This Week in Startups - AI itineraries, newsletters, doodles, and more with Sunny Madra! | E1778

Episode Date: July 18, 2023

This Week in Startups is brought to you by… .Tech Domains has a new program called startups.tech, where you can get your startup featured on This Week in Startups. Go to https://startups.tech/jason to find out how! Notion just launched Notion Projects, which includes new, powerful ways to manage projects and leverage the power of their built-in AI features too. Try it for free today at notion.com/twist. OpenPhone. 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 6 months at openphone.com/twist * Today’s show: Sunny Madra joins Jason to demo 6 awesome AI products and features including Copilot2Trip(4:25), Hoppy Copy(19:40), Stable Doodle(34:31), and much more! * Time stamps: (00:00) Sunny Madra joins Jason (4:25)Sunny demos Copilot2Trip’s AI travel assistant (6:30) OpenAI’s fine-tuning ability (11:26) .Tech Domains - Apply to get your startup featured on This Week in Startups at https://startups.tech/jason (12:26) Comparing Copilot2Trip’s itinerary capabilities to Google Bard’s (16:51) What will become of Google Bard and ChatGPT (19:40) Sunny demos Hoppy Copy (24:41) Journalism in the digital era (28:55) Notion - Try Notion Projects for free today at notion.com/twist (31:01) The San Francisco Chronicles article and where AI struggles in journalism (34:31) Sunny demos Stable Doodle (36:32) Bringing back the voice interface (39:42) OpenPhone - Get 20% off your first six months at https://openphone.com/twist (43:01) Google Bard’s new Multimodal capability (50:39) Sunny demos 10Web’s AI website builder * Check Out Copilot2Trip: https://copilot2trip.com/ Check Out Hoppy Copy: https://www.hoppycopy.co/ Check Out Stable Doodle: https://clipdrop.co/stable-doodle Check Out 10Web: https://10web.io/ Follow Sunny: https://twitter.com/sundeep * Read LAUNCH Fund 4 Deal Memo: https://www.launch.co/four Apply for Funding: https://www.launch.co/apply Buy ANGEL: https://www.angelthebook.com Great recent interviews: Steve Huffman, Brian Chesky, Aaron Levie, Sophia Amoruso, Reid Hoffman, Frank Slootman, Billy McFarland, PrayingForExits, Jenny Lefcourt Check out Jason’s suite of newsletters: https://substack.com/@calacanis * Follow Jason: Twitter: https://twitter.com/jason Instagram: https://www.instagram.com/jason LinkedIn: https://www.linkedin.com/in/jasoncalacanis * Follow TWiST: Substack: https://twistartups.substack.com Twitter: https://twitter.com/TWiStartups YouTube: https://www.youtube.com/thisweekin * Subscribe to the Founder University Podcast: https://www.founder.university/podcast

We're now on the elevator going up. Yep. This is an elevator going up. It's going to go to the 1,000 story, which is AGI. It's going to go to the moon. Let's consider the moon as AGI. Where are we right now? Are we at 10,000 feet, 100 feet, 30,000 feet where airplanes fly?

Nobody knows exactly on the way to AGI. But I can tell you we're on the elevator. We're on the elevator going up. The speed is that much different. You concur? Oh, 100%. This weekend startups is brought to you by dot-tech domains has a new program called Startups. Tech, where you can get your company featured on this weekend startups.

Go to Startups.com slash Jason to find out how. Notion just launched Notion projects, which includes new, powerful ways to manage projects and leverage the power of their built-in AI features too. Try it for free today at Notion.com slash twist. and Open Phone brings your team's business calls, texts, and contacts into one delightful app that works anywhere. Get 20% off your first six months at Openphone.com slash twist. All right, everybody. Welcome back to this weekend startup. Sunny Madra is back with us, Sandeep, the co-founder of definitive intelligence.

They provide AI enhanced data analysis for your public and private company data, including autonomous data science agents, right? That's the thing we're going to talk about and we've been doing this this week in AI kind of episode every week because Sunny and I are down the rabbit hole. This is not a drill. It's for real.

And I am making some great progress over at inside.com on my secret AI projects, making some progress. I'll show those to you, Sunny, when I see you next week. And let's get right into it. It's demo or die time. We will show you the demos we're doing at YouTube.com slash this weekend

or you can just search for this weekend startups and find this episode or if you're on Spotify we publish video on Spotify and if you search for this weekend startups on iTunes either are two different

RSS feeds one for video one for audio but we will sports cast it as always so what do you got for us today Sunday? I didn't even know that iTunes did video I got to try it there I always just watch it on YouTube doesn't do video but you can publish a second fee with your video file.

So they actually do support video, but you have to do a second video file. And then Spotify decided they would come up with their own proprietary way, you know, because they bought the ringer and call her daddy and all that stuff. Yeah. And Joe Rogan. So they came up with their own proprietary way to do it. So the industry is kind of in a little bit of a, I don't know. And then Twitter, I see our stuff going on there.

Yeah. So for Twitter, we will publish the full episode now. Shout out to Elon for supporting that. And they have a mini player. So I don't know if you've had that experience. And then the mini player will play on your iOS device. Yeah. Outside of it, just like YouTube does.

Yeah, you can be in other apps and watching the video. You can be another apps. So that's really a breakthrough. So we are independent in I am. What a lot of podcasts do is they only send you to the feed because they want to move up the rankings. We gave up our rankings on iTunes. For distribution. For distribution.

Both we gave up all in and for this. So we would probably be both podcasts would be free. for higher. But we get, you know, you see the views. The view counts are public. You start getting tens of thousands

of views on the YouTube one. I'm sorry, on the Twitter ones. You get tens of thousands on the YouTube ones. YouTube is hundreds of thousands. We get hundreds of thousands. Yeah, already.

I mean, it's, this weekend startups is not meant to be all in. A lot of people are like, this weekend startups is small and all in. The topic of this weekend startups is a niche topic. Yep.

The more we explain the acronyms and dumb this down, the wider it would get. But we want this to be for the top 5% of the tech audience. This isn't for everybody. This is where we're kind of going deep. If, you know, people who watch this, I have friends who watch it.

I don't understand what you're talking about. Or I don't understand half of what you're talking about. That's okay. Yeah. You'll catch on if you're truly an entrepreneur. If you're not an entrepreneur, there are other entrepreneurial podcasts that are generic. This is deep dive.

Yeah. And we don't do the politics. There's no politics here. Yeah. I mean, if you want Biden, Biden, Biden, Ukraine, Ukraine, Ukraine, Hunter, Hunter, Hunter, Trump, Trump. All right, demo time. Demo or die.

Okay. So this one is cool. And by the way, said, you want to do this live when we get back maybe in August? Yeah. If I got it, we could just do it like an AI, demo or die kind of thing. Let's do it, yeah. Yeah. So this one, you know, in fact, I think J.

You talked about this particular use case of creating an itinerary using ChatGPT. Yes, I have a company doing this. Yeah. So these guys kind of take it to a nice. another level where, and so maybe it's called. So this one's called co-pilot to trip. Okay. And so.

Compilot to trip. Yeah. And so I've got one queued up here already and we can do another one if we want to, but I can say, you know, it starts up like it's a chat interface and says, I'm traveling London from San Francisco. It says,

great. London's a vibrant city. How many days will you be there? I say three days. And then it comes up with a three-day plan. And what it does kind of a little bit more interesting than, than chat GPT. is it, it creates an itinerary, puts it on a map, and then gives you a link to all,

like kind of a little bit more detailed link to the different activities it's suggesting. And so, and it has it in this chat view, and I can click over here into a itinerary view that you could print out old school style. So I see it's saying culture, day three, culture of London, start your day with a visit to Tate Modern, 9 a.m., free. 11 a.m. Explore the vibrant neighborhood of shortage. Itch, short-ich. Discover street art scene free.

Enjoy a delicious lunch at Borough Market. Visit fascinating natural history museum. Take a relax. This is a little bit too much for me. Take a rocking boat ride and reaching canal. Have a memorable dinner at Duck and Waffle. Amazing. Yeah.

And see how I did it. One's exploring the city. The next day is about royal stuff and the third days about culture. And so, you know, I think they do a really, really nice job. And this is like, I think an interesting way to think about, you know, A lot of people are like, oh, my God, people are just creating wrappers on top of Open AI. And, you know, I think that does exist. But in this case, these guys are creating additional value where, you know, Open AI probably

doesn't go in terms of getting this level of depth and curating experiences. And then they can fine tune their model over time as well. That's one thing that Open AI has been heavily talking about, you know, the last, you know, a couple of weeks is they're going to enable a fine-tuning API. So you can take all these, you know, customizations that you're building around your prompts and everything else and fine-tune the model into it so that your results are different than the general Open-A-I result. Okay, that's interesting.

Let's pause on that. Yeah. Open AI, the company, is creating a fine-tuning model for their API. For their API. So programmers who are using it will then be able to fine-tune it. And fine-tuning here might be, hey, we only want travel locations. Hey, we want to know the cost of it. hey, we want to block out things that aren't travel, yada, yada, I wonder who gets that learning.

Does that learning accrue to the chat GPT 4.0 model, or do you own it? So the way they've been explaining it is that you own it. That's something proprietary that you're doing with the model. And you are fine-tuning it for your particular use case. And you're, you know, kind of, you know, same way you would do with prompts, but instead of having to do it with the limitation of prompts, you do it sort of over. a longer set of time and say, hey, we, you know, eliminate these things, add these things, focus in this type of way. And so that's something that becomes proprietary between you and

Open AI. And I think that's- Well, so that's an issue because what I saw previously happen was Yelp had that relationship with Google search. Okay. And over time, Google watched what Yelp was putting into the index. They watched how they semantically organized it. They told Yelp, hey, can you put the opening and closing hours? And can you use this kind of of data formats, you know, and then eventually they launched a competitor of Google. So I would just say eyes wide open to anybody with opening eye that if you give them the reinforcement data, they're likely to include that in the court model. And then you might be teaching them the roadmap to train their data to then replace you.

I will say, though, J-Cal, they do a pretty good job of being very explicit to say they have a do-not train as part of, if you have an enterprise agreement with them, they explicitly start with a do not train. And then they, you know, make it very clear that they will not train off your data. And so I think that's an intro, you know, and I think they've done that because of the concerns, like you've talked about that have happened in the past. It's just when you're building on another person's platform, understand that what you consider proprietary and what they consider table stakes will change over times.

So I'll give another example. Yeah. When the iPhone came out, people created calculators. one of the best-selling apps in the app store were calculators and flashlights. That would turn on your camera flash permanently. You built apps and then you saw a note-taking apps were very big, like Evernote. If you were to look at the Apple Notes app today that comes with your iPhone and the early versions of Evernote, eventually I think the Notes app has become probably on parity with the original version of Evernote.

the notes app within iOS? Apple's notes. Yeah. Now you can share documents in it. It's got you can put any, you can clip any data into it. Evernote, but yeah. Yeah, I would say Evernote 2.0 is the equivalent of it.

So what Apple did was, they said, we will always be a little bit behind you. You will just very slowly at features. The features we saw in communications apps. Yep. wound up in iMessage, right? Yeah. So it's almost like a joke.

People have Android or have a, other platforms are like, yeah, you know, iOS builds natively into their product, what, five or six years ago existed. Existed. Existed as a standalone app. So just be wary of that. You'll have a little bit of a lead. You can always build a vertical thing.

And who knows if chat GPT will take that same approach and attack these companies. But yeah, co-pilot to Trip is great. Yeah, but you're still supporting companies doing this. I think we have, you know, that ecosystem still has a chance to break out, right, before, you know, the native platform. I do. I mean, I think, and I think this is a type of use case that folks should go after, right? That it's not just a straight wrapper around Open AI. They're creating like an additional data set, some curation, some other things that are there. You know, the way to think about it is,

say, let's think about Expedia or Kayak, right? What are those businesses today? They were businesses that understood how to interact with the travel API that was provided by like Sabre. And then they built an entire set of experiences around that. And now, you know, obviously anybody can go to Sabre and build something, but those guys built businesses. I think it's going to be very similar with OpenAI where as long as you build some curation and proprietary data sets, which is, you know, they're trying to do around experiences, I think there's, you know, there's still solid opportunity to not get consumed by the apex aggregator.

DotTech domains is introducing a new program. It's called Startups.t.c.ch.com. Let me break it down for you. Basically, they will use this minute or so of ad time to shout out an awesome tech startup that's using the dot tech domain. Today's featured startup is called dealmaker.com. So let's hear about dealmaker. Dot.

Basically, they're making fundraising simple for founders. That sounds good to me. Their platform handles all your documentation payments and more. And their upgrade plans even give you access to marketing experts and SEC licensed transfer agents who will help manage your CapTables records. If you're thinking about raising around, go check out dealmaker.com. today. Here is your call to action if you want to get featured in this ad.

Go to startups. Dot tech slash Jason to learn more. Startups plural. Dot tech slash Jason. Don't wait. Go there now. Get your dot tech domain link like I have and go to startups. dot tech slash Jason today.

And now let's think about Google Bard. Yeah. If you think Google Bard, if we go to Google Barred, if we go to Google Bard and we say, hey, give me an itinerary for my trip to London. Yeah. And it's a, it gives you the same itinerary. And you say, okay, book me tickets for Big Ben,

book me tickets for get me a reservation at this restaurant and then do this for me. Do you think that their intention is to have that built into Google Bar? And I think I know your answer. Yeah, I think. They will want to own that consumer business. Yeah. Let's see. Let's try it here.

and let's see what it comes up with. We have Google Bart open here. I was going to use it for a different example, but we have it. Look, at the end of the day, the thing about travel is, it's highly monetizable, right? It's been probably in the top three or five monetizing areas of the internet since the beginning of the internet. Because there's transactions that are big. Exactly. And so I think you have to really, you know, take a step back to think, will they, will they not do it?

No, they're going to do it. The point is, like, you know, and look, this is incredible. right. We have a startup over here in this tab that created this really nice exploring the city, Royal London and cultural. And here we have, you know, sort of the top tech company. Google with unlimited resources. Unlimited resources. And look at their kind of morning, arrive and check in your hotel, afternoon, take a bus, get a delicious meal at a pub. Right? And so look, can they bridge that gap for sure? But I think, you know, this is what entrepreneurship is about. And maybe if these guys do a good job, or the company you're supporting is a good job.

They get acquired and this is what makes it in there. So I think this is really an interesting time that this is not, you know, this is interesting and it does a half decent job, but it's nowhere near as good as what the co-pilot folks have done here. Yeah. And so you have to be prepared. I think as entrepreneurs, the best advice we give people is build a sticky service

where, you know, you're putting in your notes, maybe collaboration, and then you constantly have to be adding new features and studying the customers. and you have to be bigger than the, faster than the incumbents. Yeah. I guess the question is,

proprietary data set. Like what they, like I like how they've done that, you know, the London, Royal London and get, you know, explore London and then,

you know, cultural stuff. I really like, because that's curation that they're doing on top of the models. Do you think at some point they should create their,

take one of the existing LLMs that are out there and train their own? And then they would have even more control of the destiny because Google, let's say, this company starts succeeding.

Yeah. And they raise $25 million and they're making $3 million a month. They got $30 million in revenue. They got $25 million in cash in the bank. Would you advise them if you were an investor on the board? Yeah. Would you tell the founder, hey, go take, what's Facebook's language model? Lama.

Lama. Go take Lama. Yep. Go find a bunch of proprietary data sets, cut deals with them. Go start building your own data sets and then, you know, do this. You know, it's fine that you built on chat, GPT4. find you've built on Google's API, but you really want to start investing in a proprietary data set.

I think it's a really good question. You know, the way I kind of give advice to startups is crawl, walk, run, and which phase are you in? So in particular answer to that question, not in the crawl phase or the walk phase, but in the run. So in the crawl, you know, you'll just be burning your energy and cycles trying to do that, and it's a lot of heavy lifting to get that off the ground, although it's getting easier in a laying chain over the last couple of weeks.

have added much easier support for multiple LLMs. But you want to be in the walk or the run phase, mostly in the run phase of your startup where it's really working, it's scaling, you've got a lot of customers and revenue, that's when you want to get to that. So when you have customers, revenue, and resources, aka money,

and developers, then you kind of do that. You can build on somebody else's system until such time that you think, you know, giving them the data might be problematic. And that's really, I think, my concern, maybe I'm a bit paranoid, but if you watch a lot of the things that people were building on top of the Facebook API,

they rug pulled everybody. You look at a company like Apple, they wanted to have a thriving ecosystem. You look at a company like YouTube, they want to have a thriving ecosystem. See, it's unknown right now if Google and chat GPT4 is but two examples want to be an API provider

and they want you to own the customer relationship or they see themselves as owning the customer relationship and they are the sole source of customer interaction. if we were to look at the two companies, what does your gut tell you five years from now, will these both be consumer products primarily, or they both be APIs,

or will they be both? Well, you know, they both have a little bit of conflict, right, because you have Google's cloud business who is offering, you know, Vertex API,

which is their LLM platform, which has Google LLMs and support for other LLMs, most of them outside of OpenAI. And so in that model, if you show up as like a GCP customer, they're saying, hey, kind of like Netflix and AWS, right?

But on the same side, you know, Amazon has Amazon Prime Video, which competes with Netflix, which also runs on the same underlying infrastructure. So I think you're going to have these scenarios where your service provider, in this case, like your cloud, is going to be offering you primitives that you're using, but may also be competing with you. And you have to, and, you know, what does Netflix win on?

Content, user experience, algorithms. and so they don't really worry that they run on AWS in the same, you know, in the same way that, you know, Amazon Prime Video does, because they really kind of win out on a better user experience there and content and everything else they're doing. Yeah, so this is going to be the key case, you know, 20 bucks a month for chat GPT4. Yeah. That's a big number. Doesn't seem like a big number, but that's not an easy consumer number. That's like Wall Street Journal, you know, the information, you know, like you have to be. have to be a deep insider to be able to for 20 bucks a month, 30 bucks a month for a service

for an enterprise and even enterprises, they like to pay eight bucks a month for Slack, not 25, right? And so it's a very $300 a year to consider purchase. It's not a mass market product. Google though, they own consumers and that's their business and they own advertisers. And I see this as the ultimate advertising. You put together somebody's dossier for their trip and you have a Google wallet in there or the credit card information, they're going to start confirming these tickets, I believe.

So that would lead me to be thoughtful about how much value you're providing. If you can't provide in the specific vertical more value than Google can, you can be running in front of a freight train. Yeah, it's going to be competitive, but I do think

when I see this example that we just showed, I think startups are innovating really fast and coming up with better experiences and kind of get out there and monetize users, create those, you know, the simplest experience, like zero friction, clicked to buy. And like, that doesn't exist in co-pilot yet today, but in the co-pilot to trip, but I think the faster they do that, the better off they'll be. Yeah. Okay. What do you get next on our list of demos? This one's interesting because this one is like a mix of using two AIs together, one from Google and one from these folks. So this is a service called Hoppy Copy.

Copy. Copy. H-O-P-P-Y. Copy C-O-P-Y. Yeah. And in this case, they have a bunch of different creators, but they have a newsletter creator.

And I know this one's important because I really like the daily inside newsletter I get. And we got some props on that recently in one of our private chat. Yes, one of our friends. And so what this one does is, it's a newsletter creator.

And how this is interesting is they wanted some details. And what I did for the details was, So it said, hey, who is this from? And it said, hey, give me a little bio. And so what I did was I went here and I just said Jason Calcanus bio. And then inside my Google, I have this enabled now, which is their Gen. AI. And so I'm just doing a search here for the folks listening.

And so it says, do you want a generative AI powered overview of this results? And my search query is Jason Calcanan of bio. And it does it a toolbar or something? No, no, no. This is just Google. Google. Oh, this is the Google Lab. experiment.

Yeah, exactly. So you can enable this now. Yeah. You have to go to that Google. Are they calling it Sandbox? Is that the name of it? Yeah. I think it's called a Google Labs, right?

In Google Labs, you go and turn this on. And then inside your search, you get this, any search you do, you can kind of pre-pend the results with a generative AI result. And so in this case, there's the standard results below. Wikipedia is the first one for J-Cal. but up here what I see is like sort of a summarized one right which is Jake How is an American entrepreneur angel investor you know your birthday here I hope that's accurate and yeah the book and you know it's even got your net worth on here yum yum so okay great but you know wow doing great you're doing great wow Uber stocks up yeah exactly and it talks about you know it does a pretty good job so yeah I mean it's my Wikipedia page basically yeah and and I didn't have to go there.

So I copied that over here. And basically that's what I jumped into the from. And then it said, you know, who is the target audience of this? And I kind of went back into there and I said, audience of the inside.com newsletter. And I did the same thing I hit generate.

And basically, you know, it says, here are the details of the audience for that. And so again, I don't know where it got this from, but. That's hallucination.

That's not available anywhere. But yeah, sure. Okay. But well, maybe tell us, Is this accurate to any? No.

Okay. So, dangerous. But I don't think that's accurate. Yeah, no. Okay. We should check it. But anyways, yeah, because I was wondering where it got it from because I don't know if you're publishing, like, you know, where the audience is from.

I'm trying to think if there's a sales document. It could have scraped. Yeah. It's possible, but unlikely. All right. And so, but anyway, I use that because, you know, I didn't write it myself. And then it said topics or interest of this newsletter. And I said, latest tech news, right?

That's what the inside newsletter is about. And then what it does is it creates like the sections and then you can have it write them. And so obviously it wanted a little overview of J-Cal. And so it wrote this little blurb here. And so this is the newsletter on the, so I'm in the UI here. And on the left is the proposed sections on the right are the ones that I'm putting into it. And so on the left here, you know, it's got like a unicorn spotlight that I generated. And then I can I can push that over into the, you know,

into the document, which I have here on the right hand side. And it's got other suggestions. I can do a Tech News Roundup. So if I click this generative write button, it'll go and it'll go off. And it'll start like kind of coming up with the topics for it to talk about. And it's got a whole bunch of different things that's come up in Startup Spotlight here. Ask Jason anything. It's a recommendations.

Yeah, exactly. So exclusive offers and events. Basically, this will be built into substack or Beehive. It's basically these are prompts. that will help you brainstorm what to write in your newsletter. So this has previously been done by a newsletter consultant. Or just a general writer who reads other newsletters or a consultant might give you this information.

Hey, here's some ways for you to brainstorm doing something interesting with your newsletter. Yeah. And they just make it super easy to come up with this Ask Jason Anything section. And you know, you can push that over into the document. and it makes it kind of over there. And I think they've just done a really, really good job of, you know, like it even came up with the baseline Q&A for this section, for Ask Jason Anything.

Yep. And look, I've never really done a newsletter. You do this as part of your business. So I'll hand it over to you here and say, like, you know, how do you think about this and how many people work on this now? And could you guys leverage this tool? Well, you know, what we do at inside.com is we have verticalized newsletters and we have writers.

I think people getting paid $70,000 a year. Yeah. You know, it's above average for a starting salary at Business Insider. And the starting salary of Business Insiders like 50K, 55K. So we call them analysts at our firm. They analyze the news. They put the top seven news stories in order of importance, and they summarize it with bullet points,

a format that Axios, I think, cribbed from us. Or, you know, maybe people just come to the same idea at the same time. I'm not sure. So that is a pretty cool format. It really involves two things. One is curation, going and finding the most interesting stories in the world, and two, summarizing them, thinking about the reader and getting them through the newsletter as quick as possible. That was my premise. This is built a couple of million dollar a year of business at Inside.com with, you know, I think we have 30 newsletters.

It's a nice business. It's an incredibly hard to scale business, and it relies on advertising. people will pay for a newsletter like Casey Newton's, let's say, or somebody's voice, Barry Wise, if it appeals to them and it's like biased. And when I say bias, it has a point of view. I shouldn't say biased. If it's, you know, got it. So platform is Casey Newton's.

Eric Newcomer has one. So you take a top writer. They can get to thousands of paid subscribers, a couple hundred. My premise with Inside was to do it for verticals. The problem is people aren't willing to pay for that. because it's not enough. But it's enough to get an advertising business going. And so what I've been brainstorming on and grinding on,

which is a little too soon to announce is, how could AI help this, right? And it's obvious that writers are now taking stories and putting them into chat chipte and saying summarize them. So if you were to take a story right now and cut and paste it and say, just give me these bullet points, it would not be as good as our writers. In my estimation, it would be 60% as good, 70% as good.

When will it be 100% or better? I don't know. What do you think? Well, you could probably get it to 95% if you just did the work and said, if you took the corpus of inside data and said, here were the stories and here were our summarizations, and use that and going, this going back to earlier conversation to fine tune an existing model to say, this is the format in which we summarize, right? There's some, you know, flavor there that, you know,

you obviously inspired and the team runs with, I think you could get to 95% results pretty quickly. And how would you do that? You would. With a prompt or by making... You would do this with like creating your own training set, right? So you'd have the, you'd have like, say, two sets of data, right? And then many, kind of many iterations of it.

One is the original article and then your summary. And you do that for, you've done it thousands of times or tens of thousands of times. And you would feed that into this, you know, the fine-tuning APIs of these models and say, hey, this is how the inside.com voice summarizes long-form articles into our bullet points. And then the- How many articles do you need to train it? The more you give it, the better. And then what you would do- What do you think a baseline is for it to understand?

I'd say like 100. Okay, got it. Yeah. And then what you want to do is on top of that, once it's done that, and then as you have it generate new ones, you want to give it what's called like reinforcement learning and say, this is good, this is bad. So further over time, it's also learning that it's what is taking- What software and platform would you use to do that? I mean, like all of the large-scale LLMs that have APIs support this now. Got it.

And this is quite tying back to that. This is that fine-tuning API, right? And so this is where you would do all that in the fine-tuning API. Now, you can do this outside of the models as well using kind of other frameworks like Langchain and kind of vector databases to create like a similar experience. But I think the best results will come from the fine-tuning APIs of these LLMs. Fantastic. You know, I run everything on Notion.

It's incredible. I've talked about it here hundreds of times. I talk about it on Twitter. I love Notion. Helps me run everything inside our company. But the most important thing, it turns out, It's not just documentation. It's not just reporting project management. We are increasingly doing more and more complicated, detailed, and collaborative projects. And listen, we are putting on a 200-person

event entirely on Notion. Sales, pipeline, AV, checklist, schedules, all of that in one place. And Notion is about to make project management even better. Today, I'm excited to share that they just launched Notion projects, which includes new, powerful ways to manage projects and leverage the power of their built-in AI features too. Notion Projects combined project management with your docs, knowledge base, and AI

so you can stop jumping between tools and stop paying too much for them too. Notion is the critical tool that everybody's using and startup land and big companies

are starting to get into it. I know because I'm starting to have them share collaborative documents with my team. So here's your CTA. Do your most efficient work with Notion Projects.

You can try it for free today. At notion.com slash twist. That's all lowercase letters. Notion.com. slash T-W-I-S-T. When you use our link, you support this show and all the people who work on it to bring you all these startup lessons, incredible guests, news programs, AI, all the great stuff going on here. So go right now to notion.com slash twist. I think this is going to be not the future of journalism, because in journalism you have to call people on the phone.

So I've been thinking about what is a journalist, right? Yeah. They should be asking the who, what, when, where, why. That was the original training, you know, and then you interview people who witnessed something, you research it. But what is journalism today? Journalism today is like, I got the who, what, when, where why from tweets. Yes. And then I crafted a story based on tweets or emails to my inbox.

Like, that's kind of what journalism has become in the digital era. Journalists don't pick up the phone and talk to people, right? Very few of them do that. And when they do, it's considered like, whoa, that's crazy. I don't know if you saw the San Francisco Chronicle sent to reporters down to, is it Honduras or Nicaragua? Where are the drug dealers, where do they claim the drug dealers in the tenderline are from? Oh, like that MH17, that whole kind of thing? No, it's not MH17, but they did this incredible article where they went to this country.

I think they're Honduran. Okay. And they found the town where this was all occurring. And it's Honduras. And so this hometown of San Francisco drunk dealers. So there's this one area there. All of the mansions, and when I say mansions, I mean like an average house in America. A regular home in Texas, like a strip home, have Warriors logos.

The basketball teams on the outside of the houses of the rich people. They have San Francisco 49ers. It turns out these folks from Honduras who came here to drug deal, then take the money back. And then they build themselves mansions. And on the gates, they put 49ers logos and the Bay Bridge. Wow. And they went there. Two journalists went there for a little dangerous assignment.

And this is going to win them a Pulitzer, I predict. That kind of journalism. That's not getting done by AI. No. So what I think is like really interesting is how much of journalism or could journalism be augmented, right? We talk about copilot. So what's the journalism equivalent of copilot from, you know, what's the journalism equivalent

of GitHub copilot? Well, you know, it's interesting that you say it. And so I hadn't thought about it this way to you gave that example. And what's, you know, what probably the. will force to happen is, I think, to your point, journalism has gone in a direction where it's mostly summarizations. And it's not the, and it's not sort of original report. Not original work. Now, if AIs can do the summarizations really well, like we were talking about, it'll force journalists to go back to first principles and create better content because that's exactly what's

going to happen. Yeah. And I think that that's really interesting because, you know, it's, it's not super exciting when someone is just taking a set of articles. and summarizing it and then, you know, putting it together for me. It's like when they do the real work, like what you talked about, what you showed there, that's what is, you know, why I subscribe the New York Times or Wall Street Journal. I want to read sort of this. Exactly, yeah. And that's worth paying for.

Aggregation is not going to be worth paying for. And so that, unless it's some incredible form of aggregation, you know, if the aggregation was so good that, you know, it knew your next story based on your previous behavior and like a for you type algorithm like TikTok. Yeah. you know, maybe on the margins, that would be interesting. Like, okay, you've read these five stories. We think this is a six-month-inch, interesting one with you based on feeding in your click stream or how much data you put on something.

But, so anyway, I'm deep in the rabbit hole. If anybody has any ideas of how journalism and AI will change, Jason at calicanus.com, I'll take any free ideas or thought you have. That's my email for life. What's your next great demo? All right. Okay, let's keep going here. And by the way, I think bias is the other thing that I think, you know,

could be interesting, and I've been thinking about as a vector. Awesome. So this is another, you know, interesting kind of demo that, you know, stable doodle by stability AI, which is, they create one of these image generation models. And I want to talk about two different things here. One is sort of the capabilities. So we've always, you know, done this in a meeting where you're doodling something and say, I really need like a chart that shows something like this, right?

And I'd say, you know, they'd kind of say, this is a, this is an example of a chart that's rising. So I just doodle the chart here for the folks that are listening. You know, we do this in meetings all the time. And then I gave it a quick description of what I'm looking for. And then I hit generate. And what it will do is it will generate images that are, you know, generated based on that chart. And this could be useful, you know, if you're doing a presentation or something like that. And you're not able to quite describe what you're looking for.

you wanted in a particular format. Holy cow. Yeah. This is really good. Yeah. So I started drawing the worst drawing of a bulldog ever. Oh, wow. That is like, you know, the jowls of a bulldog.

And then I used English bulldog as the prompt. But it got the perspective I was looking for. Yeah. And like, this is pretty good. Yeah. That was kind of the perspective I was looking for. Yeah. And this one's not bad either.

Yeah. And, you know, not bad. That's what the doodle helps with, because, like, you could type that in one of these generators, but it'll give you a bulldog in a perspective that you didn't want, right? Like, you wanted, like, a headshot with, you know, kind of lying down. And so this is a new addition to their model. And I just said, all white. Yeah.

And so now I got UGA. Yeah. Amazing. I think, you know, this is a superpower again, right, where combining the, you know, human an ability and augmenting it in this kind of way. Because, you know, I think a lot of people have been experimenting with the prompts, so they get frustrated because they don't get quite what they want. And then they can't fully describe it.

I think what you just did there with the bulldog was a really good example. This is where I think the voice interaction would be go a long way. If I could be talking to this. And I'd say, that's a good bulldog, but I want it all white. And can you make it a little more wrinkly? Yeah. And I'd like it to have like a more of a smile to its face. Right.

Yeah. I don't want to typeos and prompt. I just want it to change as I speak. Yeah. And that's where I don't understand why we lost the voice interface to chat GPT or to AI. Yeah. Siri and Alexa were the start of this, but we don't have it actually, you know, a working yet. Here is like an image my guy's just created, a man with umbrella in rain next to River.

you know, like for a first shot, not bad, but now what I want to do is fine tune it by talking to the AI. So I would like to encourage people to build voice interaction here so this can move faster. I think this is a perfect example where talking to a computer doesn't make you look stupid. Like if I say, you know, like, hey, Siri, call this person, I don't feel stupid because it feels like it's going to be accurate. Yeah. When I ask it to order more coffee or Alexa to order more coffee, it never gets it right. But this is an example where talking to a computer because of the chat GPT and language models could actually be good.

I'd like to have a talking shopping experience with Amazon. Yeah. Where I just talk to Amazon on my desktop or my phone. Hey, I'm looking for a gadget as a gift for Sunny to thank him for being on the show. I want something that's in the $500 range. And that's fun. and interesting. Boom.

And then I say, oh, something more designy, you know, and up the budget to a thousand, you know? Yeah. I think we're just living off

kind of biases from past bad experiences because, you know, and now we're past that and I think the interface is voice, right?

And Bill Gates has kind of always said this for a long time. And I think we're just going to, I mean, I think it's a great idea like to start that way and then have kind of the back and forth between voice

and, you know, sort of, like that, that's the superpower kind of humans. Or, you know, like Amazon does this already. I open up Amazon and it's like, order things that you ordered in the past, or here's some things you might like. So they're really using AI on the back end, obviously. Yeah. And they know that, like, I've ordered some coffee 10 times in the last 10 years or toothpaste or deodorant.

So they're upselling me all the time. You know, imagine I came there and I was like, and it was like, hey, do you own any pets? And I'm like, yes. What, what do you own? I own two dogs. Or what type of dogs do you own? Oh, English Bulldogs. How old are they?

Oh, one's two, one's one. What are their names? Boom. Now, because I'm talking to it, and it's just changing the screen. And it's like, oh, do your dogs have any behavior problems? I'm like, yeah, they run away from me. Oh, have you tried a shop collar to train them? Oh, have you tried a clicker and, you know, doing treats?

Or do you want to order this book about this? That's the kind of interaction I want to start having. I want it not just to anticipate what I want, but I want it to interview me. Yeah. And that's going to be the holy grail. Are you still using your personal? phone number at work at your startup in 2020. Stop.

Such a common mistake founders make. But open phone has totally rethought every detail of what a business phone should look like in 2023. Open phone makes it so easy to do this and so affordable that you have no excuse. And you really don't want your team using their personal phones for business. Why? Well, it could get creepy. People start texting people on your team.

It could be that they leave your company. And the salesperson has all of these text threads going with all your. clients and they bring them to your competitor. Do you want to deal with this nonsense? You don't. I can tell you open phone is amazing because we use it. Our sales team, our ops teams, we use it daily. We also started using open phone for Angel Summit communications. It's rated number one on G2 for customer satisfaction. And let me tell you, those G2 rankings, those are dogged battles. If you win that, you really have to be the best. Twist listeners love open phone. My sales team uses it. Ops team uses it. Customer support uses it. And you know what's great about it? You can create a

shared phone number like we did for the Angel Summit, with multiple employees being able to field those calls and text and keep it all sorted. It's affordable at just $13 per user per month, but Twist users are going to get 20% off that already ridiculously affordable price for six months at openphone.com slash twist. And if you got an existing number, Openphone will port it over at no extra cost. Head to Openphone.com slash twist to start your free trial and get 20% off. And it's interesting that you touch on that because that is the you know, we touched on this before, I think in previous episode, which is open AI's big focus is personalization. Because every time you go back in Open AI, it sort of resets from, and you

have all your history, but the history is not tied together. And I think that is the next unlock we'll see. And that's when we'll start to see the kind of use cases you're talking about. And multiplayer mode. Yeah, of course. Yeah. Like, I want to be in there with my event planning team for All in Summit or for the Angel Summit. And I want to just talk to it. And when I would go in, And it's like, okay, all in Summit is 70 days away. Where are you at? What do you want to work on? Speakers, marketing and ticket sales, the website, or, you know, the parties.

Oh, we're going to work on the parties. Okay. Which part of the parties? The location, the food, the entertainment. I want to work on the entertainment. Yeah, yeah. And I asked it the other day. I was like, because the first theme is going to be like a James Bond Casino Royal

kind of situation. I was like, what themes would you do for this and what activations would you do? When I gave it a role, you're a party planner, you're pitching somebody who's a really a hard-to-please customer I put in there, and you're going to have to sell them on these ideas. Give me 10 great ideas for activations. It's like a photo booth. Yeah. You know, casino games like roulette.

And I was like, that's exactly what we're planning. And I was like, those are obvious ideas. Give me more. And then it gave me some other ones. It's pretty good. Higher actors who can play James Bond. And I was like, oh, that's interesting. Yeah.

That was in the back of my mind. Let's get that going. Nice. Yeah. That's exciting. See, I mean, and like you said, you want to keep that contextual in a multiplayer mode. And look, I think that's less than six months away. I think with the rate, things are innovating.

These are not things that, you know, we're, I think that are too far away for us. Yeah. Awesome. Next one. What's your next one? Yeah. Okay, this may be the last one for today. And I got one more after that, but we can save for next week.

So this was one I was really impressed by. So I'm just pulling up, you know, Bill of Wall's thread because, you know, he did the hard work here. and this is a little bit of a nod to the enhancements by BARD last week. So they added multimodal, which means you can give it photos and have it give you explanations. And so I thought this was really groundbreaking in terms of- Okay, on BARD, Google's Bard, Bard.Guard.com, you can do multimodal. I can give it an image. Yes.

Got it. Okay. an image of a tweet, an image, just an image straight, and you can give it like sort of a screenshot of anything you take. And so what Billowell did here, and I just want to give him credit for this, so I don't want to recreate it, is he took some recent memes. And so here's a meme of, you know, that Chris Hurd tweeted, we're so excited to return to the office because of culture and it shows like an empty cubicle. And then what he did was he turned around and he asked Bard to explain, you know, explain why, you know, why this is funny. And Bard writes an explanation. And this is pretty incredible, right?

It says, and it explains it's funny because the cubicle is bare bones and nothing's in there, but yet we expect people to come back to work for a vibrant culture. That is really impressive that it's able to kind of extract that, you know, kind of humor and corporate culture knowledge. from that image. So what do you think of that? There's a couple other examples here that we'll go to. This is why the writers and the actors on strike are, I believe,

and listen, it's been pretty clear, I'm not a fan of unions. I'm a fan of everybody fighting to get the best deal they can and a competitive environment for labor, just like there's a competitive environment for startups. I understand some people disagree with me on that, but as a high performer,

I would rather negotiate the best deal for me than have somebody else negotiate for me. Your mileage may vary. You may have a different explanation. This is why the writers are underestimating their value, overestimating their value, and not understanding how quick this is,

or the writers are smart and the studios are smart, and that's why they're in a standoff. If you're a comedy writer and you see this explain why this is funny, you should be able to understand that it can do the reverse eventually.

If it can explain the joke to you, it can make the next joke. Okay. These explanations, I saw that when it went viral, that thread, and it nailed it over and over again. It explained the joke.

Yeah. And so this is why... This is a good one J-Cal, right? Yeah. Threads versus Twitter daily downloads, right? You know, obviously shows Twitter downloads not really increasing

because they didn't do anything and threads was going nuts. Twitter downloads are a flat line because Twitter's been around for X number of years. Everybody's got it installed already. And then this is its explanation. This is an image that shows a number of daily downloads with, you know,

threads in Twitter. The graph shows a significant surge or $30 million on July 5th, but then it has declined. The text in the image said it's over, which suggests the author of the image believes that the threads is failed to compete with Twitter, yada, yeah, it's incredible. It's like, it's really, really interesting that, and this is supported by the data in the graph, which shows that threads has not been able to maintain the same level of popularity as Twitter. It is, I mean, this to me is like really, really the next level stuff. And it's just getting started.

You know, I just had Ryan from, I just recorded an episode with Ryan from Qualtrix. And, you know, we were just talking about how this all started in November, December, of last year. Yeah. End of November. It's July. Yeah. So we'll be at the one-year anniversary of. of this basically at Thanksgiving.

Yeah. In this first year, this has been a chaotic, inspiring, and never-ending series of, wow. And it's not ending, folks. What you have to understand. And I also just had Tim Urban from Wait, but Why on. Tim Urban made the famous image that said, like, here's progress. And it's like a slow, slope of progress. And then there's a human.

And then it goes straight skyrocketing up. You know, classic hockey stick style. And it says, you are here. Guess what? We're now on the elevator going up. Yep. This is an elevator going up. It's going to go to the 1,000 story, which is AGI.

It's going to go to the moon. Let's consider the moon as AGI. Where are we right now? Are we at 10,000 feet, 100 feet, 30,000 feet where airplanes fly? Nobody knows exactly on the way to AGI. But I can tell you we're on the elevator. We're on the elevator going up. The speed is that much different.

You would concur? Oh, 100%. Yeah. And look, even between November and today, the capabilities that we're seeing in the examples that we're just showing here, the clip art, doodle thing. Yes, that's my point. Yes. This explanation are like what was happening in November was blowing our minds and now what's

happening in July is blowing our minds even again. We got on the elevator. Yep. At Thanksgiving last year when this came out. And at that time, I told everybody in our company, if you're not using this every day. You can't work at this company. Go buy it for 20 bucks, get Chatschip before. Everybody get on the train. Get on the elevator.

We're going up. And, you know, I think I asked my team, I like to talk about how I manage my companies, because sometimes people find that helpful. I told my team now, because of Share GPT being built into ChatChipT's little share icon, and then people can pick up your thread

and start working with it. I said, include your links to your share GPT, in your end-of-day report. So as a remote company, we do a five-minute start a day. You say what you're going to work on in Slack. We trust you to do the work. We're not monitoring.

I don't do keyboard clicks. Although I do have some data, it's interesting to think about it. And then at the end of the day, I ask you to reply to that and say what you got done and just share it with the team so everybody knows. Five minutes in the beginning a day, five minutes end of day,

and then you get to work from home. The other thing I get it to work from home is you got to come to a stand-up with your camera on and be dressed for work and, you know, showered, shave, whatever. I mean, I technically don't know if you're showering or whatever, but you have to be presentable on camera, I guess, is the ask. Yep.

So at a certain time, you have to start work, a certain time you got to end work, and you got to say what you did. I think that's reasonable and you get to work from home. Totally. And now people are starting to include them. And somebody did a search for the sales team of, hey, All In has sponsorships this time around. We're going to try to sell a couple sponsorships.

Yep. And in the sponsorship stuff, they just said, hey, what are other conferences that compete with Allend Summit or other big tech conferences and business conferences, who are the sponsors, put in a table, put it in a CSV, and we don't have the ability to compare it to our pipe drive, which is our CRM. But eventually I want to be able to just say,

which of these companies are not in our CRM? Or just, I had somebody doing this already. Who are the advertisers on other business podcasts? Tell us if we've ever talked to them. Yep. That will be automated, you know? Yeah. Again, like we're, you know,

I'm surprised it's, you know, that's less than three months away, that type of function now. So. Yeah. I mean, I'm like, where's pipe drive? Get to work. Like, you know,

they just need to do a plugin. They just need to do an open AI plugin. If they had an opening I plugin, we'd be done already. Exactly correct. All right. Let's go for a final demo. Great job this week,

son. Yeah. So the final one is, you know, one that I'm actually using. And I was using this because I needed to build a new corporate site for definitive. And basically,

it's called 10 Web. And what it is, you know, it's folks that kind of like have a website. site builder, but it's AI-driven. And so, you know, you give it your little bit of a description of your company, the products that you have, and it will go and it'll create like the first, you know, kind of the template version of your website for you. And so this ties back into, you know, anyone that's trying to build something that is looking for like a quick way to

get off the ground. You can write your summary yourself. It can have it enhanced through one of the choose your AI of choice, generate an AI of choice. And then you can take that and stick it in these other tools that can further enhance it like we were showing you even with Hoppy Copy. And so this is the one that, you know, TenWeb, I want to give them a shout out. I think they're doing a great job. They made it quite simple. I had just done this quickly before because we needed to create a new corporate website for ourselves and this, you know, this took me less than 30 seconds to put together. And you can use this as a jump off point for kind of building, you know, your home base for your for your business. Wow. Incredible. Yeah. So it asked you your business

type. I could pick a digital agency. After I said, I was creating a digital agency. I could pick this. I think I'm picking the one that you picked. Yeah, that's one I picked. Yeah, yeah. It looks like a nice one with tiles. Then I hit next.

And enter your company name. Calcanus Solutions. We make your dreams into reality. You can even click that in hands with AI above that if you want. If you go back, it'll further expand that out for you. Oh, let's say, generating. Or digital any special in bringing your vision to life.

Yeah, great, even better. marketing non-svents. Service one. SEO, newsletter, whatever you want. Yeah, yeah, yeah. Service one. Brand identity. SEO, content, video. Next, formal, informal, finalize, boom.

Sign up a Google. No, I will not be. Yeah, it's great. I mean, these wizards already exist. And, you know, Squarespace has got a great wizard and some other people have great wizards, but adding a little AI magic, to it makes it even faster and then the ability to talk to it i'm sure squarespace will have this

kind of stuff built in shortly like it's going to and then i think being able to say hey uh talk to it a little bit can i make it a little bit more um uh organic looking can i make it a little bit you know or enhance this section yes or let's put it and what they do is like they kind of tied in like hey can you add a blog section or something like that and i think it's look it really lowers a barrier and I think it's going to be incredible. Like I said, it's... Everything, I literally had this, it's very interesting. I had this internal discussion.

I said, put up, we're doing a new accelerator for restaurant startups, like Cloud Kitchen startups. I'm doing, we're going to be announcing it next week or whatever when I don't know when this is coming out. But anyway, we're doing this little test where we're going to try to give money to people who want to create restaurant experiences inside of Cloud Kitchens. And I said, make a website. And then so somebody was using like hard, more like hard coding it.

Yeah. And I was like, stop. And I just told the person running the program, go pick a Squarespace website, use a 10-money-go. And they were 10 hours into building a website. I'm like, what are we doing here? Yeah. Like, did we not?

And it was just somebody wanted to be more custom or whatever. And I was like, stop what you're doing. This is not why we're here is to be website developers. Yeah. Squarespace will get us 99% of the way there. And then you customize Squarespace. You'll not start with a blank sheet of paper. Stop.

Yeah. You know, and I think this is where entrepreneurs will learn like a lot of good lessons with AI. You know, 99.9% will be done by, you know, an AI powered service. And you can just go back to your core business, like the travel startup you start at the beginning. All right, listen, time to go. What a great job. Everybody who wants to work with definitive intelligence, they're too busy. They've got too many customers.

No, maybe they'll be able to find a spot on your dance card. No, we're always want to talk to folks. Always want to talk to customers. You can email Sunny at definitive.io. Sonny, S-U-N-N-N-Y at definitive. If you got a demo for Sunny, send it to him. Maybe he'll feature it here. If you're a big customer, you know, boom, I get 5% of any referral, so that's good for me.

And we'll see you all next time. I don't. I'm joking. I'm in a festival. I kind of get a little bit. We'll see you next time on this week in AI slash startups. Bye-bye.