Looking Back on 2023

0:04Hello and welcome to Generative Now. This is the show where we talk to the builders who are creating the world's most exciting AI products and companies. I am Michael Magnano. I'm a partner at Lightspeed. And as 2023 wraps up, we wanted to take a moment to look back over our first 10 episodes of the show. I've talked with leaders in AI ranging from the head of AI at DuLingo to the founders of companies like Volley, Rewind, and Hidden Door. I've learned a lot, even much more than I expected to over the course of these conversations. And so I wanted to share for you a few of my favorite moments from the season so far.

0:41Here's a clip from Connor Zwick, co-founder and CEO of Speak, talking about the time that he and his co-founder used to sneak into Berkeley to learn AI. And so we just decided we're going to spend 18 months of our life doing nothing but just trying to become as knowledgeable as possible about everything that's happening. And that's basically what we did. I think shortly after that, we ended up discovering there were some pretty good courses that were just coming online for all the latest stuff at Berkeley. It was just remarkable because literally none of this, this discipline didn't really exist a few years prior.

1:14So like having a university course on like deep learning in 2015 or whatever it was, not a, not a thing, nowhere. We were like, what if we just show up? So it like had just started, I think it was like September of that year. And I remember going to Berkeley and we like take the BART there, me and Andrew, we have our backpacks on, we have like our notebooks and we're literally walking. I think that's like, was like the first week of school because it was like the first lecture. And we're getting stopped by like an undergrad. And he was like, hey, my name's, you know, whatever. And he introduces himself and he's trying to make friends because he thinks we're also just like these undergrads going to school.

1:51And we felt like such imposters because we literally were. And we end up like getting into the building and finding the classroom. And we were expecting a giant lecture hall. And it's literally like 12 people. and it's like a graduate level course about this like pretty specific topic. And so there's no one in the course. It's like no one had ever even really heard of reinforcement learning. They didn't even know what it was. And so we just like walk in and sit at the back of the classroom. Everyone's staring at us. It's very clear. Everyone kind of knows each other from other courses and they're like, who are these people?

2:26But everyone's a little too awkward to actually like confront us. And including the, I think the lecturer definitely knew for sure. and uh yeah it was it was uh so we just literally would go every single time um the so so the the lecturer ended up being john shulman who ended up being one of the co-founders of open ai um he like he was the one teaching it even though peter abiel was like the actual professor but like i'm sure abiel like had less less knowledge of reinforcement learning because it was such a new topic than john shulman did who was like i think grad a grad student at that point Maybe, maybe.

3:00But anyway, like, yeah, it was like literally I'm sure the people that were in that room, all if I knew who they were, like they probably have all gone on to like join research labs and such. Did he ever confront you? Did he ever? Did he ever? Never. Hey, guys, I know you're not in this course. What are you doing here? Yeah, never. Did you ever have to submit any work? We so everything was on a public everything was on a public website. And so it was amazing. He would just give us the website and everything was there. he also was, I think there was a period where it was literally three or four weeks in a row where he was literally making the courses we went and he like was too busy.

3:36And so he didn't create homework. And so there literally was no homework for like three or four weeks. And then he finally reveals this thing and he's like, okay, this is going to be your new environment where you're going to like build your models and then it'll test on this. And I'm pretty sure this was like the first version of the OpenAI playground or gym, which is like their big reinforcement learning thing that he ended up doing at OpenAI. And he literally like spent four weeks on it to do our, like to basically give us like a homework environment to build models in, like, you know, casually producing this for a student.

4:06So it was pretty cool. Let's hear from Max Child, co-founder and CEO of Volley, talking about what it's like for a startup to be first to a new platform. You know, big moment in my life was got my first Amazon Echo Alexa device and started playing around with it um it's for my parents for my birthday and um you know i started playing around with it and asked to play music and you know telling it to play spotify from 15 feet away um and it like worked and um for anyone who had used siri or any other voice system at the time uh the leap forward of alexa like cannot really be overstated how much better it was i mean like it seems boring in retrospect, but being able to talk to a speaker from 10 feet away and say something in a pretty arbitrary way and have it actually do it was like, maybe not an iPhone moment, but it was like a 60 % of an iPhone moment.

5:02It was like, Oh my God, like, this is like, this is nuts. Like voice recognition actually works finally. Like it didn't work a year ago and now it works. Um, and so James and I got really excited about that. And we realized there was a developer platform on Alexa. And so we were like, why don't we take all these chat ideas we have for Facebook Messenger and start just making them voice versions and see if that's fun, right? And so we did it as a weekend hack project. I think the first game we launched was a spelling bee game, like the Scripps National Spelling Bee, which was a terrible idea because they were not very good at recognizing individual letters.

5:40I was going to say, I feel like it needs to have a very high degree of accuracy for that to work. Yeah. Yeah. Like B, C, D, P, like T are all pretty hard for that thing to recognize, especially in 2015 or I guess it was 17 at the time. And so that was a bad idea. But the second idea we launched was a storytelling game, kind of like a little bit of a heritage from our pet app. But it was a choose your own adventure, interactive fiction, medieval times kind of game called Yes Sire, where you would like be the lord of a medieval castle and you would make decisions. And it was all based on yes, no answers, because we knew those would work like 100 percent of the time.

6:22And we wrote, you know, we wrote all the content. You know, my brother actually helped out on some of the content because he was an aspiring screenwriter. And that kind of was like, in so far as we had ever had a hit, that was like a hit. Like we got 10 ,000 users in like the first month or something on that with like no marketing or anything. And we were like, oh, wow. Okay. This is like the most successful thing we've ever done. Like we have to do this. We're like, like, like immediately, like we are Alexa developers. We're like, all right, you know, it happens. Right. And there were, and there probably weren't that many back then.

6:55So you're sort of early mover to a new platform. Right. Exactly. Which, you know, is one of those things that like startups in DC is always say like, oh, you want to be first mover to a new platform. But like, let me just say it sucks. Like, it's not fun. Like, like it's, it's, it's being an early mover to a new platform that no one believes in is a relentless process of being told that what you're working on is dumb. Like, I mean, it is every investor thinks you're dumb. Every other founder thinks you're dumb. Like, you know, even the platform is sort of suspicious of like why people are even doing this.

7:26Alexa itself was like, why are people building these apps? It's just a process of every day having conversations with people where they're like, yeah, why are you doing this? Why are you building Alexa games? Who gives a crap about this? And we were just like, I don't know. It's fun. here's a crazy story from gaurav misra the co-founder and ceo of captions talking about how they had previously shut down one of their products to later learn that it had just been generating revenue for them in a bank account and they magically discovered five hundred thousand dollars so i think that was the big question for us right is how do we make a talking video platform is what we were thinking at that time right and so we thought let's start with the tools, right?

8:14So when we thought about the tools, we were like, well, let's start with something we can do for talking videos, right? Where people can have it easier somehow, right? And so that's where the idea for captions came was, oh, let's have maybe automatic transcription, right? Because that's not something that existed at that time. I think a lot of people forget that TikTok, Instagram, like nobody had that at that time. It didn't, you know, exist in the form that that that captions made it where it's like appearing on top of the video and stuff like that and so a lot of people at that point there was a movement around accessibility and a movement around like you know people would sound off and i want to understand what they're saying and so a lot of people would manually do that you know they would manually put all those words and there was been an hour like a whole hour doing that manually transcribing their own audio exactly Yeah.

9:11So, you know, we kind of noticed that. We saw that. We're like, hey, perfect. This is the place to begin because it caters to talking videos. It's the perfect thing to start. And we made it, you know, made the app. I think it took like a couple of days to put together the first version, launched it, and kind of went to sleep and woke up the next morning and it was at the top of the app store. So that was a surprise. Oh, my God. Yep. It took you three days. Yeah, very fast. And was this the first thing you tried? So you were like, okay, here's this problem of transcription, three days, build it, boom, top of the app store.

9:46Yep, it was the first thing we tried. And actually we were doing sort of VC pitches at that time. So the timing was absolutely perfect, right? Because in the middle of the pitches where people were starting to question, like, well, is this really going to work? We had something to show off. Like, wait, it is working. You can go check it out. Right. So that helped us a lot. Wow. What a story. I mean, it's so rare that that actually happens. And are you like, while you're pitching, are you pitching the social media side too? And you said you want to start with the tool. Yep, we are pitching exactly that.

10:19But I think people would obviously question, like as I would, if anyone told me this idea, right? It's like, okay, sure. Maybe if you pull off the tool, which is pretty hard to begin with, you know, how do you even make that into a social network? That's the whole other challenge, right? Right. It's hard to actually believe that, right? and so but people were getting even stuck up on like well actually making a tool is pretty hard yeah um and so the timing was absolutely perfect because we could then show like hey wait actually we did make a tool and it is working that's awesome so where so where do you go from there i mean the thing is exploding you're raising money yeah i mean then we set on our our path to make a social network which oh you basically we did yeah we did and that lasted about about a year or so.

11:04We spent a year, you know, really trying every possible idea, trying every possible angle. And I mean, the bottom line is, it's really hard. I think, you know, everybody knows this. And I think we knew it too, but we really wanted to do it, right? So hard to blame us for it. But we pivoted to doing more photo stuff at some point. And that started to really work. Um, so I think there was also a little bit of a movement at that time around, like Instagram was becoming more video. And so people were like, well, where's my photos, right? Like, where do my photos go? And so there was a need building that we could see on like platforms like TikTok and stuff with people talking about, you know, well, where do photos go?

11:54Right. And so we started doing photos for a while that worked fairly well. We had like a U.S. only app that we built for photo sharing. And that took off. And, you know, we hit, I think we hit like 50, 60 ,000 DAO in the U.S., which solid, you know. Yeah, totally. And so we ended up going and raising a Series A on that, actually, which was like early last year. Yeah. And at this point, is the talking video tool still happening? It's like running off to the side while you're building the photo app. It is. It's there. It's kind of sitting there. And we actually, you know, we're a little bit worried because it's costing money.

12:39Right. Yeah. Yeah. Like it's not a lot, like$28 ,000 a month or something, but that's a quarter million dollars a year. Right. Yeah. It's not nothing on a seed company's balance sheet. Right. And so we were debating whether we should just kill it. which was we discussed for a bit and we were like, actually we decided, you know what? Yeah, let's shut it down. So when I went to actually go shut it down, I had this like random idea of like, why not just put a paywall? And that way if, and the paywall will block the entire thing. And so if no one pays, then it's essentially shut down, right? Right.

13:25For all purposes. and if people pay, then it'll pay for itself and we don't have to shut it down so we can stop worrying about it. And so I put the paywall and I shipped it just like on the weekend without telling anybody actually. I think our iOS engineer, I told her, I was like, oh, if you see PR coming in from me, don't worry about it. And yeah, and that was it. Then we forgot about it basically. And then we were working on our photo sharing app which was doing really well. And that's all we were focusing on. And we raised a series A for that app. We hit, you know, 50, 60 ,000 Dow and literally like millions of photos being posted.

14:09Things were going pretty well. After the series A raise, we, I went into my account, my like a personal Apple account, which was separate from the company one. And the personal account had the Captains app on there. And I go in there and it's made$500 ,000 basically. And yeah. And you're looking at, yeah, the ARR growth that is just like going straight up basically with zero employees, just zero people working on it, you know, zero customer support. There were like 1800 open support tickets that no one had answered for six months. Right. And it was just kind of just going, right? And so that's kind of where we had to really think about, you know, what do we really want to do here, you know?

14:57And we got really excited about the original vision with, you know, captions and all the things we had thought about with ML and AI and all that stuff. And, you know, yeah, it kind of made clicked for a second a little bit more of like, you know, this is all the things that I've worked on in my career in one thing. Here's Hillary Mason, co-founder and CEO of Hidden Door, talking about definitions of artificial intelligence versus machine learning versus generative AI. One question I actually wanted to ask you going into this is what is the difference between machine learning and AI? Are they the same thing?

15:36Are they different? So I've been poking at this very question since I'd say at least 2014. I have a slide I like to use when I try and explain this to CEOs. And my answer hasn't really changed, which is that I would say, and let's go back to your earlier question. Like data science is the ability to build statistical models on data that allow us to make useful predictions. machine learning is the ability to design systems that improve with the introduction of more data and to have feedback loops so we're able to make those predictive models and then we're able to sort of you know deploy them out and see where they drift and fix them up again it's really the ability to there's a fine line there and a nuance but i'm going to say like more or less that's where i'll draw the line for the purposes of today's discussion a lot of what we call data science could also be called machine learning and vice versa.

16:33These are, it helps me to say they're more at this point marketing terms. And they also, to me, they get your mindset around what frame you're approaching the problem with. So with data science, it's saying like, okay, we have a question, whether it's a business question or a systems question, and we want to take the data we have and sort of build a model that's going to help us answer that question. And then machine learning, you're sort of saying like, okay, I want to make a class of model perhaps that can mimic some of what I see in the underlying data. I don't know. It's very fuzzy. Yeah. What about artificial intelligence?

17:13So that one, I would say that term came back when deep learning started to be a useful set of techniques. and I believe again it is primarily a marketing term used to differentiate from what has come before and and if we go back into the history of artificial intelligence as a field like you know there's been many rebrandings um machine learning itself was a rebranding of AI coming out of an AI winter um where the whole field essentially stole statistics and like started using it a lot and then rebranded a bit but again same people same same kinds of work um and so you know if you asked me in 2014 or 2016 i would have said like ai is a term of use that is largely driven by people who are now able to approach problems with deep learning and that has given us a step function forward in terms of certain capabilities in certain domains comes with a ton of trade-offs but it's not I would still say that most things that we call AI systems are machine learning systems in the sense that like if somebody comes up to me with a regression model and they're like look at my data science I'm like cool if they come up to me with a regression model and they're like look at my machine learning I'm like maybe not and then they're like look at my AI I'm like oh no not okay like that's not it but also what is like where do we draw those lines and right now like I'm doing it on vibes.

18:44Like there's probably more, more clear ways to do it. So we shouldn't think about it as such, Hey, we have a problem to solve. Should we use AI? Should we use ML? Should we use data science? Cause it almost feels like that's how people use these terms now. And it sounds like that's the wrong way to think about it. I don't find that to be a useful conversation. Um, I find it more useful to be saying like, can we clearly state the problem we're trying to solve? Do we have a quantitative objective function we're trying to optimize for? If not, how would we even know what good looks like? And almost trying to get into like, is it meaningful to solve this problem?

19:20How is it meaningful? How meaningful is it? Is it even worth the investment of time and energy to do it? And if so, what are we going to do with that? Because there are obviously, I mean, I'm the sort of person who can ask a million questions and get really deep down rabbit holes and then realize none of that work has actually changed my decision. So I should probably not have done it. I have people say generative AI is a pure marketing term. Is that fair? I don't think so because generative functions, again, are a thing and they are a thing that says like, it is the kind of function that can take many examples and then produce things that sample from the space of those examples to resemble but not equal.

20:02And that's itself a research point, the underlying examples, I actually quite like that we're using the phrase generative AI because it is wrapping a little bow around that particular set of capabilities. And now I do think it gets overused to mean chatbots, which are terrible in many ways. So it definitely is an overloaded marketing term also, but I think there's a fundamental grain of truth in it, that it is worth mentioning the huge step forward function and capabilities around a lot of this stuff and sort of calling it out as an area of capabilities, whatever techniques you might use to actually produce that.

20:45I think we could definitely do a podcast episode that debates this because I've had people tell me the exact opposite and both of you sound right. Yeah. We should do that at some point. Let's fight. check out this clip from victor reaper belly co-founder and ceo of synthesia talking about how new tools create new content and experiences that we couldn't have previously predicted comparing it to how the drum machine created electronic music the best way i think to think about it is just looking at the history of media technology and technology in general right because i think history repeats itself all the time and even if it feels new now that we're in the middle of a cycle, I think we're going to just see the same thing happening again.

21:28When there was a time when to make music, you had to have real instruments, you had to have a huge recording studio and there's very few people in the world who could afford to do that and knew how to do that. Then something like a drum machine was invented and the drum machine was actually originally invented to replace drummers. That was the intent of the drumming machine. But it turned out the drumming machine is just very far away from being something that you can replay the drum kit with. But what it did open up was you can buy this little machine and you don't need a drum kit anymore and you can use it with headphones.

22:00And all of a sudden we get like house and techno electronic music, right? Along with synthesizers and all these other things. And what slowly has happened is that what used to be an entirely kind of analog workflow with real instruments now is completely compressed in a MacBook, right? But you can, as we spoke about earlier, you can use software instruments, you can use samples, you can play an entire orchestra on your keyboard if you want to, right? You're not really restricted by having to go out in the physical world. But of course, people still play piano. People still play guitars. And they have some soul you can't get out of a machine.

22:33We like it as humans to do those things. And we always think it's different this time, but it just rarely is, right? So I am even... Internally, you know, we're very kind of like, we don't want to be prescriptive around what we think this is going to look like. We want to learn with our customers and sort of, you know, one feature at a time, build our way towards that future. but I can't wait to see where we are in three years time. It's going to be very exciting. It's so true. And it's so interesting, right? Like it almost feels like we're in this phase of AI right now where we're trying to map everything that gets done to existing formats or legacy formats, right?

23:06And to use your example from earlier, it's like when the drum machine came out, you know, the first use cases were probably, how do I replace the kick drum in this rock song? And it's like, no, no, no, no. This isn't, actually, this isn't for rock. This is a whole new genre. And it's called electronic music, right? If you take that example, you can do things with a drum machine you could never do with a real drummer. You could literally have 10 drummers on 10 different drum tracks in a machine that's the size of a VHS machine or something like that back then. You could play at speeds and different patterns of overlapping, which you really couldn't do before.

23:43And that is the exciting thing about new technology. Absolutely. And I think it's so, I think a good mental model for people who are building AI today really is like, what is it you can do with this technology that you couldn't really do before? Here's a clip from Scott Belsky, Chief Strategy Officer at Adobe. Well, let's talk about AI, generative AI. I mean, that's what we're all here for. And like I said, I feel like you've had a front row seat. For me, sort of the aha moment, I think everyone had different kind of aha moments for AI. But for me, it was maybe nine or 10 months ago when Dolly 2 launched.

24:20That was where, you know, sort of the light bulb went off for me. But what about for you? Like, what was the aha moment for you and maybe for Adobe as well? Well, I think that, I mean, one of the things that brought me, I was at Adobe, as you know, and I left for a brief period and then I came back, you know, as chief product officer five and a half years ago. One of the things that brought me back was this idea of, like, creativity for all. You know, I was like, oh, my gosh, you know, what are humans going to be doing in the future when a lot of stuff is just achieved through compute? and creativity really needs to become the next productivity.

24:49That's how people are going to stand out at work and in school and whatever. And when I came back to Adobe, I realized that the creative world is just full of friction. You know, it's really hard to get in. Like most people don't have the skills. Creative confidence probably peaks in kindergarten when everything we make, our teachers will put a star on it or our parents will put on refrigerators. And then we quickly realized that there are these things called critics and we don't have most of the skills that we would want. and then suddenly we kind of like go away from creativity, most of us. I always thought that was a shame.

25:19It's like, why does creativity, you know, why does creative confidence go down as opposed to up over time? And then, of course, there's also like the future of digital experiences and like what's next and 3D and immersive and whatever. So that's what brought me there. And then over that time, I just realized that the friction is there because it is hard. Like these tools are hard. And if you think about creativity as a box with the floor, you know, kind of like pretty high, it's like hard to get in that box because of the skills and the capabilities and the expense of the tools, whatever. And the ceiling of the box is what you're capable of, given you only have so much time in a day.

25:51The objective was to make the box bigger, which, by the way, increases the TAM of Adobe and is good for the company as well. And there's been a million ways we've tried to do that. We've tried to improve the first-mile experiences, tried to create new products and web products and all these things. We'd all make little dents here and there, but none of them were truly transformative. And then you see Dolly and you see some of these new capabilities that we were also brewing in our lab. And there was a light bulb moment for us where we were like, wait a second. Simultaneously, this lowers the floor of the box.

26:22Suddenly people can just get in the box by just prompting something. You can start with something and just start to speak a natural language and make it into something in your mind's eye. That's an amazing way to lower the floor of the box. and you raise the ceiling of the box because suddenly people who are illustrators who never knew how to animate could just like use generative AI to make an illustration animated, you know? And you start to go down the rabbit hole of 3D and immersive. It used to be privy to people who knew physics and math. And now suddenly everyone can do it with, you know, prompts and with other capabilities that are, you know, made possible by algorithms.