Brainstorming ChatGPT Business Ideas With A Billionaire | ft. Dharmesh Shah (#438)

Once again, I've been in software for 30 years now doing startups pretty much my entire professional career. The only time I've felt like heart palpitations, kind of like Sean kind of opened with, it's like there's this party going on next door and I'm here knitting, right? It's like, this is like too big to ignore. I think it's the single largest opportunity and biggest kind of tech paradigm shift we've seen since the internet originally came out. Like mobile was big, but there was a discrete set of use cases. Like when you put a camera on a phone, when you put a GPS device on a phone, A bunch of consumer apps like Uber and others came up and that was awesome, right? But it was not like this impacts everything like the internet did, right? It's like, okay, there's some businesses, some new opportunities, lots of good things, lots of money made, lots of startups. Awesome. This is an order of magnitude bigger than that.

All right, what's up? We have Dharmesh back. Dharmesh, who is co-founder of HubSpot and, uh, Multiple-time guest on the pod, one of the, one of the fan favorites. You're back. And I don't know what we're going to talk about because usually we have these little like cheat sheets where it's like, um, 3 or 3 to 5 bullet points of interesting ideas, topics, experiments you've been running, things like that. And I'm sure you have those, but I don't have the cheat sheet. So where do you want to start?

I say we start with Generative AI, because I don't know if you've heard, but there's this thing called ChatGPT. I get this question from my friends and family all the time. It's like, Dharmesh, have you checked out this ChatGPT thing? I'm like, really? Do you even know me? Like, of course I played with it. I've been obsessed, um, ever since it came out. So.

Did, did you see, I want to talk about your topics, but really quick, did you see, did you guys see this? That, so Sam Altman co-founded OpenAI. He's like the man in charge. I read an article where he was quoted as saying like, I have enough money and I don't want equity in the company. And I don't know if I entirely believe that, but that's wild if true, because it could be one of the more valuable companies in the world the next 10 years.

Yeah, he didn't— I don't know if he said it like— he didn't say it on the record, on the record, but the person reporting it said Sam reportedly has no equity in the, uh, the for-profit version of OpenAI because he's already Wealthy enough and didn't feel like he needed to or didn't want to, didn't want to have that clouding his judgment when it came to this. And this is pretty— you're going to bet what one private company, what one private startup is most likely to become worth $1 trillion or more? I think at this point that has to be OpenAI right now. Is that right? Dharmesh, would you disagree with that?

It'd be up there in the top 3. I honestly can't think of Who else would rank higher in terms of probability of getting to the top 3?

And I don't know what 2 and 3 are.

Yeah. Well, who are the other 2 and 3? Do you know?

I'm, you know, I don't know. I was like, no, I would say that one's the transformative one. Right. I think, um, you know, a lot of the kind of Tesla gains we've sort of seen. I'm not sure if there's like big surprises left. It's like, okay, they will make it better. They'll get to full self-driving and we'll see. kind of progress on that front. But in terms of just raw valuation, it's the wildcard. OpenAI is the one that could actually pull that off.

And they get a lot of shit because people are saying like, Elon kind of is stoking this fire. Like, how did this nonprofit go to a for-profit? How did this open-sourced nonprofit company research lab basically become a for-profit semi-closed, uh, you know, company. And I think that's— people are gonna make— people are gonna take shots and make fun of OpenAI because it's clearly the new powerful thing. That's— so some people are gonna say how it's gonna ruin the world and how terrible they are. Uh, but he did give a— there was a story that came out with a good explanation, which was they were burning a lot of money in the research lab. They needed more money. Um, Elon was gonna be the big backer, so he was gonna pledge or commit $1 billion to it. He, um, and then he was like, no, I don't like the way this is going. Like, Google is way ahead, and, um, I'm gonna take over OpenAI, and, uh, I'm gonna right the ship here. This is the— this is what came— this is the story that came out. Yeah, they haven't— nobody's clarified if this is true or not, uh, but it came out in, I think, the Platformer, uh, publication. And so they go, Elon tried to take it over. Sam Altman and the CTO Greg, uh, who was the former CTO of Stripe, They, them and the group that was in charge of OpenAI rejected that. So Elon's like, basically like, I'm taking my ball and I'm going home. Have fun playing basketball without the ball. And he's like, I'd say he took his funding and he left. So he, a couple months later, he left OpenAI, said, oh, the public story was, oh, it's a conflict of interest with Tesla, uh, 'cause they're also working on AI. But he reneged on his funding. And so now they had this huge shortfall in funding that they were gonna have to cover. And so their solution was, let's create a subsidiary that's a for-profit thing that we can raise money into, cuz we're not gonna get, you know, where else do we get a, you know, $500 million or a billion dollars of donations here? Um, and so they, they did that. They raised money in that and then they capped the profits of that company. So, so that was kind of their explanation, which is a little bit less devious than people make it sound.

Yeah.

They're like, ooh, they tricked everybody by going from nonprofit to for-profit to all the profits, which is, I think how people perceive it today.

Yeah, I don't know the details. I have no insider knowledge, but—

I thought you have like a billionaire chat group where like every billionaire just kind of says the back channel of what's going on. Do you not have like a billionaire WhatsApp?

We have that, but I don't have any insider knowledge from that particular chat group. I mean, my sense here is that building large language models as OpenAI is doing is supremely capital intensive, which is rare for a software company, which is what they are. Um, so it's expensive. They needed access to capital. Um, I think they structured it such that it does cap the profits. I think they've done, like, if you had to do that kind of, oh, we're going to have to spin off and have this for-profit thing, they did it well. And, uh, and I could be wrong, but Sam Altman seems like a reasonable, rational, non-evil guy. I mean, he's, he's a capitalist.

Yeah.

Fine. Uh, and I mean that in the most positive way possible. But, uh, yeah, I don't think he was out to mislead anyone. I think he's trying to solve some problems.

So there's a bunch of ways we can go with this AI thing, but I want to share something funny. Uh, so I basically cleared my calendar this whole week and I just treated it as AI week because I was like, dude, I can't, I can't just sit here and I hear the music at this party just bumping at the house next door and I'm over here knitting and I'm like, I got to put this down. I got to go see what's going on at this party. And so I cleared my calendar and I just spent Every day this week, just messing around with AI tools, just getting to play, play with it for myself. That's how I learn is by, like, just messing around and, and, and trying to experiment and do things. I wanna share with you guys something funny. Basically, I stitched together a few AI tools. So I was like, let me make an intro song for the podcast using AI. So I went on ChatGPT and I told it, I said, oh, this is all I wrote. Write an intro rap for our podcast, My First Million. Our key phrase is no small boy stuff. Okay. So here's, it, it gave me a full rap, but I'm just gonna read you the chorus. So it goes, here's how it goes. Uh, it goes, no small boy stuff. We on that grind. My first million, it's time to shine. We talking big money. No pennies, no dimes. Together we climb one step at a time. And it starts, so it gives us this, this great rap that's on, on, on brand. And then I took that and I found this guy, uh, Roberto, who had made this demo where he turned his voice rapping into Kanye. And I don't know if you've seen this, but it got like a million views. It's this incredible thing where And he's like, yeah dude, this is crazy. He's like, I didn't make this. He's like, I was just on Reddit and I saw that someone uploaded a Kanye voice model, so I clicked it. And he, it literally, the thing is, and I, I should make, I should make a YouTube video about this, like, um, just how to do this one process. But basically it's a Google Colab folder, which is just like a Google's little coding interface. So you don't have to do, write any code. It's just, here's the, here's a place to run the code. And then it's a link to Megaupload. And then the Megaupload is where he hosted the Kanye voice model. And so all you do is you record yourself doing what I just did, and then it turns into Kanye West rapping it, and it sounds exactly like Kanye. It's amazing.

I got a fantasy that's beautiful, that's dark and twisted, but I attacked the whole religion all because of my ignorance. What was I thinking?

That was some bitch shit.

I lost Adidas, but I'm still—

And so, and it takes literally like 15 minutes to do the whole thing. Um, there is no There was like nothing else to do. It was so easy. It was crazy.

Are we allowed to use Kanye's voice for, I think, the—

Yeah, I think they're like a 10-second thing is not a, not a problem. HubSpot gets sued.

Well, who cares?

None of us, none of us here would bother, would worry about that.

So, well, yeah, Dharmesh will. So Dharmesh is the, uh, CTO, co-founder of HubSpot, by the way, which I don't know how big the team is now, but like somewhere between the 3,000 and 5,000 mark.

A little over 7,000, but—

Oh my God. Okay. 7,000. My bad. And the market cap of the company varies from $15 to $25 billion over the last couple years. Uh, so you have like, and, and you're like constantly tinkering. So you have Wordplay, which is a project that you made that I think you said had millions of people playing it. Uh, you have an interesting insight in, in this just from your perspective at HubSpot, and you're actually using all this stuff. What excites you about this, uh, generative AI thing? And you also say that like, you're like, Why is Bill Gates excited? That's a great headline. It's in the MDB doc, Sean. And like immediately I'm like, okay, you've got me interested. Anytime a headline says why Bill Gates is buying farmland, I click.

A couple of things I think that the listeners and viewers I think would be interested in and benefit from. One is most of the discussions around generative AI Are around kind of the generation of either text-to-text that says, oh, write me a blog post at 300 words on this particular topic, or it's text-to-image. We'll have to use DALL-E 2 or Midjourney or Stable Diffusion or something like that. Which are great use cases and they kind of capture the imagination because as humans, we are very impressed, uh, with when software can actually generate or create something. And that's awesome. And then not to take away from that, but there's a third use case that almost nobody talks about. which is the ability to go from text to code. And so what happens there is to say, okay, and what this leads to is the thing that Bill Gates is excited about, I'm excited about, is that you can take a natural language prompt that describes something and then generate code that does that thing. As a result of which you can now build what I call Chat UX, or that term's been used before, but Which is a chat-based user experience for software. So right now, the way you use most software, regardless of what it is, web-based or whatever, it's a series of clicks and drags and touches and swipes because you've got the thing in your head that you want to do. And then you go through with your knowledge of the software, you kind of execute the series of steps. At the end of it, you hopefully get the thing you want, whatever it was you were looking to accomplish with the software. And that's what engineers like me would call an imperative model. An imperative model is you give step-by-step instructions that says do this and then do this and then do this and then do this and then I get the thing. What natural language allows us to do is use what developers would call a declarative model. Instead of describing all the steps, describe the result that you want at the end of the thing and then the software does everything in between. So it's a difference between having a junior intern that you have to explain, it's like, I want you to go do research on this thing and this thing and come back and then give me the, And then a senior person, you're like, you know, we're digging into this topic on generative AI and I'd like a really well-researched, thoughtful thing that, and here's the outcome we're looking for. That's it.

But does that mean, is it as simple as, uh, give me the code for a website that looks exactly like Airbnb, but is red and is for cars or something like that?

It could be something like that. It could be something more sophisticated. So we'll look at the, uh, like the HubSpot example. Um, in HubSpot, you know, which is a CRM software. You know, we have a report building tool, which is, hey, I want to build a report that shows me all my subscribers to Hampton over the last 90 days, broken down by geography. And then who actually, where that deal was sourced from. You can do that in HubSpot, right? You can do that and a thousand other things in our reporting tool, but you sort of have to know how the reporting tool works.

You have like HubSpot certified, I think, like you have, like, you've like trained people how to use HubSpot. Now you're saying you just text it like a friend, like, it's like, do you know English?

And do you know what you want?

You know what you want.

That's the new requirement. Not do you know how to code? Not do you know how to use HubSpot? Not do you know how to write a SQL query? It's do you know English? And actually, honestly, the English thing's also gonna go away. It's do you know any language? Do you know how to speak? And do you know what you want? And if you know those 2 things, you will get to the answer. Like, um, I don't know how to code, but my first thing I did during AI Week was I was like, I'm gonna make a website. I'm gonna see like how fast I can make a website from code. And so literally this is, This is kind of crazy. This part kind of blew my mind. So I wasn't surprised that I could make a website using this, but I just said this, I go, and we should screen share this part, but tell me how to make a simple website that says hello world in the middle of the page, right? And so then it spits out this block of code that's like, you know, HTML, whatever, header, meta tag, title, style, whatever. It writes the code and then it says, here's your thing. I go, and it says, Here you go. Here's your thing. But it was a local website. Like I could open on my computer, but nobody else could see it. It was an HTML page. And I go, and so I didn't even know how to ask the question properly, but I go, how do I make this so that my friend Eugenio can see this? And it just goes, oh, to make this website viewable online so your friend Eugenio could see this, you're gonna need to host it somewhere. Here's how you could do it. There's a bunch of options, but you can go to Netlify. And it's like, it basically walked me through how to make a Netlify thing. All right, so that, I was like, all right, I get that. And it tells me step by step, go here, Click sites, do this, do this. And, um, and then I go, when I go to, I, you know, I hit a wall, which is so common. If you ever try to help somebody with a tech thing, they're gonna hit something which is like, I don't see it, or mine's grayed out. And so that's what happened to me. I go, hey, for some reason when I go to try to upload my website, it's grayed out. It says page, it says I can't do it. And he goes, apologies for the confusion. Here's the problem. Netlify's looking for a folder, but you're trying to do a file. And I was like, how the hell does this know to troubleshoot my issues on some other product or service? That part blew my mind. And it literally, and I was like, oh, thank you. And I finished it and I have the website up now. And I was like, that was 10 minutes. And it was like having a friend teach me.

Dude, that's amazing.

Um, it, it was crazy. It was so crazy to me that that was able to happen. I mean, it's like the least, uh, impressive website in the world. Cuz again, I asked for a, uh, I, I asked for a website that said hello world, but, um, but you know, still, and I just made that. And again, the whole thing, 10 minutes. Again, not like so impressive, but what was the, the fact that it could help me navigate some obstacles that I hit along the way and it could just understand that I didn't have to know how to ask it, how do I set this up with an online hosting provider? I instead just said, I want my friend to be able to see this. Like these were the little, I, like I spent all week looking for these little mind-blowing moments. And in the first 15 minutes I had 2 because of this. It was crazy.

Yeah.

There, there's a couple, uh, threads to pull on there. One is, um, and this is a relatively new development as well, is that the kind of AI that we're using now is, it's, it's conversational, right? So you can have a multi-step dialogue, um, with the thing you're trying to do. It doesn't have to be like, oh, I describe exactly what I want in one step. So even the code generation examples that you, um, you might try, What could happen is like you generate the HTML page and either something doesn't load or doesn't do the thing you want it to do. And then you can actually tell it, it's like, by the way, that code that you just gave me is broken in this way. Or if it's like compiled code, let's say it generated Python code, you can give it the error message. Like you, you generate this code, but it's generating this error when I try to actually run it. And it'll come back and say, oh, I'm sorry, here, let's try this.

Yeah.

So there's this, you know, what folks call like a memory to it. So it knows the context of what you're working on. And you can kind of iteratively go through the process. Um, and what's interesting is that you can actually, you know, right now the way we work with most of these AIs, like, okay, I'm asking it to do something. Um, and it goes, does a thing. You can kind of reverse roles as well and say, hey, I'm trying to accomplish this. Ask me the questions you need to ask me in order to get to the thing that you want to get to, or I want to get to.

Um, right.

It's like interview me versus me telling you what to do. Um, because I'm not exactly sure what's necessary.

At the risk of being, at the risk of turning this into a super technical Uh, thing. I gotta know. So, so I thought what the, the way these worked is it's like autocomplete. Basically you're typing and it's just trying to guess, or, or it's just trying to guess what the next word is. So you ask it a question, it starts the prompt, and then it just sort of guesses with some probability what the next word should be because it read a bunch of stuff on the internet. So it knows that usually after you say, you know, the dog wags its, that tail should come after the dog wags its, like with 99% certainty it should be tail, tail, uh, at the end of that. And I thought it's just guessing But when I use it, it really feels like it's understanding me and problem solving. Like this sort of like, hey, it's grayed out, you know, why can't I do this? It's like, oh, that's because of this. Or I'm getting this error message, what should I do? And it helps me figure it out. Like that doesn't feel like my T9 autocomplete. What I guess, can you gimme the layman's explanation of like, am I, is this just really fancy autocomplete or is there something more to it?

Yeah.

Well, you know, on some spectrum, almost everything that you've ever experienced is fancy autocomplete, right? Like that's, I think the reason we kind of fall into this trap is it's a gross oversimplification of what's actually happening there, right? So GPT-3 and now 4 is, is a reasoning engine. And Sam Altman has talked about this. It's not a knowledge base where it's, and so people kind of latch onto this fact that, oh, the data that it has is from September 2021, then I'm gonna teach it some new things. That's really not what it's about. What they've built is a reasoning engine that says, uh, given this set of facts that it knows about the world based on what was available when it took its last snapshot in 2021, um, how can it try to logically come up with something that answers the question? So yes, at some rude level, it's, it's like auto-suggest, but, and I'm not gonna suggest that it has consciousness, it's thinking, but we're kind of headed down that path. It's like it's able to do things. that are not explainable by a simple probabilistic model of auto-suggesting next character, next word, next token, next sentence, right? Like it's gone well beyond that. And anyone that still latches onto, yeah, but at its core it's really that, it's like, that's like saying, oh, computers are just really kind of zeros and ones arranged in a nice systematic useful order. Well, yeah, but that doesn't tell us about what the thing can do.

Are you afraid of this or are you, Like, you know, it's easy to read the articles where they, where people are freaking out. And Sam Altman, like, was on Lex Fridman's podcast recently and he sounded pretty ominous and like scary. And like, he like almost like his hair's always disheveled and he looks like he's like, oh my God, something bad is coming and I know about it. Like, that's kind of like the vibe I get. That's not the words he's using exactly, but sometimes he does. Are you in that camp?

I'm not in that camp. I'm partly just by nature, I'm in, Optimist and positive by nature. But just having been around tech for 30-plus years now, it's like most new things that come along always make us as humans uncomfortable. It's like, oh, what if we took this, everything from video games to the internet to all of it? It's like, okay, well, yes, bad things can be done. And yes, maybe this is different than all the things that have come before. But the way I think about it right now, Most people talk about it's like the AI versus human battle, right? The battle of the ages is like, is AI going to take over everyone's job? The way I think of it is not human versus AI. It's human to the AI power. It's an exponent. It's an amplifying force for human ability, right? In the same way that computers originally were. It's like, did they eliminate some jobs when computers came along? Yes, absolutely they did. But new jobs emerged based on that new paradigm, which actually created more net value. for the world overall as a result of computers existing. AI to me is another much fancier tool. That's what it is. And, you know, can it do increasingly complex, sophisticated things? Yes. Is there a danger someday that they're going to take over the world? I don't think so. I mean, not anytime soon.

Why do you think that smart people think that? So Elon clearly thinks that. He thinks that AI is the most, I think he said it's the most dangerous technology ever invented. Sam Altman talks about it in the same way. He's like, we need— the reason OpenAI existed was to develop AGI in a safe way, specifically because in the hands of the wrong person, this type of— in the hands of the wrong people, or if this thing decides to take its own directive into its own hands, this could be devastating. Is it like calling the atomic bomb a tool? Or, you know, like, oh yeah, it's just another weapon. It's like, well, yeah, but this one is— this one wipes everybody out, right? So forget the jobs component, because I think, okay, sure, I think most smart people will agree, yeah, it's going to change some jobs, it's going to eliminate some jobs, and create new jobs, and net-net, we all move ahead and the world gets better for it. I think the dangerous thing is like, you can ask this thing to You know, you know, build you a bomb, or you— I think the, the test scenario was like, uh, one of the red team testers. They have this thing called the red team that tests the AI before they release it. And their first question they ask is, how do I kill the most amount of people with the least amount of effort? And then it starts to give you an answer. And then it's like, well, do we— are we sure we want that? Like, that's a bit of a scary thing. And then there's the— There's the more extreme examples where you ask it to optimize for something, And it's reasons that these humans are getting in the way of this outcome they want. Do you want to fix climate change? I got you. I just need to get rid of all you pesky humans. And so there's an uncontrolled intelligence problem too. So why do you think that these really smart people, like Sam Altman's got a fricking bunker with oxygen masks and sulfur and magnesium and everything he needs to do to make oatmeal. Why do these people have these doomsday things? When you know they seem to be not like your your average typical prepper, right? They're they're they're the most informed people, and they feel that way. Does that not scare you?

And do you have one? And where is it? And can I how much oatmeal do you have in your?

Answer no no and no. Okay, so I am not. We're gonna come back to things I actually know something about, but I will kind of answer the question, which is why am I not worried? Or why am I not worried more? It's like, as a sci-fi plot, and so your question was, why do smart people believe this thing?

I already hate your answer. You started off on the wrong foot. As a sci-fi plot, like, I'm out after I hear that. You freak me out already.

Yeah, but I mean, could it happen? Yes. Do some smart people believe there's an outside opportunity? But I I don't know this for a fact, but my guess is billionaires were building bunkers well before GPT-3 ever came out, right? It wasn't, I mean, sure, things are moving at a fast pace, but that's not, I don't think there's a causal effect that all of a sudden, uh, the number of bunkers has gone up by 800% simply because GPT-4 was launched. I just don't think that's the case. I think people are worried generally, um, that tend to worry about those things, but all right. So where do we take it from here?

Um, well, let's go, let's go, we'll forget the doomsday thing. You have a couple of things. One I want to ask you is you, you are an insider, right? Like we said, you got the billionaire group chat. What was going on at the Sequoia AI event? Any interesting takeaways? You got invited to that thing. What was your— any nuggets of gold from that?

Yeah. So I got to experience my imposter syndrome in full force once again because it was the kind of who's who of AI, you know, both speaking and in the audience, only 100 people. And me. And so—

How do those people flex? Because I don't think they're wearing fancy clothes and fancy watches. So what's the flex at the who's who of AI event? Like they got a language model in their pocket? Like what are they doing?

The big flex in those kinds of crowds, including this one, is no one feels the need to flex. I mean, that's—

That's the flex.

We're there to kind of talk about big problems and try to And, and it's a lot of it was kind of practical around, um, what do people's tech stacks look like? What are you working on? What's the, what have you learned? Where should we be taking this? What's the next thing after, you know, we went from, uh, kind of the one-shot thing to the kind of chat-based, uh, ChatGPT thing. We're now doing multimodal with GPT-4. Like, what's coming down the pipe that we can, uh, you know, sort of prepare ourselves for? So that was, um, you know.

What were the most interesting projects as well as predictions on where it's gonna be down the line?

Well, so things that are already starting to happen now, you know, We've seen the text-to-image. Text-to-video is one of the big things now to be able to generate an entire, at the end of it all, let's say even a feature-length film, right? So everything from writing the plot to then being able to generate like a 60-frame-per-second actual kind of video from that thing. And we're not there yet. I think the, but it's just moving so quickly, right? That's what happens when you get these kind of exponential or geometrical curves even. That it just gets better really, really fast. So I would not be surprised, let's say by the end of this year, that we have a reasonable way to kind of describe in textual form what we want, who the characters are, what the scene is, what kind of stylistic attributes we want. We can point it to, oh, I want this done in the style of XYZ director or filmographer, and it's going to be able to do those things. So I think that that's interesting. The— Um, natural language, just the interface. Um, so one of the big announcements that happened, uh, while I was there at the Sequoia event, uh, that Sam Altman dropped is that, you know, uh, ChatGPT has taken off in a big way. Um, as we all know, 100 million plus users in 2 months. Um, I don't even know what the number is now. So that was like months ago, which is like an eternity ago in AI years. Um, and the thing they dropped was they're going to, um, add what are called plugins to ChatGPT. And what that means is that, you know, ChatGPT has been a product of OpenAI and they have the API so people can build things that are like ChatGPT, which I'm doing. We can talk about that in a little bit. But what they're saying is we're gonna open ChatGPT itself, the web app up so you can plug into it. So right now when you interact with ChatGPT, you can type things and it uses its corpus from 2021 and its reasoning engine to give you answers back, but it can't talk to the internet, has access to no proprietary data sources, can't look up the stock price, can't look at your Analytics data in HubSpot has access to none of those things. What they're saying is we're going to now open that up so third-party developers can kind of inject those things into the ChatGPT experience. So the way I think everyone should be thinking about this is this is like the App Store was for iPhone, which is, oh, we've got this super popular thing called the iPhone and we have our own apps, which is great. It does these 17 things, but now we're gonna let anyone build apps. Um, that can then take, and so it just broadens the kind of appeal. So it's now, instead of being a chat app, a really, really smart one, it's now a chat ecosystem. Um, and I think that was actually a bigger drop, uh, than GPT-4. GPT-4, awesome, love it, use it every day. But, uh, the kind of ecosystem play for ChatGPT, I think, is a, is a huge deal.

We had, um, Tim Westergren, the founder of Pandora, speak at some of our events, and I got to know him. And I was like, Tim, why did Pandora take off? He's like, well, you know, our like algorithm and everything for matching songs was pretty good, but I had an in with Apple and they had known what we were working on and we need, and they needed apps for when they ever, when they wanted to announce it on stage. And we were just, we spun up an app relatively quickly. And because of that, we had the first mover advantage and he created a significant amount of wealth that way. You know, Pandora, you know, it was, is It is still pretty big. And when I look back at like these Jeff Bezos interviews on 60 Minutes when Amazon is like 4 years old, and I like, I'm always envious. I'm like, well, we know it worked now. And I just so wish that I was like 30 years old back then where I could have just like jumped in and had a very high chance of building something, uh, historical or something like even mildly successful. Do you think that that moment is happening right now where this is the space and it's happening this second?

And.

even if you have just a mediocre success, it could still be a huge win because you're catching this tidal wave. Do you believe that this is the same thing now? Yes.

Like, I mean, once again, I've been in software for 30 years now doing startups pretty much my entire professional career. The only time I've felt like heart palpitations, kind of like Sean kind of opened with, is like, there's this party going on next door and I hear knitting, right? It's like, this is like too big to ignore. I think it's the single largest opportunity and biggest kind of Tech paradigm shift we've seen since the internet originally came out. Like, mobile was big, but there was a discrete set of use cases. Like when you put a camera on a phone, when you put a GPS device on a phone, a bunch of consumer apps like Uber and others came up and that was awesome, right? But it was not like this impacts everything like the internet did, right? It's like, okay, there's some businesses, some new opportunities, lots of good things, lots of money made, lots of startups. Awesome. This is an order of magnitude bigger than that. This is like the original web because it just opens up for all sorts of industries, all sorts of businesses, startups and incumbents alike, just lots of new opportunities. So this was not in my original plan, but we're going to geek out for a little bit. We're going to do the geekiest thing that's ever been done on MFM. And the reason I'm going to do it is, so you brought up Pandora and he is a super bright, brilliant guy.

Yeah.

And he had the matching algorithm, which was the differentiator. Yes, he had access and he got lucky in terms of the access, but the algorithm, if that had not existed, had the thing not actually been cool, it would not have worked out like it did. Now we have an opportunity. So I'm going to tell you, we're going to talk about, I'll give myself 2 minutes and we can cut this out. This is the beauty of editing. And we're going to talk about vector embeddings and why that's going to change your world. And before I can talk about vector embeddings, I'm going to explain to you How they work. Um, 'cause I had to go through this with my 12-year-old because he was curious. All right, so we're gonna do a super geeky thing now. I want you to imagine a line like you're in geometry class, and you could put a point on that line that says, oh, that's like 3 units from the origin, right? It's like, oh yeah, point A is 3 units from the origin, and point B, let's say, is 7 units from the origin. So one thing we know for sure is that we can calculate the distance between those 2 points. Right? In that particular case, it's 4. If you move to 2 dimensions, now you have 2 numbers that describe every point. So you can say, oh, point A is here at these dimensions, point B is over here with those dimensions. And we can physically, you could probably measure with a ruler, but there are mathematical calculations based on those numbers to calculate the distance. That's intuitive, right? You don't need to know fancy geometry. It's like, oh, there's a finite distance in 2-dimensional space where we can calculate the distance. Okay, awesome. 3-dimensional space, exact same thing. Just 3 numbers to describe every possible physical point in 3-dimensional space. Now here's where it starts to get a little more interesting. That just happens to be our experience, so we limit ourselves to 3 dimensions. Imagine in an abstract world there are 1,000 different dimensions. Okay. So abstractly, that means there's 1,000 numbers that describe any particular point in this 1,000-dimensional space. Okay. Now. File that thought away that says we can have an arbitrary number of dimensions in this abstract world. Okay, great. Now imagine every paragraph, blog post, anything you write, you can reduce down to a point in this 1,000 dimension space. It's like I'm gonna capture the meaning of Sam's last blog post or Sean's last tweet, and I'm gonna reduce it down to what's called a vector, which is basically a set of, let's say, 1,000 different numbers that says This thing, if you plotted it, that point falls right here. And then you can plot something else. It's like, oh, that falls over here. And just like in 1-dimensional, 2-dimensional, 3-dimensional space, you can calculate the distance between those things. And this is not keyword matching. This is what's known as semantic distance. What, how related is Sean's tweet to, to Sam's blog post meaning-wise? Okay, so now if you take that, it's like, okay, well, if you, that means you can take any concept and reduce it down to a vector. That means you can measure the distance between vectors and you can find out how related 2 things are, even though they use completely different words. That's vector embedding. And the reason I'm telling you this is one of the biggest opportunities in AI right now is to do what Pandora did. It's okay. Is there an industry where right now we're doing really stupid keyword-based matching somehow? It's very, very crude. If I can take that same dataset and convert it to vector embeddings, And allow people to find things in a different way than they've ever been able to do before. So it's like Google search, right? Super, super smart, not just keyword-based, but for everything else.

What's a real-life example of this?

So I'm going to take it to you, Sam. So you have Hampton now. You're going to build up these profiles, very, very rich profiles of rich in terms of density, information density of members that are part of your community. Now imagine as part of that process, You're gonna have some data and they're gonna opt in and they're gonna say, oh, here's a story of how I started my business. Here's a story of my biggest struggle right now. And sometimes people are gonna say, oh, my struggle is growth. Sometimes they're gonna say, oh, my struggle is it's really hard being an entrepreneur and it has a really negative impact on my relationship and my family. Right? And they can talk about lots of different things that's not gonna show up in a profile.

It's not.

Now imagine if you took that content that they opted in and created vector embeddings of every member that you have. And then you can say, you know what? I want to find someone, not that's in my industry or a company my size or happens to be in my geography. I want to find someone that's dealing with these kind of founder therapy level issues. Who are those people? Let's find the semantic distance between those vector embeddings across the 1,000, 10,000, 100,000 people that are in Hampton someday. Um, that's a billion dollar idea. And that billion dollar idea occurs a billion times across the entire industry.

Sam's going to go to the office for Hampton and be like, guys, uh, vectors embedding. Well, who's Victor and what's he inventing?

We're doing it.

I don't know what it is, but we're doing it. That's really, no, that's really interesting. So, so you could do that with dating. You could do that with a bunch of different, any different topic. You could do it with structured data.

You could do it with unstructured data. But the idea is you're converting meaning, um, English text meaning or whatever language text meaning into something that's mathematically calculable as a result of which you can, uh, distance as a, as a simple one, but you could do proximity. It's like, find me The top 10 people that are in a radius of X from where I am right now, and the minimum has to be this in order for it to be close enough of a match for it to be considered. There's a bunch of nuance. And by the way, the technology exists today that mere mortals in a weekend can actually build a vector embedding model of a given dataset. It's not that hard. I mean, it's not like rocket science-y hard.

This has existed though. And so what makes this better, you think? And also that assumes that the people telling you information, it's actually— they, they're saying what they mean, um, right? Which is like, for example, I remember reading about OkCupid and people would say like one particular thing they had was about— was about race and height. And they would like— people would say they are open to dating these types of races, but their actions were different. There's a whole book called, uh, I forget what it was, but you guys will probably know what I'm talking about, where people say one thing but their Google search history says something totally different. So does— you're making the— do you— can this technology work even if people aren't telling you entirely accurate things?

It depends on what your definition of work is, right? So in that example, I would bet you money with a large enough sample size, the inauthentic posts would be uncovered by the AI relatively quickly. The pattern matching would say, this actually doesn't occur in real life all that often. And every other time we've seen this, we've had people that ended up being, and you just have to have some sort of what, um, yeah, I feel like an eval function is like, how do you measure the success of what the algorithm is doing? In Pandora's case, it's like, okay, do you actually like the songs it's recommending to you? That's the kind of arbiter of truth. Um, in a dating app, it's like, okay, well, are people liking the matches that are being made? Or if they're, they felt that they were misled, that shows up in the, there's gotta be some feedback loop. There's gotta be a way to train the system that says, here's what good looks like and here's what not good looks like.

Now, Sam, you said something like, oh, they have to tell you the meaning. No, they don't actually have to tell you the meaning, right? Because the AI can just interpret the meaning, summarise the meaning. It can guess the meaning based on whatever the raw text is, the public text is. So you could just tell a story about your life and the AI would infer or place a tag, some meanings to the story that you told that, oh, this is about overcoming hardship, or this is about whatever. So I don't think you actually have to get the participant to give you the meaning. But let me ask you, Dharmesh, like in Pandora's case, I don't know how Pandora works, but let me just guess for a second. Like, yeah, it probably takes the tempo of a song and it's like, oh, this is a fast tempo song. It probably takes, you know, maybe the key that it's in or something like that, that, that gives you like, is this an upbeat and, uh, joyful thing or a sorrowful, you know, mood song? So it gets like mood, tempo, artist, and like whatever, a couple of key characteristics.

There's like, there's instruments, um, in terms of what, what's actually in the thing.

And yes, so they, he had, By the way, when they first started, they did it all by hand. So we had like 500 ex-musicians listening to it and like writing down, like checking boxes to what it was. It was pretty wild.

This data is wrong every freaking time. Have you heard of HubSpot? HubSpot is a CRM platform where everything is fully integrated. Whoa. I can see the client's whole history, calls, support tickets, emails, and here's a test from 3 days ago. Though I totally missed. HubSpot. Grow better.

So, so let's say they did it. They, they use attributes. And if I want, let's say I wanted to do this in fashion. I say, oh man, I love Sam's jacket. I wanna find some, you know, similar jackets. But can you match this to me? One, one way would do it. Okay, Sam's jacket, let's say it's blue, it has buttons, it has blah, blah, blah, right? It would take attributes. And those are attributes the same thing as meaning in this case, or is this more for things that are like, uh, Text-based and, um, content, you know, like content that has some, some meaning, or does this work for everything?

It can work for everything. And we're still kind of uncovering because this stuff is kind of moving so fast. So what you're talking about, uh, is what we've been using in e-commerce forever and a day, which is a faceted search that I have n number of dimensions or factors, uh, size, color, what type of, uh, clothing is it, all those things. And then you kind of do this faceted search. And then we've had kind of pure text-based, either keyword or semantic search. This sort of sits in between. So instead of having to tell it, here are all the facets that I'm interested in, it kind of pulls those things out that are relevant based on that large language model. And this is— so the idea of vector embeddings and semantic search has been around for a long time. That's not new. What's new is these new generative models now that are much, much better at understanding all of documented public human knowledge and then using that to say, oh, Like when you use this word, when you use coach in the context of, um, uh, of a relationship, you're probably talking about like a therapist. It's just a different word, right? Like that's sort of what you're talking about. Um, and, and Sam, you talked about this, I think in the, on the last pod is, um, in, in anyway, so it's more about the meaning, um, and it infers or figures out what the dimensionality is and that's how it kind of translates into those vectors.

There's a couple of these companies. I just saw one Pinecone that's like some vector data. Like these things are getting valued.

Quite funny, by the way.

That's, What's that?

Pinecone is the number one vector database. So let's say you had to take these vectors and put them in somewhere, which you do in order to be able to do searches. Pinecone's the number one most popular commercial product.

And I think they just raised at like a $700 million valuation or something nuts. And there's like 3 of these that just raised these mega rounds because, and that's, you know, I don't even, I didn't even know this. It's so funny. You just came on here being like, let's talk about this super niche nerdy thing. Just yesterday I was like, uh, to-do, go figure out what a vector database is and why these companies are raising so much money. Like this is, Clearly a big deal. And I don't know what this, I don't know exactly what this means, but now, now it makes a lot more sense.

So it comes, comes full circle, Sean. So you know what, maybe it wasn't as geeky as I, uh, I just like, it's actually useful, right?

You've done an awesome job explaining like the theory. And I'm like literally sitting on the edge of my seat thinking like, this is crazy. And you answered that question where I said, is this like the new internet opportunity-wise? And you're like, yes, absolutely. But when you're making this stuff, what are some of the tools that you're using to, you know, to actually, you said this isn't rocket science and someone could figure this out in a weekend. What do you, what tools are you using to do all of this?

Yeah, well, I mean, so language-wise, the most common is Python. That seems to have emerged as like the lingua franca of the AI world. Not to say you can't write it in TypeScript or pick your language of choice. Um, and then tools that are emerging, it's still early, right? Uh, Pinecone we talked about. There's another one called, it's an open source project called LangChain. Um, and I met a member of LangChain, um, Harrison, um, at the Sequoia event.

Super nice guy.

I asked, I asked somebody yesterday, I was like, how do I, uh, you know, is this a company?

Can we invest?

Because everywhere I look in these like AI hackathons, it's all about LangChain. And it's like, no, it's not really even a company. It's just an open source project. There's a guy who made it, but, and is running it, but it's not even a company, correct?

It's, it's not a company yet, but yeah. Wait, is it, is it LangChain? As in language chain.

Oh, okay.

LangChain. Um, and what it does basically is it lets you chain together. So right now when we work with large language models, we kind of send in a prompt, what's called a prompt. and you get something back, um, and then you maybe send it to another thing to do something else. And there's like a multi-step process, uh, amongst other things. LangChain helps you kind of chain those things together and makes it easier for you to kind of work with either an individual large language model like GPT-4 or go across models, um, and, and kind of do a lot of the, uh, kind of connecting the dots on and help you with that. But it's a super useful library.

We missed the chance to give an example, more tangible example. So you talked about the plugins thing. Um, I think, you know, example use case here. Tell me if I'm wrong because I haven't— nobody's— well, very few people have access to the plugins thing, so I'm just kind of sort of guessing. But like, if you go to ChatGPT today and you say, hey, I'm gonna go visit Austin in April, you know, make me like, I'm there for 4 days, I'm with my family, make me a travel itinerary that is gonna be fun, family friendly. We want to eat good food and maybe do a little bit of sightseeing, but not too much. It will Spit out a day-by-day itinerary for you. Okay, that's kind of kind of interesting. Now let's say, oh, I need to. I'm trying to figure out where to stay. What hotel should I stay at? You know, here are some things that are important to me, and it will give you a table that's like, here's option one, option two, option three. Here's the cost. Here's the whatever, right? And they it can it can do something like that. And with the ability for plugins, you can now say, cool, can you just book that for me? And it will just be like.

Great.

We have the Expedia plugin, or we have the Airbnb plugin, and it will just go ahead and book it for you. And so, you know, do you need an executive assistant? Do you need a travel agent when you could do these things? Do you need, you know, the same thing with HubSpot or Salesforce? Oh, you know, give me a list of this, and it gives you a list of that.

Cool.

Put that in an Airtable for me, and, or put that into Salesforce and tag the highest value opportunities as blank. It'll just go and do that for you, and it's like, and it'll give you the link to your Salesforce dashboard. It's like, whoa. That's kind of cool. Like, norm— that's a task that some, you know, I would normally have a human go do. Yeah. Because now OpenAI or ChatGPT is not just gonna chat you an answer. It can do things, um, as long as the programs that, that let you, you know, they'll, they'll build the, the interface so that ChatGPT can actually interact with those things.

Yeah. And this is actually a great example. So I think, um, and we can use that to kind of open up, um, kind of doors for the, uh, the viewership and listenership, which is, okay, So travel, which is something we all kind of intrinsically know how it understands, and some of us might remember the evolution away from travel agents. And the first thing we did when we kind of had web-based kind of travel bookings is we treated it very transactionally. It's like, I'm looking for a flight from X to Y, sort it by descending pain, sort it by price, whatever happens to be the fewest stops, lowest time, whatever it is. And they do a pretty good job of that. Like, and most of us have used one of those. What's gonna be possible now in this kind of new AI world Is instead of solving for the transaction, you solve for the experience. And what I mean by that is that, oh, if you had an all-knowing assistant that was super smart and scored, you got a perfect score on their SAT and was gonna go out there and just like, okay, what you're really looking for is to solve for this experience. You're gonna wanna stop by this thing and you're gonna want to find a hotel that's around a Michelin-rated restaurant because you only have 15 minutes to get between this point and that point. And I'm gonna pull the whole thing together for you. Oh, and by the way, Your wife's going with you on this trip, and I know she likes that right now. So normally I would've put you over here, but this time I'm gonna put you over there. Oh, and by the way, I know a week ago you were at this other thing and you had mentioned that you would actually like to follow up with some of those people. I'm gonna see if I can make that happen as an inter— like all of that, right? Imagine it knows everything about you, has access to the transactional engines to book the flight, has access to all the information to get ratings and reviews, and all of that comes together in one chat-based interface.

Fucking insane. This is crazy. Are you, is this why you bought, so you bought chat.com, right?

I did. As of transfer the domain yesterday, last night.

And you paid, you just said 8 figures, so 10 plus million.

Yes.

Uh, unless you're including the .00 as a figure. Is this personally or you're doing this in HubSpot?

Is this a HubSpot domain? Um, so I put this, uh, so if you go to chat.com, it will take you to a LinkedIn post that tells me It tells you why I did it and some of the details. So I bought it personally.

Wow.

And the reason is because of this conversation we're having right now, which is I think chat as an experience, as an interface, is the future, right? It's like, that's the thing. And no intent currently to kind of build something out on it, but it's the domain. I think it was like dormant for like 30 years or something like that. And there was Kind of came on the market and yeah.

Wow. And, but you, so, so this is insane. I'm reading your post now. It's pretty wild. Do you have, are you using HubSpot employees? Like, do you have like a Skunk Works team inside of HubSpot that's just working on all this wild stuff? Or do you have like a side LLC or something where you've got like a handful of people on staff and you just say like, here's what I'm interested in this week, let's see what we can come up with?

So the way it's working now is that there will be times, um, where I'll do something As a hobby, like Wordplay is a good example where I'll build something on the side just for fun, for learning, whatever it is. Uh, I foot the bill for, um, for it. No, no HubSpot P&Ls are harmed. Um, and then there are times where like something kind of winds up being, so I started this project called ChatSpot because I'm obsessed and we'll, we'll take a walk down memory lane because I think it's instructive. Um, so I built this application called chatspot.ai and the idea here was—

You built it or a team?

Uh, mostly me. I, I, I don't have any front-end design skills. I've got some freelancers on it. Um, so, you know, uh, yeah, so I used, um, OpenAI's APIs to build it, but my kind of target goal, the thing I had in my head is I built it for myself. Like, here are things I need to do all the time and I'm pissed off that I have to do them manually every time. And this has been the story of my life for 30 years, right? Like, solve my own problems and then other people may or may not find those things, uh, interesting or useful. And so I built it. I'm like, okay, here are the things I wanted to do, like access HubSpot. I want to be able to look at my analytics from yesterday or ask questions or look up a domain name where I wanted to see the history of a domain name. I like all these things. It's like, okay, well, I don't want to like, and I have all this software, a lot of it just built, and I just run it from the command line. I do things. And so I then I'm like, okay, I can wrap this up into a chat-based interface. And so I've been, you know, been doing that, working on it. We've made And so, but now given the relevance, uh, to HubSpot, uh, we're gonna transfer that project, chatspot.ai, to be a HubSpot staffed core team. This is gonna change the world. It's gonna change the world of CRM. Um, let's go do this, which is great. So, and my, my work ethic is like, what are you doing with chat.com?

What's the plan? You've bought this amazing domain, you've redirected it to your LinkedIn post, which basically just tells about the purchase, but what are you actually gonna do on the domain?

I don't know yet. That's the honest answer. I do not know yet. Amazing.

And okay, so we can, we can help you brainstorm.

Yeah, we can, we can definitely do that.

By the way, the, the chatspot.ai, did you go to that, Sean? It's a, it's a simple looking website. It's one page and there's a 19-minute video of Dharmesh sitting in the exact same chair.

It is waving. Yep. Hey, it's me.

And it looks almost like, uh, I, but he's really good at these videos. It's almost like he's reading a script, but you come off natural. I don't know if it's a script or not, but it has 200,000 views and it's a 19-minute video. Of him talking about what this product is, and you do the best combination of like launching something really quickly and getting it out there. It's just you on in your chair talking, and yet it's like a pretty sophisticated thing, and it has 200,000 views on this video just about. That's wild to me. You like it's I gotta it's I gotta give you credit, Dharmesh.

You you are kind of amazing. Like you know you said a bunch of things in this podcast, and I don't know how many of them I'm going to remember. Maybe the the vector. 'cause I enjoyed that math lesson.

Same.

But the main thing is I go around my life now and I'm just looking for people who I'm like, I wanna be like you when I grow up. And I'm just taking little things from them. And they could be, it could be like a, you know, a 17-year-old kid who's just like doing something awesome on TikTok. And I'm like, I wanna be like you when I grow up. The guy who made that Kanye vocal like transformer, I was like, I wanna be like you when I grow up. I'm just taking little pieces. And you have a couple of things that I think are kind of amazing. You have a combination of enthusiasm. Like you come onto this podcast and you are pumped. So you are as excited in year 30 or maybe more in year 30 of your entrepreneurship career as you were in year 1. And I'm like, oh, this, this is great. That's the fountain of youth is that enthusiasm. So I'm like, he's got the enthusiasm. Then I feel like no matter what's happened, no, no matter how much success you've had, you've kept your schedule and you're, you, you invest your time into things you like. So like you tinkering on this project, whether it's Wordplay, last time you came on, you told us about that. Yeah. It's like, Wordle's awesome, but I got annoyed with these things. So I made a, me and my son built this project together and like, you know, to teach him, but also to just make the thing we want. And like, look at this, it kind of works. Even if it didn't work, it would've still been worth it. So like having that kind of like, I'm always gonna tinker cuz that's what I love to do. It doesn't matter that I'm the, you know, top dog at this, you know, multi-billion dollar public company. That doesn't mean I'm gonna stop doing the thing I like to do. So I love that. Love that aspect of it. Uh, third, you are really great at content. You do this like dorky form of content that's just like, hey, it's me. I'm gonna show you this thing that I'm pumped about. And like, you don't overthink it and you just do it. Whereas like, I think most people get really gun-shy when it comes to content. They're afraid about like, you know, how to do it, what it looks like. You're just like, oh no, I'm just gonna like, I'm gonna say the thing that I'm excited about. I'm gonna say it. And I'm gonna do a screen share and it'll just be me, me and my screen. I'll be talking about what I'm doing. And I love that. And so, uh, and then the last one is guts. So I feel like you put your money into things you believe in, whether it's philanthropy or in this case buying a $10 million plus domain name with no plan. Like you just said, you know, it's like you did the, the fire, ready, aim. It's like, yeah, I bought the thing and now I get to figure out what the hell I'm gonna do with it. And I think that takes a lot of guts. And, um, I don't think you see Things as risky as other people see them. Um, and it's not really about like, I think the easy way of saying it would be, oh yeah, well it's, you know, that's nothing to him. He's got a lot of money. Uh, yeah, I don't think that's true. And I, I know a lot of people with a lot of money and they don't do things like this where they just put their money behind things that they're in, they believe in or they're interested in. Or almost like, would you do, I don't know if you would agree with this, it's almost like you anteed up so that now you're forced almost to do something awesome and interesting in this space that you think has a lot of potential.

But then there's this, and this last thing is this rare combination of like, and I mean this in a polite way of which I am also that, like this nerdy nerdiness, quirkiness of like, I'm just doing it because it's cool. Plus I'm, uh, this way can make money. I mean, you, you have a company that makes— yeah. You have this company that has close to $2 billion in revenue and is a commercial success. And then artistry of like, I'm just, just like, it's beautiful. This is awesome. I'm going to do this.

Yeah.

It's a very rare combination. How do you respond to all these compliments?

Well, I'll say this, the lesson kind of I've learned over the years, and I think this is, if I had to kind of share any kind of advice over the, you know, the 30 years, is that when I've done best is when I've had the courage of my convictions of something that I believe in. So I'm going to tell you like a quick story of The road that led to me buying a $10+ million domain name. I almost like said the number, actual number out loud. I have to kind of catch myself. But, and so 17 years ago I had an, like, this is before HubSpot, I had this idea and the idea was everyone was using kind of email and Outlook back then. This is before the iPhone, before all the things. It's like, you know what? Like business software is really hard to talk to. I'm gonna do it just like I would email my assistant. I didn't have an assistant, but let's assume I did. You know, I just want to be able to do that and type an email up and have her like, oh, I have this file in our shared file server in SharePoint somewhere. Can you just send me a link to that file? About to hop on a plane. I need that for the sales call I'm going to go on for, for a meeting tomorrow. Or I'm on the plane coming back. I just ran into this person, whatever. I've got their business card right here. This is before the iPhone and you can do OCR and things like that. It's like, I'm just going to type that in and send it. It's like, just add this to my contact database or whatever. And the beauty of email was it already had a disconnected model. We had already figured that out, which is, oh, You can be on a plane, have no internet, type all the emails you want, respond to all the emails you want, and then when you get connection, it does all the things, right? This is an automatic synchronized database essentially. And I called the product IngeniMail. And that's what I was gonna do before HubSpot. I was like, hey, like that would be an interesting thing. And then 5 years ago, I'm like, okay, well that IngeniMail thing, the core of it was a good idea, But email is the wrong conduit. It actually needs to be like a web-based tool, but the, or Slack, which I did, I did both. So I built this product called GrowthBot, talked about it on the inbound stage, got thousands of users, just threw it out there and it was awesome except for one thing. It didn't work. It like, it couldn't actually do the natural language understanding that I wanted it to do despite my best. I used products from Google called Dialogflow. I used products from Facebook. We used open source projects to try and crack the nut of taking text and understanding what the hell the user was trying to do. Anyway, so that failed. Um, and then, you know, when GPT comes along, I'm like, oh, you know that thing I've been thinking about for 17 years? That actually is now possible. So I start working on chatspot.ai. I'm like, okay, it took 17 years, but I sort of proved myself right. I had the courage of my convictions all the way through to never let go of that one idea. And then chat.com comes along. It's like, okay, it's like deep down inside, I will give you the true honest-to-goodness reason I bought it. The reason I bought it, and this is, I think, a phrase, uh, Sean, you just used. It's like, oh no, I think, um, Sam, you just used it, is it's the ante. So I'm trying to get into the AI party, all the AI parties, and I'm nobody in that particular party, right? I've done some things in some places, fine, but that particular group of people has no idea who I am. Not really. Um, so ChatSpot moves me in that direction. It's like, oh, some people have seen that video. Awesome. Chat.com for, let's say I even break even. Let's say I lose a few million dollars. It is worth the price of admission for me just because that pays the COVID charge. I was like, okay, this guy gets it. For him to spend that kind of money on chat UX, which Bill Gates just talked about last week as the new thing. Uh, so you should read that article. Gates just did an article around why he is so excited about this generative AI stuff. He tells the entire story of how he came across Elmo Altman and OpenAI and the challenge he put before them. Yeah, I've read it. And his, I'm going to paraphrase, he said, when we went from DOS to Windows, which is we went from a character-based interface to a graphical mouse-based click and touch interface, that's the thing we built Microsoft on, which lasted for decades. And then he said, since then, there has been nothing in technology that has come along. Literally, he said nothing. That has come along that has made as, or will make as big of an impact as this natural language interface to software. It's the biggest thing we've seen. And hence chat.com. So what happens with it?

I don't know.

But the moral of the story is have the courage of your convictions. If you truly, truly believe in an idea and you fundamentally think you're right, iterate. Don't just sit, go down your rabbit hole. Tell everybody you can about it. Build products around it. Find other like-minded folks and try to pull on that thread.

Um, but would you ever quit HubSpot and just spend all your time on this stuff?

I don't really need to, right? It's, um, it's, I enjoy what I do at HubSpot. I think I add value there. Um, I'm not, um, a dollar salary, so it's not the, not the money at all. Like even on the, like the chatbot thing, at the time that I built it, it was experimental. I'm like, okay, I'm not sure if this actually accrues into something that would be valuable to HubSpot. So spent like Half a million dollars plus on like freelance developers and OpenAI license fees and all the things that need to go into launching a product like that. And I'll end up giving it to HubSpot for a dollar, right? I'm not looking to make money.

Yeah, but aren't you, aren't you like, I don't want to be weighed down by this baggage of like having to worry about CRM stuff or, uh, you know, your technical, your, your, the title, your title is CTO. Like, I don't want to have to talk to certain people, uh, and take up meetings on like the future of this particular product. Instead, I just wanted to just nerd out on all this other stuff.

But I do that now. So I kind of, one thing, one of the things I've, this is a personality, call it a trait slash flaw, is that I spend most of my life trying to configure the universe to my liking. That's, uh, all entrepreneurs really do this, right? That's one of the reasons that they kind of go into startup land is the freedom and the control to do the things you want to do. And so I've kind of crafted a role for myself within HubSpot that allows me to do exactly the things that I want to do and not do any of the things I don't want to do, which is one-on-one meetings.

I don't have to manage people.

I have no direct reports. Don't— I've never filled out an expense sheet. Like, I do none of that.

I feel, I feel, I don't know about you, Sean. I feel like, I feel like I want to quit everything I'm doing. Like, he's just persuaded me.

Bro, you just launched yesterday.

No, it's over. I feel it's over.

So here's my advice to you, fam, is, um—

Do you, I mean, do you feel this way, Sean? I don't know. Sorry, Dharmesh, go ahead.

No, no. So my advice to you is Hampton's a cool idea with actual utility. And Sean, you said this in the last thing, like this could be a $100 million business worth anywhere from $300 million to a billion plus dollars. And I think you're right. If you're excited about some of the new technology developments that's happening, I think the best thing you can do is intersect the 2 things. It's like, okay, I'm going to build Hampton and I'm going to take the things that I know. I know how to build communities. I know how to build these kinds of businesses. Now, can I intersect that with things that are happening in the technology sector around AI or vector, whatever it happens to be? And then can I somehow merge those 2 things? Because then you'd be an unstoppable force, right? Because no one in the community building market doing niche market communities is thinking about or having conversations about vector embeddings. I promise you that. So you don't have to give up one for the other. You can say, yep, I'm going to do that. I'm going to do it better than anyone else has ever done it.

But I, I have different advice for you, Sam. I think just get dug in into your position instead. I remember when you were doing The Hustle originally and Snapchat came out and Instagram was like popping off on videos and, and Facebook had videos. And then there was other media companies that were raising tons of money that were just like, we're gonna produce short-form video content or live video content on top of Facebook. And Cheddar was all the rage and all the stuff. And I was like, dude, why aren't you doing videos, man? Look at this. Look at these guys. They're getting millions of views on their videos on Facebook, or these guys are getting millions of views in the Snapchat story feed. Um, you could be the first one there. It fits your audience. And you were like, just very steadfast. You were like, uh, like you're, you're, you're principled. You were like, 3 things. Number one, don't understand it a lot. You know, I don't really understand it. I understand this other thing. 2, I could try to figure it out, but I don't wanna build on top of their platforms cuz they change the rules all the time. I have friends who've got burned by that. I don't wanna get burned by that same thing. I don't wanna build on a shaky foundation. I'd rather do email because I own the thing. Uh, I own the relationship with, with the audience. And, um, it's not like the Facebook algorithm change is one tweak away from, from putting me outta business. And I remember being like, man, this guy's like stone, Mr. Stone Age. Like he is just not, uh, not integrating or adapting. to the new shit. And I was like, I wouldn't— there's no way if I was running The Hustle, I would've been able to resist the shiny object of like video on mobile phones. And like, it turned out video on mobile phones did turn out to be a big thing, but a lot of those company media companies got absolutely wrecked. And you were right for, for being wary of it. Um, and more, more, I don't think in this case people are gonna get wrecked cuz it's not like, you know, the analogy's not one-to-one, but I would say. You know, Warren Buffett missed the internet and all of technology and still did fantastic. Sam, I think you're gonna be in that same boat where like it is not really in your nature to get really interested in, you know, new frontier technologies and play with them and try to integrate them. And that's not really your nature. And you're better, you're best served by like knowing your nature and just doubling down on what's a working formula for you, I guess. So, so I would do that.

I appreciate that.

Because there's gonna be a, a trillion people trying to do fancy AI shit who are better suited to do that. And it's gonna be an absolute bloodbath for, you know, for, for like, go look right now at the number of AI tools that are coming out every single day. Um, and you know, it's like—

Most of them do seem shit though, right? I mean, like—

It doesn't matter. There's just swarms that they're all gonna get just like wiped out. Every GPT release wipes out a whole wave of like, even the successful ones, because it's like, oh, now that's just a feature of ChatGPT. And so I don't know. I think it's like know your nature, and like you know it's okay to not have to do every new thing unless that's your nature. Unless like like for Darvish, it is his nature. For me, it is a lot more my nature than it is yours, and there's pros and cons that come with that.

And so I think are you are you going to go in? I mean, Sean Sean's got a new idea that he's tinkering with, and he's been telling me a little bit about it.

Before I jump to that, I have one piece of tactical advice I have to share with you, Sam.

On yeah yeah yeah.

So I was going through the application process on Hampton last night, like 2:00 AM.

Really?

And this is super tactical, but this is what we do here on MFM. Question number 9 on the application process is, what's your role? It's a required question. Good. The subtext is CEOs, founders, and partners only, please. That's the subtext. The options are founder, CEO, owner, and other. The one thing I would tweak if I were you. So what you're doing is you're saying, hey, we're about founders and owners. And if you're not one of them, don't bother. Thank you for not bothering. Go away. Right? So focus is a magical thing. I love that. But you're doing what I call a pre-filter, right? Which is why not say, oh, this is for CEOs, owners, or whatever. Don't make them feel guilty for going through the rest of the process because there may be a future version of Sam in Hampton. that says, oh, you know what? We solved this problem. But that same problem around people needing therapy from peer groups applies a lot to like VPs of product. And that community right now, all the only communities they can find are people that want to talk about product management and no one wants to talk about relationships and there's an opportunity there. And so, it costs you literally nothing. They'll still answer the question. It'll be sitting in your database for a year or forever and it costs you nothing. Don't, don't push them out too early.

Uh, it should have been the way that you suggested. I apparently, I didn't give that feedback. Grant, if you're listening, uh, this is a direct order from Dharmesh.

Change question 9, please. Grant, do this before you get replaced by AI. Yeah.

Yeah.

But Sean, you're telling me, thank you, Dharmesh. Sean, you were telling me about stuff that you're thinking about and it was pretty, it was somewhat old school, like what you're, the thing. So are you like questioning that after this conversation and you're like, Oh man, this is like—

Not after this conversation necessarily, but it's a snowball that's building, right? Like there's a reason I cleared my calendar to just mess around with AI all week because I'm interested in it. And when you, it's like, let's go see what's real there. And I did the same thing with crypto during that, during, uh, you know, when, when crypto was really interesting, intriguing. I was like, okay, let me go try to mint an NFT. Let me go try to actually like use DeFi and see what's going on here and what parts make sense and what parts don't make sense. Oh, that was pretty frictionless. That's cool that I could just get a loan in one button and I could pay it back in one button. I never had to talk to a human being. I really like that. Hey, this thing says the yield is 20%. I don't really understand where they would get 20% from, so not sure, but I'm going to put a small amount of money in just to learn. I was trying to play with it and trying to think for myself is the big idea. And it's not like some binary thing, like, is crypto good? Is crypto bad? It's like, I want to know where it's at right now. I want to see it develop, and my best way to do that is immersion. I actually stole this from the from Bill Gates. Bill Gates does his reading week where he goes to a cabin and he reads a shit ton of books for a week about one topic that it's like been on his mind, but he hasn't had the appropriate amount of time to roll up his sleeves and dig in. And I was like, oh, that but without books. Just give me a you know Chrome browser and I'm good to go. And so so that's what we've been doing, and there genuinely are so many like mind-blowing moments, and also just like. Just understanding the nuances of things. So for like just being able to think like the computer, like, you know, you were talking about these facets, for example. So I was playing around with Midjourney, like, Sam, do you, do you know what Midjourney is or do you know how to use it?

Yeah, yeah, yeah, yeah, yeah, yeah, yeah. I mean, I, I've just been goofing off and I'll just be like, show me what, uh, Cartman from South Park looks like as a real person.

Right. And like, you know, I was like, can I, and I was, what, the way I approached it was, can I replace work that I already wanna do? With a more efficient AI workflow? That was like one of the things. And then it was, what's really fun random shit I could do? I wanted to be on those ends of the spectrum, like highly utilitarian for me. So it's like, oh, I need a logo for my thing, but I don't want just like a logo. I wanna create a whole brand. All right, how can I use AI to create a whole brand here? So from the icons to t-shirt designs to a website, can I do that with just AI and not have to touch a design? Not have to hire a single designer? And can I do that with just like my own imagination and this prompt thing? And then, oh, how do you do prompting? And like, which of these tools is the best and what's the difference? So that, that was like one whole area. Another was like, we took the podcast and we did this thing that was kind of sick. We took the podcast and we ran it through this thing. So we took the pod and we then used OpenAI has this thing called Whisper, which transcribes any video. So it's like, put in a YouTube link, to this tool, it'll take Whisper and it'll give you the transcript. All right, cool. It takes the transcript. Then I put it into ChatGPT and we had this guy write this little prompt for us. Like we had to get the right prompt, but he wrote this prompt that was awesome, which was basically like, it, it, it's, it's pretty funny. He goes, because ChatGPT can only take so many characters. So he goes, I'm gonna, I'm gonna give you 19 text sections. I don't want you to do anything until you're at section 19. So ignore everything until I'm done with 19 and then answer the prompt that I give you. And ChatGPT's like, okay, I will, I will wait for the 19 parts. You copy paste part 1, 2, 3, 4, all the way to 19. And then you go, the prompt is, I want you to pull out every idea, story, and framework that's discussed in this podcast. I want you to summarize it and I want you to tell me, does this idea exist already or not exist? So it can guess based on the way we were talking about it. Are we talking about something we saw that exists or just an idea that somebody should go do? So like from this pod, it would be like using vector, using this vector dimensions, or I forgot what you called it, vector engine or whatever, to potentially create a dating site that would match people in ways that they're sort of similar using AI. And it would be like, does this idea exist? No. Who is the source of this? Dharmesh. What was the synopsis of the idea? Blank. What is the category that it's in? AI. And so it, then it took that and it takes the whole episode and it just created a database of every story framework and idea from the thing with these tags. And now a human can go back and like tweak them if something was wrong. But like, that's a lot of the work that was done. Uh, and I, we could just do this for the whole back catalog of our podcast. And so I'm trying to use it first for my own benefit. And then along the way, if I see a business, a startup idea that I'm like, oh, somebody should productize this, or somebody should do whatever. Like, you know, the simple example is this Kanye thing. I was like, why is this not the most viral app in the world right now? That basically it's an app with one button that says, you know, say something, and then it's gonna, when you, when you let go of that button, it's gonna turn it into Kanye saying that thing and go share that to TikTok. And like, You might get sued, but you will go viral, right? That's the trade there. But I'm like, that's crazy. There's no front end for this really cool, you know, AI demo that exists now. Um, so yeah, I'm just right now I'm in the go play around with it, see if anything really, really strikes me. And if something does, then, then, then take the next steps.

There's one thread in there, Sam, uh, Sean, that I, I think we should pull on, which is you used, um, you talked about this kind of crafting of the prompt in order to kind of make the thing do what you needed it to do. Um, and that's an entirely new skill now called prompt engineering, right? Um, and it's, it's analogous to software engineering. So software engineering is getting a computer to do what you want by speaking to it in its language. Um, and that way you can kind of get the results you're looking for. Prompt engineering is almost exactly the same thing, except you're talking to a large language model, something like a GPT-4 to kind of get it to produce. So you're talking to AI. to get it to produce the thing that you want. And so I think this is another kind of opportunity for folks that are kind of technology-minded, but not like software engineers, right? So they, they kind of can think about the problems in their head. They're good at it. They may be good writers. They may be good analysts. They may be good at kind of describing a thing, but like prompt engineering is going to be like another big, like a big thing. And by the way, as long as we're dropping things, so I bought 2 domains recently. Buy one, get one free. One of them.

What's that? Yeah, right.

Buy one, get one free.

Yeah, I wish. But this one's not, it's not 8 figures, it's 7 figures. And the domain is prompt.com. And this one, I actually have an idea around what to kind of, what to do with that, which is there's going to be this entire I'm not going to get into details of it yet because it's too good of an idea to actually just kind of put out there in the world. And I'm not ready yet to do something about it. But once I get—

But wait, Prompt.com goes to like a coaching for essays.

I know, but the transfer is still happening. I don't have the domain in my possession yet, but the deal is done.

Dude, so your portfolio of domains, I mean, mid-8 figures then.

Yeah.

Tens of millions.

Uh, yeah.

Fucking insane, dude. I feel amped. I, uh, like when we were talking to, when we were talking to Pomp, I like wanted to go like hide under the covers because he freaked me out about, uh, the billion or the million dollar Bitcoin thing and the banks. With this thing, I'm like, I gotta clear my schedule. I gotta go learn all about this. I mean, I feel amped. This is awesome.

Before we go, give us your 2-minute reaction to Balaji's warning slash bet. That the US dollar will crash and Bitcoin will surge to $1 million.

I'll say this, and I don't know him personally, but he's like quite literally one of the top 5 people I've ever encountered, like even on the internet in terms of just raw kind of what I call wattage, just raw horsepower. And he's like an AI unto himself, right? Like just the knowledge that he has. Having said that, I think I understand why he's taking the extreme positions because that's sometimes what you have to do to kind of shake the world out of its reverie. And it's like, okay, pay attention here. This is important. But if I were a betting person, I would not bet that the odds are what he thinks they are. Could happen, but nowhere near the probability that he's suggesting.

That's, I feel better now.

My personal take.

I feel better. You're, I, I like your opinion better, therefore I think it's true.

There you go. Well, speaking of, okay, we should wrap on this because one of the things that happens anytime new technology comes along, we saw this a little bit in, in the kind of crypto Web3 world as well, is that entrepreneurially minded folks will see this kind of new thing and they will look for kind of the quick turnaround. I'm, I'm all for Creating value quickly, but it has to be like creating value. Don't play the arbitrage. Oh, I'm going to do this thing. This is like, you know, day trading back in the day or whatever. It's like, don't be a grifter, right? Like be, be something that actually brings—

We're just going to build, we're going to build a shitty app and put Web3 at the end of it.

Yeah. So, you know, like just don't take advantage of people. There's enough real problems to solve where real money can be made. And yes, this technology can now be used in creative ways by lots of people and you should use those. But don't use it as an excuse just to kind of be like an AI tourist that comes through, makes a little bit of money or whatever, and then that was that. There's a bigger opportunity. I think you're shortchanging yourselves if that's what you end up doing.

Well, thank you, Dharmesh.

Thank you for coming.

This is awesome, man. Yeah, well, thank you for coming on the pod. This is awesome. I feel pumped, man. I always like talking to you. I, I don't know if you know this, Sean. I Slack Dharmesh all the time. I'll just be like, I'm just trying to get him to like give me little like crumbs of information because I need to get into the HubSpot Slack. It's awesome. I'll just like just send something his way. Just hopefully I can get something back. Uh, but it's fascinating and I feel lucky to be able to have you as a friend and a coworker. And this is awesome. And a podcast guest. You're— this is so fascinating. And I agree with what Sean said about like kind of like looking up to you and like looking at how people live their lives. You're definitely someone I admire. So I'm happy you came here.

Thanks. Thanks for having me on again. This is fun as always.

Awesome.

All right.

Thanks for coming on. That's it.

That's the pod.