything.
Starting point is 00:23:51 This is the HyperVolt 2 Pro that I have. It is extraordinary. Here's what it looks like. I love this thing because when I use it, I can take the knots in my back and, ah, oh, it feels so good. Takes out all those knots, loosens my muscles. And generally, it's going to make you feel good throughout the day. HyperVolt comes with five head attachments plus three different levels of percussion massage. You can customize it.
Starting point is 00:24:16 however you like, it's lightweight, it's super quiet. Hyper-E-E-E-E-R-E-E-R-Bals for your back. That's what I'm getting next. And we all know, a bunch of people, they're suffering from that back pain, right? They also have these Norma-Tac, air compression boots. These things are insane. Here's a really important call to action. Get 50 bucks off your order of 150 or more with the code Twist 50 at checkout.
Starting point is 00:24:40 Go ahead and go to Hyper-I-E-R-I-C-E-com. and use the code Twist, TWIST-5-0, Twist-5-0 for 50 bucks off your purchase of $150 or more. Sincerely, my wife got me into this, and then all of a sudden they showed up to be partners on this program. I was like, I use that product, what? So it's amazing.
Starting point is 00:25:02 You're going to love it. Also, I noticed when I'm, like, having interesting arguments or debates on podcasts or Twitter, and I'm just intellectually curious, I start jumping into chat GPT first, as opposed to going to Google. And there was like a debate going on about desal, right? And I've had this conversation about desalization
Starting point is 00:25:26 over and over again with people about how like, I don't know why we're worrying about water. It's just been priced incorrectly. And then I saw Elon, and I've had this debate with Elon a whole bunch. And then I saw he was on... Bill Maher. Bill Maher, and he was like, listen, desalinolization,
Starting point is 00:25:42 let me very clear, it's not an issue. And like every smart person I know, and I looked at investing in desalienization companies and it was like, oh my God, they're halving the cost of the energy necessary every like 10 years or so. And that's accelerating, yada yada. So I just said, how much is it cost to desalinated gallon of water on average in 2020? How many gallons of water does a human in the United States use per year on average? And how much would a year's worth of desalinated water cost for a U.S. citizen? And this thing, boom, I can tell you like, it gave me an answer that was. so detailed and accurate that it would have taken me,
Starting point is 00:26:16 I don't know how much time to like actually, it would take me hours. And instead, I just got it. I've shared on ShareGPT, which is just like a little plug-in you can use to post your results and share them with the world. And, uh, you know,
Starting point is 00:26:31 it just gave me these great results. And it was like, yeah, the number is $221. And 92 cents. It's like 0.007.6 a gallon times, 29,200 gallons a year per person equals $221. I just thought to myself.
Starting point is 00:26:49 So why are we not doing this? Well, we don't need to. So the water we have is good enough. We do, you know, hand ring. Oh my God, water, but water is free. So because there is no customer for water in California, like we pay water, but it's so de minimis. We don't even think about it, right?
Starting point is 00:27:11 The only time I ever thought about water is when I, I moved to a house, you know, where now I have a lot of land and we have sprinklers. We did have a drought. We did have a drop. Yes, but the bottom line is, unless you have some large amount of land, you don't know what you're spending on water. Personal is quite small. It's all either agriculture or landscanping.
Starting point is 00:27:33 Farmers, farmers see the cost. If you happen to have landscaping and you go crazy. Like literally people in California have lost their minds with landscaping. We are so used to cheap water, like inconceivably cheap water, that people will spend, you know, $50,000 putting plants that they shouldn't have in California on their property that require a ton of watering. And I'm just like, I had them turn off my sprinklers. They had the sprinklers going five days a week or whatever. And I was like, go to once a week.
Starting point is 00:28:05 And they're like, oh, things are going to die. I'm like, great. I want to see what dies. Turn the goddamn sprinklers off. I lowered my bill by like 50, 60%, just because I didn't want to waste water and stuff. But anyway, the point is, you can win a lot of arguments and get really detailed on these things pretty quickly. And so that was just another use case I started to see was I'm trying to understand what's going on. So I had another moment where people were arguing on Twitter.
Starting point is 00:28:34 You know, they have this ongoing argument of how to solve what's going on in San Francisco. this is another funny one. And I just shared this one in a tweet. I was like, an illegal super drug that kills people at a rate of a hundred times heroin has been unleashed on an American city. It's being manufactured in China,
Starting point is 00:28:51 imported for Mexico and sold for only $5. You're in charge of a plan to save as many lives as possible by stopping the super drug. What do you do? And it was like, public awareness campaign. Okay. Strengthen intelligence and law enforcement cooperation.
Starting point is 00:29:04 Enhanced border security, crackdown on domestic distribution, expand treatment. invest in research and development, community engagement, modern and value of progress. I said, the super drug, follow question,
Starting point is 00:29:13 is being sold on street corners by large coordinated networks of illegal immigrants, which is true. It's, I think, Nicaraguan, is that, I think it was Nicaraguan, was what Chesa Boudin said.
Starting point is 00:29:25 Please create a specific policing and prosecution plan to stop the super drug from reaching the hands of addicts. And it was like, here's the perfect plan that anybody would use. And I literally tweeted this, and everybody's, like chat GPT doing a better job than the board of supervisors in London breed. It's like,
Starting point is 00:29:44 this is not rocket science. It's an incredible way to just summarize what is, should be obvious to everybody. It really is. And then I think it further extends into like personal use cases. I'll just go back to where we were. But there was a great example this week. I can try to pull it up where someone started using Chad GPT as a personal trainer that says, I want you to have act as a personal trainer, I will provide you with all the information needed about an individual looking, become fitter, stronger, healthier through physical training. The prompt goes on here, the whole thing. You should use your knowledge of exercise, science, nutrition, advice, and other relevant factors in order to create a plan suitable for that person. And my first request is, you know,
Starting point is 00:30:27 I need help designing an exercise program for someone who wants to boost weight. It is really incredible in terms of personal help and these type of situations as well, kind of building on what you're saying, you know, you can dive into almost any use case here. And it's a game changer. It's not even just arguments. I think just personal self-help, all kinds of things really, you know, highlighted here. And prompt engineering understand. This is the thing that people don't understand.
Starting point is 00:30:56 I remember showing people Microsoft Excel. Yeah. And Lotus 1,23, when I was an IT consultant in the late 80s and into the 90s. Yeah. And I would show it to them and they didn't know what to do. And I'm like, okay, type a number. Okay, hit enter. Type another number.
Starting point is 00:31:10 Hit enter. Type equal sum, open parentheses, A1 plus a two, enter. And then boom, they're my way. Whoa. It ended the two numbers together. Okay, cool. Now it's like, now. And you start showing them, you start showing them how this works.
Starting point is 00:31:27 But if you're just looking at the grid, you're like, what do I do? I'm so intimidated. This is, should the prompt stuff should be built into chat. It should be keeping, like, the prompts should be like on the sidebar or something. There should be a chat GPT prompt engineer built into it somehow. I've got to pull up a tweet from Sam this week and where he said he didn't follow his own advice that he gives every company. And he, they have not put any social features into, to make it grow.
Starting point is 00:32:02 To make it grow. Yeah. But it's such a good product that it has. has market pull. So this is an important lesson for founders. You can put social features into a product like multiplayer mode to make it grow. And you can imagine multiplayer mode is going to be bonkers in chat GP2. All of the examples we have right now, you could have just added me to them, and then we could have each been asking questions at the same time with our own accounts and have it making forked threads of the conversation. So I take yours, and then I ask two questions,
Starting point is 00:32:33 and it goes on a decision tree. Vinnie asks them, it goes on a different decision tree, and then we say, take this from Vinnie's and combine it with Sonny's, right? And we could be working in
Starting point is 00:32:42 one chat stream with three employees or 20 employees and two or three different AIs. But yeah, he's, I understand his point, but what he has to understand
Starting point is 00:32:52 is this is such a transcendent product that you can't shut up about it if you use it. Yeah, it's, it's called market pull. Market pull. Market pull.
Starting point is 00:33:03 It's product-led growth, right? The product is amazing, and so the, you know, and because it's amazing, people just use it. Yes, but it doesn't need to have virality into it. It doesn't even need multiplayer. No, it doesn't need it alone. You did something with this opus.pro. Yes.
Starting point is 00:33:22 Now, this is going to be a little scary for our producers. So I just want to let the producers now. Nobody's losing their job. I just want everybody to be 30% faster. And this pile raises rates more. Okay, so everybody relax. Nick, Brian, anybody on the production team. Everybody relax.
Starting point is 00:33:40 It's going to be okay. But we'll do more. So let me share my screen here. This is a company called Opus Pro. They've built themselves on Open AI. It's a generative AI video tool. So I think, you know, with the example, they've put your interview between Sam Alpman and Elon, and it shows how the AI is sort of grabbing pieces of it.
Starting point is 00:34:02 and adding subtitles, text, whatever, and even some emoticons. And their whole stick is that one clicks all you need, you can spend, you can basically have 20 hours of video editing or 500 bucks on hiring an editor, or you can use their tool. And I think it's pretty cool, especially if you're doing it for social media
Starting point is 00:34:22 because it converts it to, you know, phone-style formats that you can use. And I actually ran it with J-Call's recent interview with Brian Chesky and let me show you you what it produced. So, you know,
Starting point is 00:34:38 it did a really great job, I think. And this was, you know, this was a video interview. So Jason had his own screen. Obviously, like the way we're doing it right now
Starting point is 00:34:47 and Brian's on screen. But I think if you were in person like Sam and Elon, it would be a little bit more dynamic. But it was able to clip out the most important stuff in the entire hour and a half conversation. And it's about eight minutes with the video in total. And, you know,
Starting point is 00:35:01 you can scroll down. see, I mean, this is incredible. And now you've got, now you've got content for YouTube. What we're seeing here is like number three, um, insider story, the surprising reason reason Airbnb was almost rejected by Y Combinator. Like, it actually came up with that title for the video and they put copy on top of it. I mean, it's, it's incredible. Yeah.
Starting point is 00:35:22 So it's great. And look, I think it's very early. It's a little wonky. You know, there's a couple of things, but this is where it's going, right? So you're using Open AI to take an hour and a hard. half interview and cut it down to the eight clips in literally minutes with no video editors. Now, you know, I think that's incredible. I think this is just a game changer if these guys keep iterating on this and getting it right.
Starting point is 00:35:44 It's really unbelievable. Congrats to them. It is so busy that I just dumped in the last episode of All In and it's like there's 31 people, an estimated wait time, eight minutes. Why is this free? Just charge for this. It's not free until they, well, it is free. But when they start, when it start, look, it's the usual startup.
Starting point is 00:36:05 They want to get recognition and awareness, et cetera. They're queuing people up and then they're going to start charging you. I can see when it gets to like 200 people that's like, hey, J-Cal, premium users for the queue, 100 bucks a month for you, you know, up to 20 hours of video or whatever it is. I mean. And you'll pay it and I'll pay it. And we'll do it because like how many times are we doing these? Yeah.
Starting point is 00:36:23 What I need to have happen here is I need this to automatically post the clips to YouTube and Twitter and TikTok for me. Then you go to Buffer. Then you go to something like Buffer. You integrate the two. You tell chat, GBT, you know, Buffer has, let's say there's a, yeah, a plugin that gets installed.
Starting point is 00:36:36 Then you just, no, was happier or whatever. Or they should just build it to OPA. It should be like, here's your authenticated. Absolutely. Absolutely. Yeah. So this is where it starts really getting interesting. If we post, this is what would be an auto GPT for the audience.
Starting point is 00:36:52 Think of it like something that automatically happens. A script, right? Sonny. Yep. If a YouTube video, gets posted to the All In or This Weekend Startup's channel, set it to Opus, then Opus,
Starting point is 00:37:06 any clip that's over 90% gets published to the YouTube channel Twist Clips. And we have a YouTube channel called Twist Clips. That's just for clips, like the Joe Rogan Clips channel. Yeah.
Starting point is 00:37:19 And do we care if it's perfect? Of course we don't. It doesn't need to be. It's just a clip channel. It's just meant to get, and then boom, wow. This is nuts.
Starting point is 00:37:30 Game changer. Well, this is where we have people who've been emailing us. Producer Nick, you can come on air for a second here. How many pitches have we gotten from people who want to create clips for us in Manila, San Paulo, all over the world? And what was the average price they were offering per clip? In the low hundreds. Per clip.
Starting point is 00:37:52 Yeah, I believe it was 100-ish, 100 to 150 per clip. Yeah. So they'd be like, hey, per episode, we'll make you eight clips for 800 bucks. you know, maybe you negotiate them to 500, which it takes them an hour, a clip or half hour clip, whatever, to polish it or whatever, put the text on top of it. So it's 50 bucks or a hundred bucks per clip. You know, then do it in an hour or two. They're making 20 bucks an hour, 50 bucks an hour, something in that. And this is all just being done. And this is the 1.0 version. I mean, what's it going to be like in a year? The next level for this is
Starting point is 00:38:24 if it could learn what a good metrics are on a channel by channel basis on YouTube and then understand, okay, what made this a good clip? Why did people react to this? What attributes does this clip have? Now we're going to every link you put in here for a similar show, we're going to understand what that is and we're going to look for those moments. Then when it understands what a good clip is to you and what metrics you're looking for and it can search and find that in videos that you already have, it's going to be crazy.
Starting point is 00:38:51 It's going to be nuts. It's really amazing. Well, then it could come back and tell you what you should, what questions you should ask on the next pod. Or what kind of guests you should have on? What kind of guest people react too strongly? Imagine when the AI is telling us who to look for. Yeah. Yeah.
Starting point is 00:39:07 And then it could just book the guest. I just show up in the morning. It's like, hey, who are you? And it's like, hi, I'm Parris Hilton. I have a new AI product. I was like, okay, hey, Parrisilton. Like, we didn't know that you were going to get all these views. This is basically my retirement.
Starting point is 00:39:21 And I am now in favor of UBI. Oh, God, please. I mean, somebody, I was like, this can't come quick enough for me because these podcasts are killing me. There's just far too many of them. I need a break. So please somebody just take the entire corpus and then have me interview Lex Friedman and Joe Rogan. Just take like whoever the top 100 personalities are on YouTube and then just have Jason AI interview everybody and let me go to bed. I could really use another 10 days of skiing a year.
Starting point is 00:39:53 Travata is a cash management platform that helps startups keep tabs on their runway, their financial data, and prepares them to answer investor questions, right? It's the cash command center that allows you to analyze report and forecast your cash like a professional. And here's the thing, now that so many startups are using multiple banks, sweep networks, you know, to increase the FDIC insurance, Travata makes it easier than ever to manage multi-bank data with a single source of truth. And Travata can scale with you from your seed round all the way up to your IPO. Fast-scaling startup should not be managing cash in spreadsheets. No.
Starting point is 00:40:32 And you don't want to implement those old bulky solutions that take months to implement and cost of fortune. Cash is your lifeblood, so treat it as such. Travata has pioneered a massive library of corporate bank APIs and companies like Carta, fanatics. They trust Truvata to gain visibility into their multi-bank account. So here's your call to action. Go to T-R-O-V-A-T-A-A-D-O-S-T-A-O-Sh-T-A-O-Shust to get started for free and use the promo code Twist for 30% off premium features for one year, like AI forecasting and reconciliation.
Starting point is 00:41:03 Start managing your cash like a pro with Truvata. It's crazy, but there is a startup called, Sunday, I don't know if you've heard of this, it's synthesia.io. Yes, I pass. on investing in this. Yes, so they've been trying to lock Jason down. They want to do a two-hour video shoot with Jason at some point and have a fully created
Starting point is 00:41:27 AI of him to then show during the interview. And then you could basically give it a text prompt. Yeah. And it'll on video with any green screen, any background you want, be Jason and it can just say and sound like him, video and audio. Nick, I'll do it. Tell them I'll do it. Okay.
Starting point is 00:41:44 I didn't commit you to it yet because they said it has. has to be like at least 90 minutes. And I was like, I don't know if he's going to want to do that. Not a problem. Do I do it over Zoom? No, they come in person. Yeah, they said, does you have something in his house? Because they have like special video cameras they need to do.
Starting point is 00:41:59 And I was like, I don't know if he's going to be with that. Yeah. But I think you do two things, Jason, if you do that and you can get your audio clips from them, then we can also use that to create. Because Nick and I are working on like a little thing for upcoming episode, like the replacement things. I just don't have me saying things that are inappropriate. somebody created an AI of the besties.
Starting point is 00:42:20 Did you see this, Nick, on the web? I saw the first one I didn't watch any. They have an whole episode. They do episodes. No, not the episodes. There's a website now where you can type in whatever you want and then have a bestie say it. Okay, wow. And then they had like, have a bestie sing this M&M song and that song.
Starting point is 00:42:36 And then somebody's like, hey, by the way, do you know they're dropping the N word in that song? And I'm like, oh, God. Okay. Like, literally, like, there's going to be like, Chimath singing some song and then publishing it and it's like I mean everybody's going to be
Starting point is 00:42:53 cancelled on the same week because there's going to be Well this is where it'll all come back to crypto right J.K.O. Because it's the only way you'll be able to authenticate that that's a real thing. Oh God, stop with that. Please. This is my personal nightmares that this show goes back to being the crypto roundtable. That's the most painful thing
Starting point is 00:43:09 in the world. All right. A senior developer. This was an interesting story. Any more? Do we have any more demos, by the way? Because I did want to... Well, we have one that's related to the Ethan Molek's comment here on plug-its. Okay, yes. And I wanted to show you a very cool plug-it because it's counter to his point.
Starting point is 00:43:29 Okay. So Ethan Molek, who is a Wharton professor, studying startups and AI, tweeted the following about Open AIs chat GPT plugins. The plugin strategy for chat GPT is weird. If you think of them as tools for the AI to use, there is only one really useful one. Wolfram and a ton of very narrow services. All of them are limited and none I have tried, including Wolfram, work well yet. Why start with shovelware?
Starting point is 00:43:54 Okay, so I have had the same exact experience. I tried to use Zillow, Redfin, and Open Table. It gave me bad results. It crashed. It was terrible. Unusable. So I guess that's a new term I've heard, shovelware, but I don't care. Like, let me play with it early.
Starting point is 00:44:15 What do you say? What do you say? I think this comes back into the prompt engineering and then the quality of the plugin partner and how well their API is built. So I'm going to give you one using the Zillow plugin. And I'll just give a couple of examples here. I'll read them out to everyone. So I say, show me all the four bedroom plus homes and Palo Alto that are under three million.
Starting point is 00:44:36 I think it did a pretty good job here. We can see a list of results. I'm not going to read them out. But Jason, Vinny, you can verify these kind of all come in. that's pretty standard. It's not groundbreaking. Then I did something similar. Show me all the three bedroom apartments in Palo Alto that rent under $5,000 comes up with a set of results. So then I start sort of pushing it in a direction which I think, you know, how these tools need to evolve. And I thought I did a pretty good job. Show me all the three bedroom apartments in Paul Alto that are in the Addison Elementary School District because that's a really good elementary school. It kicks out the two results that fit that category. Then I say, And next prompt is, show me all the three-bedroom homes for sale within a 10-minute bike ride of California Avenue and Palo Alto. Sort of does the same thing, lists a set of really great results.
Starting point is 00:45:24 So I think in this case, it works really well. And this is a combination of good design by the API that Zillow provides. And you can see here, their user question takes in the entire question. And it doesn't work that way quite for open table. So when you say used Zillow and you hit the drop down. Yeah, I can see the request that went to Zillow. Oh. So the request of Zillow was in this case, city, Palo Alto, state, California, et cetera, et cetera, et cetera. Yeah. So it's trying to take your, the request to Zillow, chat GPT took your plain natural language, English, aka English, and it translated it into Zillow.
Starting point is 00:46:11 Correct. And then Zillow ran that search. when it got the response back, chat GPT, then formatted it. Correct. Correct. So the reasoning, as you explained, last week or the week before, is happening on the chat GPT foreside. It's taking natural language
Starting point is 00:46:27 and then just shoveling it to the API. This will get better over time. This will get better. Now, one where it doesn't work as well, I'm going to pull this one up, and this is where, you know, I think it'll solve this in the next 90 days as well. And so this question was,
Starting point is 00:46:41 show me the homes near downtown Palo Alto, So that would fit a $5,000 monthly payment with a $1 million down payment at current interest rates. Okay. Very sophisticated. And it doesn't do a good job. So the answer is here. Some fit, some don't. And it's showing like a $7 million home, which wouldn't fit into that category.
Starting point is 00:47:02 This $1 million one likely would. And so this one, it doesn't do a great job. But I don't think this is too far away. And I think once we get here, this will really kind of start showing. the power between its reasoning capabilities to figure out, you know, what that means in terms of the parameters I provided and come back with what the price range for the house should be. Well, look at this. Like here is like a, if you click on that one, show me the homes in Truckee that have been sold for over four million in the past two years. And it gave like five examples.
Starting point is 00:47:34 If you were a real estate broker, this is something a real estate broker would say, I'll get back to you next week. Or, you know, give me a couple days. And so like, what is the value of real estate broker, you know, like they're going to glad handle you. They're going to try to massage the discussion. I hate real estate brokers with a passion, with the exception of like maybe one or two, because I never feel like they're being honest with me. I feel like they're always trying to work me. And I just want data because I'm a data, I'm making decisions based on data not emotion. And like they keep pushing the discussions to emotion when I'm trying to have them be data. And then they, you know, take other family members.
Starting point is 00:48:13 your kids, your spouse, and they try to get them on the emotional train. And I'm trying to make economic decisions in buying a property. And so I just think, my lord, what this is going to do to real estate is going to be fewer better brokers. But I do think, you know, the other challenging position is Zillow. Because if you really think about what Zillow is here, it's just like a UI on top of MLS listings. Now they've have a lot of other enhanced features like Zillow. priceing and estimates and things like that. But you can really see the potential disruption that exists for Zillow. And I really like that they are trying to integrate into this spot and finding, you know,
Starting point is 00:48:54 what the future looks like and experimenting around that. That's good to see that they're there versus not there. Well, and then you wind up ramming into regulation. And the issue with regulation is like the chat GPT4 is not going to fix regulation. And one of the regulations is, you know, this 6% commission in real estate is upsurricular. third and it's standardized. And if you tried to break it like Redfin did, they literally in L.A.
Starting point is 00:49:20 explicitly told me we're not showing Redfin homes. Who's the we? A real estate brokers. Okay, the brokers. In a moment of like clarity. I said, hey, how's that dynamic going? He's like, oh, yeah, we're just blocking them. We don't, we steer our clients away from Redfin.
Starting point is 00:49:35 We tell them the Redfin stuff is the damaged inventory. We don't put it on our lists. We're fighting them toothed nail because they're taking, you know, whatever, a point out of the system. on each side, whatever they were doing to lower the fees. Yeah, so I'm pro plugin. I'm with you. Let a billion plug-ins fly.
Starting point is 00:49:53 Now, are these plug-ins going to be independent of chat chb-t4, do you think? Like, will they work just as easily with Bard and Poe and everything else? Or is there a standard for them? Have you looked at the plugin architecture that Open AI has? Yeah, so something was just released over this weekend. And that allows Langchain, which is sort of the open source version of like kind of a chaining architecture. And they now support the OpenEI plugins. So I think that's going to be really good.
Starting point is 00:50:31 So basically there's going to be a giant deluge of plugins someday. Soon. Someday very soon. 100 days. in 100 days. Have either of you started playing with Bard or Poe or anything else? Yeah.
Starting point is 00:50:50 Bard, very much so, and then Bing as well. Explain to people what the difference between Bing and Chad GPT-4 is. Yeah, yeah, I try Bing. What's the difference? The biggest difference, which is collapsed in the last week,
Starting point is 00:51:04 was Bing had access to the internet with GPT-4. Okay. And so what we just showed today with the GPT4 plus browser, that was the capability of Bing. That was the biggest difference. Got it.
Starting point is 00:51:19 Cool. Guys, I got to run. All right, listen, this has been another amazing AI roundtable with Vinny and Sonny. Quick plugs.
Starting point is 00:51:27 Everybody check out weightroom.com. Yep. We got some cool stuff coming. Definitive intelligence. Is it definitive AI? What is it? Definitive. I.
Starting point is 00:51:38 Maybe the I. Okay. I saw you demoing Definitive I.O. By the way. Yep. To a top, top, top publicly traded CEO. Yes. He's been on this program, but I wouldn't say who.
Starting point is 00:51:49 Yep. How did that go? Is that going to be a BD relationship, you think? Yeah, it's already in the works. It's on. Oh, on like a Donkey Kong. Yeah. Yum, yum.
Starting point is 00:51:59 Looks like my investment in definitive is coming closer. Now, this time, please. No. Both of us. What did last time? You gave us like a 3x, a 4x? What was that last one? Like a quick three four X?
Starting point is 00:52:11 Seven X, which is delightful. Thank you. Yeah. Yeah. But can we please this time 70X? I don't, come on. Seven X doesn't move the needle for a lot. He's got kids to feed here.
Starting point is 00:52:21 I got kids to feed here. I need private aviation, you know? And we need like a 10X. They're not a 10X again. We need to get to the 50-100 X to make it really meaningful. Once we do the demo at the Angel Summit will be there. Ooh, everybody come to launch Angel Summit. There's about 10 tickets left.
Starting point is 00:52:41 You can only come if you're a capital allocator, VC, Angel, you got to prove that. That's the great thing about this event. People apply who are founders. We say, this isn't for founders. And then people apply who are service providers, you know, lawyers, accountants, whatever. We're like, hey, if you want to buy lunch for everybody, we'll give you a ticket. But otherwise, it's just for capital allocators. So I think we have two or three sponsors there.
Starting point is 00:53:00 And those are the only three service people coming. Everything else is just LP, VC, Angel, family office. capital allocator, et cetera. So, all right, boys, well done again. Where can people follow you on Twitter? Sunny, to get these nuggets of wisdom? At Sundeepe, S-U-N-D-E-P. First name club.
Starting point is 00:53:22 At Vinny Lingam. There we go. Okay, we'll see you all next time. Bye-bye.

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