Episode 6: Beyond the Video Call: The Future of AI Voice

hello friends welcome to friends and code spaces do you do you get that reference friends friends and code spaces could we could we do a little at the start of this i don't have a theme song yet i kind of ask every guest to just give me a little which one of you has a better country accent i mean are you gonna claim it i am from north carolina Carolina oh my gosh that was beautiful can I just get a little just one bar I got friends in code spaces and a voice look at that well that makes sense he is in voice AI that makes a lot of sense awesome so welcome to the show you have no idea what we're doing here and I really appreciate that you both came and agreed to come uh and do this with us definitely wouldn't have if you said I had to sing excellent So this show is about people that we meet in real life that are doing awesome things. And we want to bring them onto the show and let everybody hear. Because not everybody gets to be in real life with both of you. And it is a wonderful experience every time that we get to be together. So I am here with Quinn and Nina. I'm Nina Kravilla, co-founder of Daily. I'm Quinn Lohman-Kramer. Almost everybody calls me Quinn. He has a difficult first name. And I didn't know your last name. It's really nice. It's on my screen, actually, right now. I could have looked at my screen, but I didn't. We're on a first-name basis. We're there. And we do a lot of stuff together. And because you are so good at doing community stuff, I love the community that you are constantly building. I love the way that you are sharing your knowledge. And I really appreciate that. And I would love to learn more about that as we go through things today and more about what you're building, because you are building some of the... coolest tech that is out there and you are making it accessible and i really appreciate that really really appreciate that we are going to start with our first game our first game is called context window so llms only know about some things and it turns out if i asked llama 32 about you i said hey llama 32 uh i am doing an interview with quindla haltman kramer there's the full name and nina kura villa i did that and uh They are the co-founders of daily.co. What do you know about them? And more importantly, you are a self-aware AI. And I want you to ask them questions that would help you make be a better AI. So I just want to say that you are our first guests that were in the window. It knew about you a little bit. Didn't know everything that you were doing, but it knew about you. So we're old. You've been around. It knew about daily too. It did. It did. I think, I think a cloud communications company that offers virtual meeting and customer experience solutions, maybe a little dated, maybe a little bit dated on what the story of daily is, but which we're going to get into. Um, because I went forward and I gave it some more there. It's interested. It's, it's definitely interested in how you, uh, overcome early challenges, which is fascinating. Um, uh, emerging trends and predictions. And then I gave it a little additional context, right? So I talked about, um, You're leaders in the AI voice space. I really think anytime there's something that's voiced, it's like, well, I'm just going to, what does Quinn think about that? What has Nina shared lately about that? The open source voice framework that you are pushing right now is so cool. You're enabling people. So I wanted to know about that. And you throw hackathons and you inspire people with demos. So like, I really, really enjoy that about you. And I said, hey, AI, what would you, based on this information, what's going to help know more about them. And so there were some categories. And if I could get you to look at those categories, and you get to choose, this is the choose your own adventure part of this game. Underneath each one of these, there is a question that AI generated. So AI generated the topics about how it can be a better AI. So the questions might be a little bit strange. So do you see anything there that jumps out that you want to see? Well, first, which Llama 3-2 is this? Because you've got to meet the LLMs where they are. Exactly. So we're at 11B. 11B on this one. What do you think, Nina? Let's roll the dice, pick one. So can we read which ones? Let's read out what's there because we have some listeners at home. So it's what we have presented before us for chosen question categories are AI voice space, co-founding challenges, always a key topic in San Francisco and many places. Human skills, such as creative problem solving. Which are the AI interested in. Innovation approach. Origins of daily.co. The pipe cat voice framework. And last category, the role of AI and machine learning in their products. Your self-absorbed question on the part of the LLM. Talk to me. Where are we going to go? What's door number one? we choose them yeah i mean should we start with ai voice space or is that going too deep too soon where should we start our origins of daily we talk about a little bit a little bit about how we got here yeah yeah it feels good let's do that context all right i'm putting my glasses on we reached that that part of the show now uh we're in beta um uh i like this one a lot i like this first question a lot that's asked here but i'm gonna not ask it i'm gonna let another ai ask this to you uh and it's not gonna as impressive as it's not going to be impressive here you know 11 labs we're going to use 11 labs to generate this and let's ask the question how did the duo balance their individual expertise and strengths to create a cohesive vision for daily.co and what skills or experiences did they leverage to build the company from the ground up did we do that well you know startups were here that's true you're here yeah well we thought that there was going to be audio and video and everything we all do online and a lot of it would be conversational real-time people talking to each other so we started daily to build developer tools platform that would sort of help make that future come true okay we wanted to build incredible global infrastructure and sdks that would let anybody embed real-time audio and video into any app or or or web page we've always admired Cloudflare because we think of ourselves as an infrastructure company. We focus on different things than Cloudflare does, but Cloudflare's network is incredible, and we wanted to build something similarly incredible. And then as we grew and evolved, the generative AI moment happened some years later. So we were already pretty mature and built out in terms of our developer tools. And all of a sudden, talking to humans was not the only thing we could do. Yeah, we started by building this real-time infrastructure for human-to-human calling. We all know, we all saw it and experienced it. When ChatGPT came out, I'm sure all your customers and developer community were coming to you and wanting to brainstorm together. What does this mean? What can we do? How can we better support our customers? So because we were enabling voice and video integration, a lot of our customers and developers were interested in things like Real-time translation, real-time transcription, how to make those real-time experiences better. Pretty quickly, we began developing orchestration and infrastructure around voice AI agents. So extending from human-to-human calls to human-to-AI calls. We published a very long technical deep dive 14 months ago now, which is forever. It's called how to talk to an LLM with your voice. Oh, okay. Because that was what our customers were asking us for in this new generative AI world. Like we've got these human to human conversations. How can we put an LLM on one end of these conversations for things like customer support and healthcare patient intake workflows and just fun social gaming stuff. So that is like, those are customers that you had already that were using those tools. They're like, wait a second. This is, well, how do I make this? Oh, that's cool. So they were coming to you and you were. like, uh, yeah, yeah, of course we can do that. Is that, is that kind of how it, how it played out there? Yeah, very much so. And so we built a bunch of stuff internally to support those customers, but then we got super interested in it just as, you know, technologists as an engineering oriented company, because I have to admit, I sort of missed the GPT-3.5 moment. Um, cause I, you know, I've been doing this stuff a long time and there've been lots of really interesting incremental improvements in AI, uh, but they were all sort of little by little. Yeah. And I think. Looking back on it, I think GPT-3.5 was probably like a big inflection point, but I didn't see it until GPT-4. But then when GPT-4 came out, there was all of a sudden this piece of building block technology that could have human-like conversational dynamics. And obviously for what we do, delivering conversational technology stack, the fact that you now had this LLM that could be a real conversational partner sort of opened up this whole new Vista. How did you deal with that? So let's get to the heart of this question here. It is asking about co-founders. I'm going to say a little. Can I share a little secret here that I know about you both? You're married co-founders, which I am so inspired by that. I seriously am. I think that there's a lot of people out there that have partners who would both like to have an idea and do it, but don't because they're like, well, you can't do that if you're married. But you're doing it, and you're married. And you're saying ChatGPT35, so it's been at least two years. How long have you been? How long has Daily Co been around? Let's see. You went through, you took Daily through YC winter 16. Winter 2016. It's the beginning of 2016. I mean, honestly, you're already married to your co-founder. Like, truly. You know, maybe there's some co-founders nodding in your audience. Yeah, for sure. It's a pretty intense relationship. We'd worked together before. So we'd known what it was like. And, you know, I think there are a couple things that make it work for us. First, we're just excited to work on the technology together. We've done it before. And honestly, in terms of making it work day to day, we have a shared framework for values, which is really important as co-founders. Oh, that's cool. And is it the same values that you have in your house? well hopefully yeah yeah yeah because i mean i think you often hear in in startup culture that your team is the most important thing and i think a lot of the values we have in in the house maybe map on to you know the values we have with the team what would you say we wanted to spend our time building something we thought was valuable yeah on multiple levels yeah as a technology as a product as a company wrapped around a technology and product and we wanted to do it with people we respected and admired and wanted to spend time with super cool and it's hard you know it's hard to find the the the tactical uh expression of your values always that's true yeah everything in life yeah but it's good to kind of have a sense that that's what you're trying to do we're not sort of startup people for startup sake which there's nothing wrong with that like we have lots of friends who sort of they they decided they wanted to start a startup And then they ideated to find the idea that would fit for them. We're just wired that we sort of took it the other way, which is that we had this idea we were interested in. And then we tried to figure out whether it was something we should or could build a startup around. And I've founded startups before and actually took two years between startups to make sure I really wanted to do it again. And one of the things that made me want to do it again was the chance to build something from scratch with Nina, who's so good at so many things. It's like a real privilege to get to work with her. I'm going to say sweet. That's really, really, really nice. That's awesome. How's the work-life balance with that? Does it ever stop? I mean, work is life. Life is work in this world, right? It is. And I think one reason that's tricky for co-founders and teams in the startup space is we all are passionate about what we do. this is why we do it, right? We want to build something really compelling and really great and work with really neat people. So it's this funny dynamic where that very drive can get you in this tangle over work-life balance. So I think as with everything, I keep coming back to this idea of you have to be intentional. So we try to be intentional about work-life balance and just try to be aware of it. Okay. I mean, we do have some funny things. I am one of those people where I'll have something on my brain and I'll want to say it and feel free to laugh at me. I just realized something clicked for me. I was like, oh, if we were co-founders who weren't married or we weren't living in the same house, I'd just slack Quinn over the weekend. So now if it's the weekend, sometimes I just say it because you've got to say it, but sometimes I'll be sitting next to him and I'll just slack him something. I don't want to interrupt this time that we're having here, but this is for you and the next time you're back at Slack. Yes, Slack enabling happy partnerships in many ways. We should talk to them about a commercial if you're ready. And we don't aim for work-life balance for us. I think that is the thing you have to sort of recognize what's going to happen and what's not and not fight against it. So we are lucky to get to do what we do. We like it. Everything's trade-offs in life, but we just sort of don't aim for work-life balance. On the other hand, a startup as a collection of people is a marathon, and we try really hard to kind of build guardrails for our team because not everybody can be pressured to work all the time, 24-7 for years on end. And one of the things we say to people who ask us about this, like how do you hire people who are going to kind of meet your bar in terms of expectations, is we just look for people who are really intrinsically motivated. They really like what... they would be doing it daily and they're excited enough about it that it's going to create like lots of kind of engagement productivity excitement collegiality but we also try very hard to say you know you've got flexibility you can decide like if if it makes sense to work this weekend great but then you if you need to recharge after that great like we totally get it we trust you awesome awesome super cool let's choose another one let's choose another one up here i'm going to give you one more I'm going to give you one more. Yeah, one more. So this is your last of this game. We've got another game coming up soon, but you get one more. Well, you were talking about community. And I mean, it differed to me, but I think it would be fun to talk about sort of open source and AI and how much interesting stuff is going on. And like why we're doing some open source stuff and putting a ton of effort into it. Okay, let's do that. So this is AI speaking. I'm going to read it for it. Can you walk me through the process of creating a voice-powered solution using PipeCat? What are some of the... typical use cases of applications that you've seen developers or businesses leveraging the framework for yeah all sorts of use cases maybe the simplest way to put it is you know when you call your doctor's office or you call the bank or the airline you're put on hold yeah you should not be put on hold okay thank you right we see developers and customers building things like Small SMB front desk, SMB telephone front desks, call center workflows in healthcare, patient-facing workflows like patient intake, scheduling appointments. It really runs the gamut. Sales outreach, sales coaching, really anything where AI can help triage or resolve the situation to free up human teammates to focus on more important things. Yeah, that's awesome. Awesome. I feel like the PipeCat framework has given people like, ooh, this is that thing that I wish I could go and build. And you open sourced it, Glenn. We did. Two things happened at once. One is we sure woke up and realized for several months we'd been helping customers solve the same problem over and over, which sometimes happens in a startup, right? And it was things like, well, there's something going on. There's a moment. But, you know, how do you know when it's? LLM's turn to talk. How do you gracefully handle the human interrupting the LLM? How do you do the context window management? That's so important to guide an LLM through a conversation and probably half a dozen other things. And so we realized we'd built a bunch of internal tooling to help our customers that was like really broadly useful. And then the other thing that happened is in January, we did like just kind of randomly, we didn't really think much about it. We did a voice say, I meet up here in SF and we didn't promote it at all, except probably. tweeted about it and 40 people came yeah and then in february we're like okay well let's do that again and 80 people came yeah and we're like huh there's there's only two dots but there's a trend yeah yeah yeah i see the graph i can picture it we open sourced all the stuff we'd built and it was really very rough github repo with no docs originally but people started using it and people started contributing back and people started writing docs um and then we had another meetup in march and 120 people came and then i think at cloud we did a hackathon in april yeah and like 150 people came yeah so there was clearly enough interest in this stuff that it felt like oh there's an ecosystem and you know we we do this stuff because we find it really interesting yeah and really valuable and let's see if we can help a little bit help this ecosystem form yeah absolutely yeah and to continue the trend line so we did another hackathon a couple weekends ago an open source real-time hackathon we worked with great partners um google Google, Oracle, Tavis, Vappy, Cartesia, Covell. And what, over 550? We had 564 registrations. Did they fit in or no? So one of the reasons it got above 500 was because somebody pinged me on Twitter and said, this time, can you make it remote friendly? Because we just never quite had the time in the previous meetups to kind of figure out what a remote participation might look like. Ironically, for your... Totally. Yeah. Cobblers shoulder and never wear shoes. But a two day hackathon, we were like, okay, we can make this work remotely. We can set up a bunch of video streams. We can have people join video calls ad hoc. We can do it on discord. So I think we had about 250 registrations from San Francisco and about 300 remote. And it was great. The remote channels in the discord for the hackathon were so hopping. Yeah. And you were asking about use cases. It was so. fun to see what developers were interested in exploring um so we had different categories we had an open source category and um maxwell uh built this amazing he's built out a drone project okay um over a few months and he uh added sort of voice enablement all right yeah so it was a voice enable or voice driven drone um his his demo project clip was was great it was all american themes he um launched a drone controlled it by his voice to go to trader joe's oh i saw that video i saw that video it was just an example of voice ai will be everywhere it went and it went to trader joe's to check the parking lot to see it because he hates how busy and at same same i mean what a problem i never would have thought to solve it with a voice drone there you go and it was so cool his stack um the nvidia integration yeah so he built the drone himself oh my gosh and the compute platformers and nvidia jetson embedded platform so he's running a bunch of computer vision models on board the drone he shot that in the sky yeah wow yeah and then he's running a bunch of llm stuff in the cloud talking to the drone wow and then he's got the voice channel which runs on our stuff on pipecat so he can say something like yeah go go go see how busy trader joe's is yeah and the drone goes so good you know goes a mile from his house so good parks over the parking lot sees how busy the parking lot is it reports back it reports back over voice right so just you know real-time voice ai in embedded systems hardware and then there was a ton of stuff around enabling the sort of workflows we were talking about so um there was a developer who actually owns a theater in town which is amazing and he mentioned he doesn't really have the resources you know to answer the phone all the time so he quickly built a phone assistant on that yeah and it's so cool right like that the fact that like you can be at a hackathon the two-day a two-day hackathon a concept that you have that didn't exist y'all like you you were doing that you were paving that way i hope you i hope that feels good that's awesome so so thank you for that um we are going to move into our next game have you have you heard of the game hot ones is this the chicken yeah you eat the hot and it's hard to kind of answer questions but this is very similar to that but it's called cold ones uh and what we're going to bring on as a little surprise. Lizzie is going to come and bring us, I don't know if it's been a while since you've had one of these. Have you had one of these recently? This is a Slurpee? I don't know if I'm allowed to say it. This is a frozen drink. Which one of the frozen drinks? You've got four colors to choose from. You should choose first. No, no, no, please. I mean, I'll choose first. What flavor are you going for? Are we talking cherry or strawberry? I think it's red. it's red it's preserved whatever red number 12 yes and then uh we've got a blue a brown i don't know and a man i think this is mango this has got a mango vibe i'll take the mango the mango unless you feel strongly about it okay no that's fine and i'm gonna go blue because i like the flavor of blue let's do that all right the goal of this we are gonna take a sip and we are gonna go back to a time the last time you had this And you have kids, right? So you might have had this. You have one? We were just saying on the way down here, you love to feed our kids sugar. It's a terrible, terrible thing. Your kid who comes to hackathons, by the way, incredible. He loves hackathons. Yeah, Cloudflare truly supports the developer community or the next generation developer community. Our child is now like, I want to be a developer. I want to go to the cloud. our kitchen because of the snacks we did it okay let's do it cheers cheers to the slurpee let's take a take a deep sip and feel it feel that nostalgia oh yeah that yeah that blue is so blue i like i can see it in my brain went blue there for a second i want to go back i want to go back to a time when you were when you were a kid maybe even when you first started thinking about maybe the possibility of coding of getting into code or building. Can we go to that? What was the first thing? Let's just say build. Let's not even say code. What was the first thing you thought about building? I'll let you go first. I mean, Legos are pretty formative for a lot of us who decided we wanted to build stuff for a living. So I read all the time. Build your knowledge. That and kind of the structure of narrative. Oh, cool. It was a different kind of building. If you can't tell a story, you can't really explain to people why the value of what you're doing is there. What was the first tech that you got into when you were like, I want to do this for life? I do think giving kids access to programming tools can be really, really powerful. And sometimes maybe people have different ways into it. Lots of people we know who are engineers started with video. games yeah i think that's totally reasonable yeah um but whatever a kid is interested in computers are so flexible now you can find some computational like enhancement with that uh and that's even more true in the age of ai right i think you can build stuff with an llm's help in ways that weren't possible even a year ago which is amazing yeah yeah absolutely yeah i'll give um maybe i can mention two things okay because The first really may be personal to my own worldview, but I mean, it relates to reading a narrative structure and almost prompt engineering. Do you remember the Pick Your Own Adventure? Like that sort of like, you know, those trees were really interesting to me. That is, you know. Now you really do build those for a living for our customers. You have to like ask an LLM to get five or six things done. a voice call that's interesting engaged with a customer yeah yeah so I mentioned reading before there's just something about organizing concepts right how you kind of tease out knowledge and language but then of course like a lot of people too you know that first time you're in front of a Mac as a kid right and using their publishing software right yeah you're like a little older than being a kid but I was super lucky to of wash up at the mit media lab as a grad student oh wow in a moment when the mit media lab was all about like this crossover of the internet and hardware like embedded hardware and virtual characters like there was a virtual dog okay that we were showing not me but like people at the media lab were sort of showing off like 1998 1999 like clippy like pre-clippy that that little dog that would bring the i think i know what you're talking about this is like a there was a big wall and the dog lived on that wall and you could walk up to the dog you can interact with the dog wow it was what tech was that much more it was a bunch of the precursors to neural networks okay i mean neural networks the precursors to today's neural networks they were running on massively less powerful hardware right uh with less powerful architectural techniques but you could go talk to the dog and interact with it physically yeah so it was it was amazing to be at the media lab like the very late 1990s and like you were you're sort of seeing the future yeah yeah that's cool that's super cool and i do feel like i do feel like where we're at right now like especially i i think that the code space that you're in right now is so much the future of where we're headed like a robot's no good if it can't talk to you right and if it can't understand what you're doing and you know i think that like i love the problems that you're tackling with like i mean i don't want to put words in your mouth but i think the biggest ones that i think that you solve is latency right like that because you don't even think about that right that's our goal in life is to drag latency down which we do all the time and i think you're right but for me there's sort of a two-layer stack one is just ai in general truly changing everything but we're at the very very early days of that and then layering on voice and conversational dynamics that you know meet the human where the human is in terms of how we like to talk right how we like to listen how we like to interact we're even more at the beginning of that I went to a family reunion this past weekend, and people are like, oh, this AI stuff, is it mostly hype? And I'm like, no, it's not mostly hype. But as soon as you get outside this little world we live in, it hasn't gotten sort of evenly distributed at all yet. Yeah, yeah, yeah. I was at a party, and they said, so what are you working on? I said, AI. And they said, ew. And I was like, I don't even know what that means. What do you mean, ew? So the bubble is real here. that people come into a meetup is is real that the you know we we there's a hackathon and people were locked out because at github you could only have so many people at github and they stopped the doors and people were out in the streets literal streets doing demos holding their laptops up and doing demos because like this is awesome fun stuff and like i love i love that that you all are pushing that forward so again if you've been doing this since the early since the early days of the internet it really does feel like that moment again It feels like the Netscape dial-up modem moment is what we're trying to figure out how to put the pieces together. Yeah, yeah, yeah. And as that revolution, the modem dial-up. I'm sure we all remember when our families or ourselves got that modem dial-up and that sound. Yeah, waiting for the email. And how amazing that was, how it gave our families access to things. I think you guys put your finger on it. I'm so excited by how. how voice experiences can unlock AI insights for senior citizens, for kids, for all sorts of things. Yeah, absolutely. Definitely getting beyond SF, getting beyond a certain demographic. Yeah, absolutely. And I think we just walked into our third game. Producer Peter does internet sleuthing. And this is my first question. a little awkward for both of you but we're going to talk about your ex your twitter your ex that's the gotcha gotcha all right um so uh so peter i did a little digging on some of this is old some of this is not as old um but this is uh from a recent hackathon can you explain what's happening in this picture here this is one of the submissions for for the hackathon remote prize track okay folks i think they were in toronto three people got together and built a robot okay that as you can probably tell from the image was incredibly foul mouth you know and it looks like it's a chicken is it it's like it's a chef it's a chef it's a chef okay so it's a chef that has has it's running google gemini llm okay and it has vision input okay and voice and voice input and output okay and it will critique your cooking skills in a foul way in a very very foul Blunt. Were you glad that, was your child there at that hackathon? I don't think he's seen that video. It was the remote track. We're not going to play that one. But here's who did win, right? So who did win the track? I think, I believe. First place remote. First place remote. What was this? This seems cool. This was the personalized shopper and stylist. And again, multimodal. So you see the developer there. And what he showed is like he just gets on. Oh, I guess this is the part where actually he's asking the assistant to help curate his shopping choices. And then at some point he actually switches to I've put something on and like what's your feedback? Or what color should I get this in? Totally, totally cool. Like that is so wild. So let's go to the next one. I love this tweet. This is from a wild. back. I don't know if these people are still in good standings with you or not. So the tweet is that they were asking for female-founded CEO-led portfolio companies. And you tweeted that women outnumbered in the latest board, which congratulations. Thank you. That's awesome. That is so cool. We need more of that, especially in this AI space. We are in the middle of this bubble. We are in the middle of this bubble and there are 50% women in this world and there is 0.2% women at these meetups sometimes and I want to change that and I love that we need to figure out how to change that. Do you have something that you think that we could do to make this feeling, to make this feeling that happened afterwards? Make it go boom there's a bunch of women here. What can we do? Well we love your team at Cloudflare. I mean I know you guys have such a list. and others i think we always again try to be intentional and always try to make sure um we aren't passive yeah about being aware of who's around yeah and realizing that actually a lot of all of us are committed to excellence and want to build great things yeah that's what drives me crazy the most the idea that we have to do that it's sort of out of bounds for um people from you know underrepresented communities women and otherwise you know to be present um my friends growing up your friends growing up some of the most passionate nerds yeah you know women in in fantasy um black women black men you know sci-fi nerds yeah just they've envisioned the future too so we just i think have to allow just be aware of who we bring into spaces right right and just recognize excellence anymore that wasn't much of an answer no it's just they're they're there can you can you just open your eyes yeah and that's what i mean about like we there are more and the number of women at our board meetings has increased since i last tweet awesome awesome congrats that's great series b was led by the wonderful rosan winchak at renegade um series a we were lucky to partner with blackie groom who's wonderful um jenny left court led our seed uh Back in 2019, our seat for the API, and that she was joined by, Jenny was there, joined by like Katie Stanton at Moxie. And April Underwood. And April Underwood. Because you've been in the space a long time, and so you've seen things change. I think networks matter. And so I'd say two things about that. One is if you make an effort to expand your network, as Nina was saying, people are around. Like, go meet them. That's often really helpful. to you personally uh just broadening out your network uh and then inviting people into your network as well so there's that cross-pollination yeah i mean i think jenny introduced it to katie and he uses to april so we got you know this really great additional firepower on our cap table because you know there's a network effect yeah yeah that's awesome i think calling it out you know like i feel like when there's a mantle it's all men right there's like there's no need there's no need for that we need to call that out i think when we see that there's people there's there's voices that are super important here that are not being heard and because they don't feel invited i think i feel like it's more inviting i think that's a really good example and i think it ties into what you were saying about the network effect because as soon as um we realize that dynamic if there's a situation where there's a mammal yeah and we figure out the one maybe there's one person we can reach out to they may know someone else yeah right and that'll drill down to the next panel yeah yeah and say this is an inviting space because like i i hope that i mean you'll throw inviting spaces and we try our best to throw inviting spaces but i don't know if we're doing it i don't know if we're doing it i feel like we're not doing it and like i'm going to call ourselves out on it you know i think we need to get better at this and i'd like to see that i'd like to see more boom more of that boom all right uh this one this is a very recent tweet that peter peter pulled up today where we're talking how there's a big price drop on some of this real-time clients but then Quinn came back and said actually it is very expensive and we talked about some dots earlier and if I look at this graph here you're talking about a one-minute conversation on the real-time API that's out because it can it starts all the conversation it gets progressively more expensive you could talk about a 10-minute conversation that's six dollars and that's expensive it looks like open I fixed that today with automatic caching though oh really which has a 80% lower cost. The cash tokens have an 80% lower cost. So we need to redo that math. We were, that's a middle of a thread in which we were trying to make sure everybody in the thread understood the math, because if you understand the math as a starting point, then you can figure out like how much it's going to cost for your application. Because that's hard, right? That's really challenging. It's really hard. And LLMs being priced per token in general is a complicated thing to wrap your head around. And then in a use case like a multi-turn conversation, it's even more complicated. So we have. have you know spreadsheets we share with people that are like pricing calculators they have assumptions embedded in them right though so the assumptions might not be exactly right for your use case so you have to learn how to do those evaluations and monitoring so you know how much you're going to be spending but yeah the first release of the open AI real-time API was very very expensive but I think they just fixed that today which is one of the things that you probably need to bet on in generative AI is that stuff is getting cheaper every it's going to get better like just just kind of lean in to that's great advice i think that like build your prototype make it work yeah well you have a bunch of stuff that you offer totally for free which is also mind-blowing yeah yeah yeah because we want people to be able to do that we don't want people to be be shy away from that so that's what we that's where that's where we head yeah awesome um uh this one i love i love this one i got to come see you at this place called solaris can you explain a little bit about what's going on over there at solaris yeah it's amazing Thomas and his co-founder Jacob have built. It's an example of what's so great in San Francisco right now. So Solaris is a community and co-working space in the mission, and it's full of AI developers and AI teams. I mentioned earlier, I think we talked about our team being remote, so it's fantastic to be able to come to this place and be around this energy and meet other developers who are super excited about everything that's happening. AI. Yeah, for sure. It's also mostly pretty young companies, which is a fun, very fun to be around companies that are just trying to figure it out at the very earliest stages. And I think back to the gender balance thing you were talking about, I don't actually know the numbers, but Solaris seems like a relatively gender balanced space in terms of the kind of ecosystem. Yeah, it's interesting how things are changing. And I think Solaris does a great job of being a welcoming I really enjoy how early stage everyone is there because I think that's one of the things Thomas and Jacob have really invested in is helping developers come together to sort of figure out your ideas and what you want to do. How many companies are in there now? So it's an, what was the building? Do you remember? It was a city, a department of public works building. If you had to get a building permit five years ago in San Francisco, you probably went and stood in a window. it's got a little bit of that vibe but then like it's like we're gonna upcycle this thing you know like and it the vibe is great there like really really cool you can feel the creativity i don't know if it's the overhead lights buzzing or it's the creativity but uh uh how many how many companies i think there must be 40 or 50 because they have both some you know small private offices but they also have a bunch of hot desks yeah um you know it's sort of we work model yeah but very much a oriented sort of take on it yep yep i saw i saw people from different companies just hanging out on the couch talking about what their day was like and just what that does what that does sharing that knowledge and everybody being ai companies is so so cool such a cool place so if you get a chance to ever go to a meet up there at the solaris building go just just go check that out you might have to take your shoes off that's kind of a thing sometimes they changed that they changed the rules when we moved into the new bigger solaris building they were like this is gonna be a shoes off space but i don't think that lasted it didn't last it's It's hard in an office. It is time for our next game, which involves a camera. This game is called Coming Soon. We're going to make a movie. So we're talking about the future a lot. And I believe a lot of the inspiration for movies comes from the past of these movies that we grew up as. In 20 years, it's going to be 2044, which sounds super cool. But also, what does it look like in 2044? 2044 with the work that you're doing because you're building that future I have said that like three or four times and I mean that you are building that future what does voice interaction look like in 20 years I mean I think we're just going to be talking to AIs all the time okay we're just going to take for granted that we've got these AIs sort of in every online space yeah that we're part of and that they talk back to us and that they listen and they're both conversational and and kind of just passive co-pilots that are along for the ride and helping us in all sorts of ways. So tactically, the ways we interact with information just should be much more seamless. Kind of getting into the details, like you don't have to open your fridge to know what's in there. So we did an internal hackathon. I said, could someone please build the interactive shopping list where I can just say. Give me what's missing. add this right right like you can keep track of when i put the pasta in the fridge i can say okay i put anchovy pasta in the fridge and then when we finish it i'm like oh the anchovy pasta is done yeah right so like i gotta cook that again please put that on my shopping list yep so i spoke seamlessly um with real time voice ai you know it's logging it and then when i get in my car i say to the app which is also on my phone like maybe it's through a speaker in the kitchen maybe it's on the fridge 2044 go go nuts where is it and like i walk up the door i'm like what am i getting again yeah and it says to me you you gotta get you ran out of anchovy pasta you want to make a fresh batch so all that stuff just seamless yeah seamless that's like five years away that's true two years away true so that's what i was that's tactically right do you think like conceptually how we conceive of ourselves will be different because of these co-pilots well i i think part of the thought experiment is If you rewound 15 years, what would it take for you to imagine TikTok? That's a good way to do it. When we pitched daily to seed investors in 2016, almost all people we already knew because we'd been around and we were lucky to have this sort of network that included a lot of professional investors. Even so, we were like conversational voice, video, video calls. You're just going to do lots of that stuff every day. And probably 10 out of 12 of the investors we pitched said, I'm not going to do video calls. Like, I prefer the phone. And we're like, wait, what do you do for a living again? Are you a professional technology investor? It did take four or five years for that to shift. The pandemic definitely sort of shifted that balance faster than it would otherwise. It feels to us sometimes when we talk about talking to LLMs, the same dynamic. Like, oh, I don't really need to talk to. right yeah but you're going to you're going to yeah i mean another way of re another thing we heard was um well zoom has figured it out yeah when we were raising like many years ago which in one way like it's a testament to zoom being so great like a fantastic technology fantastic company um but the idea that you know the evolution of video technology just stops right there so i know this is a cop-out um but i almost cannot like yeah what will it be in 2040 And I totally agree. The stuff I was talking about was like five years, but it's going to be like, I mean, if we really look out in the future, I just think we've got these LLMs as as as interaction partners in ways that just feel very human. Like they're not tools. They are part of our social fabric. You're going to have LLMs in your group chat. You're going to have LLMs in all your social spaces. And it's hard for us to sort of turn the corner mentally. and like wrap our heads around that like it would have been hard to imagine tiktok like yeah the first time you saw tivo could you trace that through and imagine tiktok probably not going forward but going back you can you could see how that happened yeah that makes sense yeah so so for this we're making this movie there's there's some plot points so we're gonna make we're gonna make a movie this is uh using the the multimodal playground a lot of inputs and outputs from cloudflare that's awesome so we have a we have your picture here uh we have we have the people are named uh quinn and nina and we're going to run this we're going to say generate describe the people we're going to find actors that will play you in this movie um and then all good plots have a problem right what happens for what could stop that future what would stop that future from happening where everybody's talking to everything not enough gpus not enough gpus but they almost give up due to not enough GPUs. I love it. What was the great Foul swag? GPU poor. Our friends at Foul and the hats. It's a GPU poor. And then finally everyone what's the finally of this? They just talk everyone talks to everything. I feel like that's it, right? We're just talking to everything. If you're going to be sort of geeky about it, it's that talking to other people has a bandwidth constraint that there has to be another person who's there to listen no more bandwidth constraints on conversation conversation too cheap to meter all right so you don't have to listen to me anymore because like i have i'll have llms to listen to me all right that's gonna solve some problems it's always a big problem yeah mary co-founder all right so look to the camera oh your slurpees are in the shot that's a beautiful shot it's gonna talk All right, so it's going to run. It's going to push this through. We're going in here. And it described you here. Yep, so we got good dark brown hair, dark sunglasses. The name of the movie is 2044, The Future of Speech. That's pretty good. That's pretty catchy. We can workshop it. In fact, I went and had it generate a little slogan, movie tagline, if you will. my voice doing the movie in a world exactly it's called a benedict cumberbatch that's that's a pretty good one 2044 the future of speech starring benedict cumberbatch as dr quinn the boundaries between human and machine begin to blur in this groundbreaking tech thriller cumberbatch and gal gadot shine as unlikely allies navigating the cutting edge of innovation yes as old conversation fade to silence the world is left to ponder what exactly does it mean to connect with the people around us when the voice of the future is all that's left whoa that's a little dark at the end craig you you know that steven spielberg made the demo reel for my previous story for the next the next appearance there's your movie poster that we generated which is just crazy that we can make a movie poster out of nothing and it's yeah that's happening. I also had to generate a review, a fake review based on the movie poster because I think that's what happens. I think what happens is that people say, oh, voice AI, I don't like that. They've never played with it. They've never actually seen the movie. So this says, I'd say three out of five stars. It's more than the same that I expected about a futuristic play, which probably based on that cover, I think that's probably what's happening. Yeah, we'd workshop the cover. So it's like all our team. and Dr. Quinn and Dr. Nina Sancho. You don't have a PhD either. I know, neither of us have a PhD. But this is a movie that I'm ready for. I think that we're building the future there. Thank you so much for coming on the show. I would love to see this. I give the review of 2044 that you're bringing us five out of five stars. So thank you. And thank you for being here. You