AI Tinkerers - "One-Shot"
00:56:13 · Joe Heitzeberg, Luke Harries
[00:56:13] S1: Small club, microphone, crowd noise, you know, and have it try to generate comedy. [00:56:19] S0: Wow. That's very cool. That's very fun. The... Yeah. The interesting thing is if you think about audio Yeah. As a modality, like, do end up... So we currently have like a sound effects model. We have a text to speech model. We have... But I could definitely imagine a world of like more and more becoming, you know, the one... I mean, to digress slightly. If you think about audiobooks and what's an audiobook, it's normally one person reading out an entire book with no special effects, no sound effects, and only one character. Like, you're listening to Harry Potter with Stephen Fry playing both teenage Hermione and teenage Harry Potter. And and the future which I could really imagine is, particularly of audiobooks, is like... And we're trying to build this with our studio product. You feed in a PDF and it goes, okay. These are all the different people. Okay. This is where we should have, like, suspense music in this particular sound effect. Yeah. I think that's really exciting for just audio overall. [00:57:15] S1: Yeah. It is. You think about... Sometimes you see museums have done that with their audio thing. [00:57:19] S0: And there are audio dramas, but they're so rare and the production The cost production's crazy, [00:57:25] S1: and then it's not that good, [00:57:26] S0: actually. Yeah. Yeah. [00:57:27] S1: Yeah. So he's he's kinda tragic because they... You know, they spent a lot and it's not that good. Yeah. Yeah. It's not Hollywood.
00:57:27 · Joe Heitzeberg, Luke Harries
[00:57:27] S1: Yeah. So he's he's kinda tragic because they... You know, they spent a lot and it's not that good. Yeah. Yeah. It's not Hollywood. [00:57:32] S0: There's there's an interesting thesis we have, which is the creator will become the consumer. [00:57:37] S1: Yeah. [00:57:37] S0: So currently, lots of people spend hours and hours, like, manually creating these audiobooks. Mhmm. And and I think that will work for, like, the real big blockbusters with the fantastic... With, like, when you get a voice like Stephen Fry. But I think more and more, it will move to, like, the end user having content dynamically generated on the fly, like, exactly... They're like, I love this voice and this voice in my language. You know, it says Serbian, and then it's just, like, generated live for you. [00:58:07] S1: Mhmm. Yeah. That'll be a... That'll be fun. It's it's great to see that. And so so automated so you could have it. Your cake could eat it too. So we've seen, you know, the rise of these... Any unexpected use cases where you're like, yeah, people using ElevenLabs in this way we never anticipated.
00:58:07 · Joe Heitzeberg, Luke Harries
[00:58:07] S1: Mhmm. Yeah. That'll be a... That'll be fun. It's it's great to see that. And so so automated so you could have it. Your cake could eat it too. So we've seen, you know, the rise of these... Any unexpected use cases where you're like, yeah, people using ElevenLabs in this way we never anticipated. [00:58:23] S0: I'm just continually amazed by how big companies are being built on conversational AI. Because if you think about the number of jobs which are, like answering the phone. And so one example, hotel receptionist in Guatemala. Mhmm. Apparently, you can build a massive business building AI just for hotel receptionists in Guatemala. Yeah. Because currently, they're spending hundreds of dollars per month on a hotel receptionist per hotel. Yeah. And so the the biggest surprise is pretty much every single niche, whether it's like dental clinic receptionist, whether it's customer support, whether it's sales, like, every niche of every country of every industry [00:59:09] S1: Right. [00:59:10] S0: If you... I think if you just go really deep in it, you could build a massive business because the amount which is currently spent. And then you use like that as your wedge into all the other tools and workflows and becoming a system of record. [00:59:21] S1: That's amazing. I have to say, like, I think for my tinkering in Gen AI and things being so amazing, and so much experimentation happened, this is very early on. I think ElevenLabs was also the first where the whisper rumor numbers of the revenue growth [00:59:37] S0: Yes.
00:59:37 · Luke Harries, Joe Heitzeberg
[00:59:37] S0: Yes. [00:59:37] S1: Was... It was the first case where I was, like... Felt enormous sense of relief because the number was large. This was a few years ago now, you know, very early before. I mean, now we talk about Cursor's revenue and everybody brags about their revenue, but before anything, it was a whisper number, and I was very relieved. I was like, great. There's there's there's gonna be big businesses in this space. [00:59:58] S0: Yes. [00:59:59] S1: And I'm sure just from what you described, it's continued. Let's see. How... Switching gears really quick, and I know this is going pretty long. I don't know how much time you have before we wrap. But back to your... The start of this conversation, you know, your Renaissance background. Right? And so you're not just an engineer, obviously, but you have this ML piece, engineering, growth, product, entrepreneurship. What are the meta ways in which you use AI throughout all these things? Yeah. So I've talked to some people around their meta workflow. You know, how do you go from customer input to like shipped code and and BRD, PRD, if you wanna call it that or... Yeah. What is your... You know, because I think I find that people like you have are using AI... AI first, AI native. Yeah. What are those tricks that
01:00:51 · Luke Harries, Joe Heitzeberg
[01:00:51] S0: you just come workflow is interesting. We've set up more and more of this recently. One I've been really surprised how useful it is is Gong. So we have a sales teams who are constantly doing these sales calls. Mhmm. And every sales call, there's just like an absolute gold mine of, like, user wants, user needs, objections Yeah. Competitor analysis, where you're falling flat, etcetera. There's so many of it. Well, we have now tens of salespeople all gathering all these tens of salespeople making tens of cool hours of data. And what we've set up at Gong is you... It automatically analyzes all the transcripts post the call for particular threads and pulls them out. So the, say, conversational AI AI team has a constant thread of just, like, fit every single call. This is, the areas of weakness that we need to fix, and then that goes into the road map. And that... I think there's definitely more we could do in terms... Like, the [01:01:46] S1: I can envision a... Something that's an agent that's looking at your code base and that that. And it's like, it's literally a twenty four seven dynamic road map. [01:01:55] S0: Yes. Yeah. Yeah. Yeah. And there's this interesting thing of like, if you view software engineering as a system or as a factory [01:02:02] S1: Yeah.
01:02:02 · Joe Heitzeberg, Luke Harries
[01:02:02] S1: Yeah. [01:02:03] S0: Like what goes in, user wants and needs. Yeah. What goes out, ship code. Yeah. And like, in the middle, you have different people doing different parts and shaping different things. Yeah. I don't think we're yet, you know, at the level where we're we're fully automating. But maybe to name a few other tidbits of like processes we've done. So one is we've done that. [01:02:22] S1: Uh-huh. [01:02:22] S0: And the other one is then like, okay. We actually sign up a user. They have a great case study. Then we now have this great case study flow, which is we do a half an hour call with them using granola with their permission. Yeah. We then take the granola and paste that into a Claude project with previous examples, what makes a great great case study, etcetera. Yeah. And then that outputs first shot, like, it's 95% of the way there an excellent case study. And you're able to get that because you're feeding in that rich granola set of those. Cool. That's [01:02:51] S1: great. That's that's And going on the blogs of case study. Heard this website. [01:02:53] S0: Yeah. Yeah. That thing goes on case study. And it will also say, like, we suggest you add this picture. Yeah. And then that goes to our to our team. That's great. And [01:03:01] S1: It pains me to think of all the people who are wanting case studies or producing them in some not AI needed way. [01:03:08] S0: I know. [01:03:09] S1: And to hear how simple that is. Right? Yes. Surviving.