Generating your AI Media Empire - with Youssef Rizk of Wondercraft.ai
00:00:00.000 | (upbeat music)
00:00:02.580 | - Welcome to the Latent Space Podcast,
00:00:07.940 | where we dive into the wild, wild world
00:00:09.760 | of AI engineering every week.
00:00:11.840 | This is Anna, your friendly neighborhood AI,
00:00:13.920 | and I'll be standing in for Alessio today.
00:00:16.320 | Yes, you heard right, AI is taking our podcasting jobs.
00:00:19.640 | We flew all the way to London to interview Yousef Risk,
00:00:22.480 | co-founder of Wondercraft AI,
00:00:24.420 | which has created the number one piece
00:00:26.080 | of AI generated content enjoyed
00:00:28.040 | by the Latent Space community.
00:00:30.160 | We ask him how he arrived at his idea,
00:00:32.440 | what the future of commercial AI generated content
00:00:35.160 | looks like, and confront him
00:00:36.800 | with the hardest question of all.
00:00:38.680 | What is his moat as an API wrapper startup?
00:00:41.780 | At the end, we even have him turn the tables
00:00:44.140 | and do a customer interview with Swix.
00:00:46.360 | There's lots of audio goodies in this one,
00:00:48.080 | and bonus 30 minutes video on youtube.com/latentspacetv.
00:00:52.760 | Watch out and take care.
00:00:54.800 | So we're in the studio here in London with Yousef, welcome.
00:00:58.880 | - Thank you, great to be here.
00:01:01.120 | - It's been such a joy listening
00:01:03.400 | to Wondercraft's podcasts over the last four or five months.
00:01:07.080 | You guys have been around for only five months.
00:01:09.800 | And as you know, I am one of your podcast's biggest fans.
00:01:14.800 | And I think that it's super interesting
00:01:17.180 | because I talk to a lot of vendors,
00:01:20.560 | effectively people who create services
00:01:23.180 | for other developers to build.
00:01:25.000 | And you are at the application layer,
00:01:27.260 | which is great and challenging for me as a podcaster
00:01:30.360 | because you have some secret sauce
00:01:32.040 | that you're not gonna share.
00:01:33.320 | But I also wanna just talk to you
00:01:35.800 | as someone who's evaluated a lot of things
00:01:37.900 | and built something that I actually use every single day.
00:01:40.920 | So that's the context.
00:01:42.760 | - Great, great.
00:01:43.840 | - How do you feel when I say these things?
00:01:45.680 | Is that exactly what you're going for?
00:01:47.560 | - Yeah, yeah, yeah, yeah.
00:01:48.400 | So it definitely makes sense, right?
00:01:50.040 | Resonates definitely on the application layer,
00:01:52.000 | and that's definitely by design.
00:01:54.240 | - Yeah, and we can talk about the origin story
00:01:55.940 | leading into Wondercraft.
00:01:57.840 | But just to learn a little bit more about you,
00:02:00.000 | you grew up in Egypt?
00:02:01.000 | - I grew up in Egypt, yeah.
00:02:01.840 | I spent the first 18 years there.
00:02:03.560 | - Cairo.
00:02:04.480 | And then you came over to the UK.
00:02:07.240 | You got your master's in triple E at Imperial.
00:02:10.320 | - Yeah, it's a funny story.
00:02:11.280 | Actually, in the UK, I don't have a bachelor's.
00:02:13.880 | I only have a master's.
00:02:15.240 | - How?
00:02:16.080 | - You do this four-year integrated bachelor plus master's
00:02:18.160 | 'cause in the UK, the bachelor's is three years.
00:02:19.840 | - Yeah.
00:02:20.680 | - But you can do one more year,
00:02:21.500 | which basically counts as a master's.
00:02:22.520 | So you graduate with one degree,
00:02:24.000 | which is called a master's of engineering.
00:02:25.640 | - Yeah.
00:02:26.880 | - So I did that.
00:02:27.720 | Actually, the reason I studied electrical engineering
00:02:29.120 | when I started, I hated software in high school.
00:02:31.640 | I was like, this is something I'm never gonna do.
00:02:33.120 | - Yeah.
00:02:33.960 | - Didn't qualify as real engineering.
00:02:36.040 | But, so the reason I applied to electrical engineering
00:02:40.040 | is that I'm a big car guy.
00:02:41.200 | I love cars.
00:02:42.320 | And at that time, around 2013, 2014,
00:02:45.440 | it was, all the big rage was starting to be
00:02:48.440 | these hybrid electric cars.
00:02:49.680 | So I was like, cool, this is what I wanna do.
00:02:50.960 | I wanna build cars.
00:02:52.020 | So I go start the course.
00:02:53.440 | Our first, very topically, our first project
00:02:58.040 | was building a little robot car that follows a line.
00:03:02.140 | And oh my God, was it boring.
00:03:04.840 | It was the worst thing I've ever had to work on.
00:03:07.240 | So, and then at the same time,
00:03:09.000 | we started this intro to programming course,
00:03:11.240 | which I loved.
00:03:12.080 | It was super basic C++.
00:03:13.800 | - Yeah.
00:03:14.640 | - I was like, this is cool.
00:03:15.460 | There's this instant gratification that you get
00:03:17.240 | from just building a program, seeing it run,
00:03:18.760 | and then seeing that it works or it doesn't.
00:03:21.360 | So all of a sudden, I started,
00:03:22.640 | what I started out with, I started hating,
00:03:24.420 | and what I thought I hated, I started loving.
00:03:27.260 | - I had a kind of similar journey
00:03:29.160 | with finance and programming.
00:03:30.720 | I thought that I wasn't cut out to be a programmer.
00:03:34.360 | So I went in the finance route for like 10 years,
00:03:37.140 | and then realized that software is eating the world,
00:03:40.320 | which is cliche, but true.
00:03:42.520 | - Yeah.
00:03:43.360 | - And I find that a lot of triple E people
00:03:44.680 | actually end up this way.
00:03:46.360 | Maybe just 'cause software is easier than hardware.
00:03:49.400 | - It's easier, and it's also just,
00:03:51.280 | hardware's so abstract.
00:03:52.280 | The only thing you're looking at is this PCB
00:03:53.880 | with a bunch of resistors and capacitors plugged in,
00:03:55.800 | and if you wanna build some board,
00:03:57.980 | you have to go ship it, send it.
00:03:59.280 | - Yeah, yeah.
00:04:00.120 | - It's just the iteration process takes so long.
00:04:02.000 | That being said, it's hugely valuable.
00:04:04.300 | - Yeah, yeah.
00:04:05.140 | - The improvements that hardware has driven,
00:04:08.200 | combined with software, obviously both huge,
00:04:10.300 | but hardware is also a huge part of this.
00:04:11.880 | - Yeah, yeah, totally.
00:04:12.780 | And you having that background is something
00:04:14.760 | that I would probably never be able to pick up
00:04:17.120 | as a non, like outside of a university setting,
00:04:20.040 | 'cause you just had to do all these courses.
00:04:21.520 | - Yeah, yeah, yeah, definitely.
00:04:23.620 | - Cool, and you actually called out that
00:04:26.840 | the Google Summer of Code was important to you
00:04:28.680 | in, on your LinkedIn, in terms of learning to code.
00:04:31.680 | Is that, is that, no?
00:04:33.120 | Okay, we can cut that.
00:04:35.080 | I don't know, it's on your LinkedIn somewhere.
00:04:36.360 | - Is it?
00:04:37.200 | - Yeah, yeah.
00:04:38.040 | - I think I did the--
00:04:38.860 | - You did some boot camp.
00:04:39.700 | - It wasn't Google.
00:04:40.540 | - Okay, okay.
00:04:41.360 | - It was, I forget the name,
00:04:42.500 | but if you have your LinkedIn open,
00:04:44.900 | I can briefly do a talk about it.
00:04:46.500 | But it was basically this like summer,
00:04:48.180 | I wanted to learn code,
00:04:49.180 | this was in my third year of university,
00:04:50.780 | so I did this boot camp in Oxford.
00:04:53.360 | - Yes.
00:04:54.200 | - It was called the Summer Academy of Code,
00:04:57.300 | and basically it was just like a one-week thing
00:04:59.160 | where you go from the basics of just building a web app,
00:05:01.860 | so all the basics of Node, React, JSX, all this.
00:05:06.860 | I was already kind of familiar with it,
00:05:08.340 | so it was kind of a repeat, but it was good to have that.
00:05:12.040 | It was super interesting, actually.
00:05:14.200 | I mean, funny story there is like when I was in Oxford,
00:05:17.640 | I was having a great time.
00:05:18.680 | At the time of that boot camp,
00:05:20.200 | I got this like, the worst gum infection ever.
00:05:25.680 | - Okay.
00:05:26.520 | - It was the worst pain I've ever been in,
00:05:28.160 | so as I'm doing this like one week in Oxford
00:05:30.240 | during the summer boot camp,
00:05:31.620 | I'm just in agonizing pain the entire time
00:05:34.400 | trying to learn about JSX or whatnot.
00:05:37.880 | - Yeah, yeah, yeah, that must have added extra difficulty.
00:05:41.120 | I know it's such a pain to learn the JS ecosystem already.
00:05:44.060 | The reason I ask is I'm a career changer myself, right?
00:05:48.440 | I also got into tech through a boot camp.
00:05:51.560 | I think that these support systems in each country, right?
00:05:54.580 | San Francisco has a lot of boot camps.
00:05:56.260 | New York is the boot camp that I went through,
00:05:58.600 | and you had the one in Oxford.
00:06:01.000 | It's super important for people to ramp up,
00:06:03.960 | and I want to encourage more of those things
00:06:05.960 | and say that it's basically okay.
00:06:07.560 | It's not as prestigious as a degree,
00:06:09.600 | but it's probably more practical.
00:06:11.440 | - Yeah, I think to everyone listening,
00:06:13.400 | yeah, I did a triple-A degree.
00:06:15.000 | For those who don't know,
00:06:15.960 | that's electrical and electronic engineering.
00:06:18.200 | Sounds similar enough to computer science.
00:06:20.840 | It really isn't.
00:06:21.680 | Like, you still do some software,
00:06:23.240 | but the thing I do want to highlight is that all of,
00:06:25.920 | I would probably say half,
00:06:27.760 | and this is a number out of thin air,
00:06:28.960 | so don't quote me on it,
00:06:30.120 | half the software engineers probably are self-trained.
00:06:32.200 | - They're self-taught, yeah, yeah.
00:06:33.480 | - And that's kind of the beauty of this
00:06:34.800 | is that you can very much learn this by yourself.
00:06:37.080 | I do agree with you.
00:06:38.080 | Learning electronics by yourself is a little more difficult.
00:06:40.760 | You just need the equipment,
00:06:42.400 | but something like software,
00:06:43.720 | it's super easy to get started.
00:06:45.320 | - I mean, now I'm having to catch up on it
00:06:48.120 | because any time it comes to GPU power calculations
00:06:51.880 | and RAM calculations, it is electrical engineering,
00:06:56.680 | and so there's a lot of actual science in there.
00:06:59.360 | You then spent four years at Palantir
00:07:02.400 | as a forward-deployed engineer.
00:07:04.680 | I actually interviewed for that job.
00:07:06.600 | I didn't make it because I slept through my interview.
00:07:08.760 | It's a different story,
00:07:09.600 | but forward-deployed engineer is a class,
00:07:11.680 | I think it's a role that Palantir invented.
00:07:14.120 | - Yeah, look, Palantir's a great company.
00:07:16.800 | I think it's like there's a technical aspect
00:07:19.960 | and a motivational aspect that they look at
00:07:22.400 | when they're evaluating it.
00:07:23.760 | I think they just have this interviewer process,
00:07:27.400 | or it kind of stabs them in the back sometimes
00:07:30.160 | because they tend to just get people
00:07:32.000 | who are super entrepreneurial,
00:07:32.960 | so the nature is they're always looking
00:07:35.360 | for the next challenge.
00:07:36.440 | I think when I thought of it is everyone at Palantir,
00:07:38.800 | or most people at Palantir,
00:07:40.080 | were the kind of people to do the group project themselves.
00:07:43.240 | You know what I mean?
00:07:45.040 | It's these kind of people.
00:07:45.880 | They just want to be challenged,
00:07:46.920 | so the reality is that you end up having a lot of people
00:07:50.160 | just leave to start their own thing.
00:07:51.400 | That's the most common thing.
00:07:52.380 | You find someone at Palantir,
00:07:53.400 | and then you guys go and start your own thing.
00:07:55.760 | - And it's where you met Dimi.
00:07:57.040 | - It's where I met Dimi, yeah.
00:07:57.880 | We actually started on the same day.
00:08:00.040 | Matti also started Vigdir on the same time,
00:08:01.880 | not the same day as us.
00:08:04.120 | So really, the thing that made it super hard
00:08:07.000 | to leave Palantir was just the people.
00:08:09.240 | You're working with such a high caliber of people,
00:08:11.740 | but interpret that as real,
00:08:15.920 | but also just people you can be friends with.
00:08:17.920 | So I still have a lot of friends from there.
00:08:20.520 | The job itself, Forward Deployment Engineer,
00:08:21.840 | is a super interesting job
00:08:22.760 | because it is kind of at this intersection
00:08:26.360 | of being an engineer, so a software engineer,
00:08:28.920 | but also still doing business-related things.
00:08:31.200 | - Yeah, solutions, architects maybe.
00:08:33.680 | - Yeah, so part of the job was a solutions architect.
00:08:35.600 | Part of the job was reviewing contracts.
00:08:37.320 | Part of the job was doing sales.
00:08:38.640 | Part of the job was coding things.
00:08:40.120 | Part of the job was interacting.
00:08:42.200 | So many different things,
00:08:43.040 | and I think that is a really good foundation
00:08:44.960 | for someone who does want to start something in the future.
00:08:47.840 | You just do everything.
00:08:48.920 | - So kind of an endorsement of that job
00:08:50.620 | if people want to--
00:08:51.460 | - Huge endorsement of that job.
00:08:52.360 | - Okay, excellent, amazing.
00:08:54.320 | Yeah, Palantir is surprisingly strong
00:08:56.360 | here in the London tech circles.
00:08:57.680 | I have a number of friends who are all ex-Palantir.
00:08:59.680 | - I think it's actually the biggest offices in London.
00:09:03.120 | - Surprising, because I think of it
00:09:04.800 | as like a US defense company.
00:09:06.320 | - Yeah, yeah, yeah, yeah.
00:09:08.800 | - Then you started Moonshot for nine months,
00:09:10.240 | which is pretty important in your journey.
00:09:12.080 | I'll bring it up to Wondercraft,
00:09:13.600 | and you started Wondercraft in April of this year,
00:09:16.040 | and it's been about five months going through YC
00:09:18.920 | in the winter batch.
00:09:21.080 | - Summer of '22.
00:09:21.920 | - Summer of '22 batch.
00:09:23.500 | Okay, cool.
00:09:24.340 | What is Wondercraft?
00:09:26.200 | - Nice, Wondercraft's a podcast builder
00:09:28.040 | that uses hyper-realistic AI voices to create podcasts
00:09:31.000 | and make that whole podcast creation process super simple.
00:09:34.080 | So a super simple example is you can,
00:09:36.160 | you publish a bunch of blogs.
00:09:37.600 | You can take that blog, put it in there,
00:09:38.980 | it'll just convert it to an audio-friendly format
00:09:40.800 | that people can listen to.
00:09:42.840 | It's just sometimes it's a bit more efficient
00:09:44.360 | to listen to things rather than read them.
00:09:46.720 | I gave the very, if you want to think of it specifically
00:09:49.400 | as what it is at the current instance in time,
00:09:51.640 | that's what it is.
00:09:53.000 | What does it strive to be is a little slightly different,
00:09:55.400 | because what it really strives to be
00:09:57.440 | is it strives to be this platform
00:09:59.440 | with the mission of expanding access to content.
00:10:03.080 | - Okay.
00:10:04.680 | - And I mean this in a variety of different ways.
00:10:06.280 | Some people just are able to consume content.
00:10:09.560 | We have this whole debate in education.
00:10:10.860 | It's like, are you a visual learner or an audio learner?
00:10:12.960 | What do you do?
00:10:13.800 | People just consume content better in different ways.
00:10:15.600 | I'm a visual learner.
00:10:17.200 | I need to see things.
00:10:18.120 | So for me, actually, it's sometimes a little better
00:10:19.840 | to read the blog.
00:10:20.900 | But if we're just talking about,
00:10:22.520 | I want to get a lot of information,
00:10:23.860 | podcasts are great because you can just do them
00:10:26.080 | while doing something else.
00:10:27.480 | There's a reason that podcast functionality
00:10:30.240 | is so natively embedded in all these smart speakers.
00:10:33.520 | It's just 'cause you're doing anything at home,
00:10:35.200 | just put on a podcast.
00:10:36.620 | So really what we're trying to do is,
00:10:40.520 | podcasts is the first instantiation of that,
00:10:43.040 | which is how do we expand access to content?
00:10:45.760 | But it expands so much more, right?
00:10:48.740 | Instead of just going to, I don't know,
00:10:52.960 | we talked about this blog to podcast.
00:10:55.200 | You can go blog to video.
00:10:56.880 | You can go podcast to blog.
00:10:58.240 | You can go podcast to Twitter.
00:10:59.680 | The permutations are, frankly, endless.
00:11:02.180 | Basically, it depends on how many platforms there are
00:11:03.640 | that people consume things on.
00:11:05.640 | But that's essentially what we do.
00:11:08.120 | The use cases for this are pretty interesting.
00:11:10.420 | The one that we just see immediate value in
00:11:15.280 | is just this ability to translate the content
00:11:18.120 | that you already have into other forms of content.
00:11:21.460 | If we just stick with that blog post example again,
00:11:25.080 | you've written, so a lot of companies
00:11:27.040 | might have this content team that focuses a lot
00:11:29.040 | on producing quality blog posts.
00:11:30.760 | Blog posts, they're good for SEO and whatnot,
00:11:33.960 | but they're not, sometimes they don't really achieve
00:11:37.480 | a specific goal or outcome that you want.
00:11:39.880 | One thing we see that is really useful for podcasts
00:11:43.440 | is they actually carry a lot more weight
00:11:45.720 | in credentializing you as a thought leader
00:11:48.280 | or your company as a thought leader.
00:11:50.080 | But like, we spent the last 50 minutes
00:11:54.360 | trying to set up this room to record the podcast.
00:11:56.360 | So it's not easy.
00:11:57.720 | It's a very synchronous process.
00:12:00.120 | Me and you have to find the time to go and sit here
00:12:02.680 | and record this, you have to come up with questions,
00:12:04.400 | I have to come up with answers.
00:12:05.920 | But this ability to actually just take the content
00:12:09.160 | that you have and transform it, it's pretty powerful.
00:12:12.080 | There's a lot of other use cases as well,
00:12:15.800 | which is just like, podcasts really,
00:12:17.960 | all they are is like, define a podcast.
00:12:22.000 | The line between, or the difference between
00:12:23.640 | an audiobook and a podcast, I guess,
00:12:24.920 | just the format and the length.
00:12:27.080 | - It's an MP3 on a RSS feed.
00:12:28.360 | - It's an MP3 with someone or something speaking, right?
00:12:32.120 | - I've actually played around a lot with this stuff,
00:12:33.760 | by the way, so I've done music-only podcasts
00:12:36.680 | where you just listen.
00:12:37.520 | Tiesto has been podcasting for 15 years,
00:12:40.200 | every single week, just DJing in his house.
00:12:42.880 | - It's just basically a radio show.
00:12:44.120 | - It's great.
00:12:44.960 | - It's just a radio show.
00:12:45.780 | - Async radio.
00:12:46.620 | - Yeah, so it's super interesting.
00:12:50.000 | But podcasts, okay, ignore the word podcast
00:12:52.200 | and just think of what we do,
00:12:53.040 | which is we help you create audio content.
00:12:55.840 | Super valuable for anyone who just needs that.
00:12:59.120 | If you can imagine a world in which,
00:13:01.400 | I don't know how to call them,
00:13:02.240 | like Calm or Headspace or any of these things,
00:13:04.240 | they can do a lot of their meditation like that
00:13:05.800 | super quickly.
00:13:06.920 | What you can get to is a point where you're doing
00:13:08.520 | these super personalized things, right?
00:13:10.920 | Because you just have the ability
00:13:12.200 | to scale the content production so quickly.
00:13:14.200 | Same with educators.
00:13:15.640 | I think there's actually, at this point,
00:13:17.120 | there's a few YouTube channels at this point
00:13:18.760 | that are all based on synthetic voices
00:13:20.560 | that produce a ton of educational content.
00:13:22.640 | - Any ones that do a good job?
00:13:24.080 | - I do, I just don't remember the name.
00:13:26.640 | - Yeah, okay.
00:13:27.480 | You can send it to me later, I'll stick to it.
00:13:30.040 | We can use Wondercraft to synthesize some extra clips.
00:13:32.880 | - Yeah.
00:13:34.440 | Now that being said, the problem with podcasts sometimes
00:13:38.680 | is, it's not sometimes, actually.
00:13:40.640 | Podcasts just have a slightly slow adoption rate.
00:13:44.880 | - Yes.
00:13:45.920 | - You're listening to a thing for an hour, right?
00:13:47.640 | Like, we, as a generation, don't have attention spans.
00:13:50.840 | It's the time and the attention span.
00:13:52.440 | Like, TikTok, give it to me in 30 seconds,
00:13:54.800 | rethink why clips are taking over.
00:13:56.160 | - 30's too long, man.
00:13:57.280 | - 30's too long, 10 seconds.
00:13:58.760 | 10 seconds with captions, I need to read it,
00:14:00.440 | and good.
00:14:01.280 | So what we also do is actually,
00:14:03.640 | and this is kind of still a beta feature
00:14:04.880 | and we're working to improve it,
00:14:05.720 | but we also let you take that podcast
00:14:07.800 | and then clip it into a video
00:14:10.560 | that you can go and share on socials, right?
00:14:12.200 | So it's this ability to take one form of content,
00:14:16.360 | produce it in a bunch of different ways
00:14:17.720 | that serve different purposes
00:14:19.280 | and be able to distribute it, basically.
00:14:21.320 | - Yeah, excellent.
00:14:22.800 | I want to go through features
00:14:23.800 | so that people have a high-level overview
00:14:25.720 | of what you offer.
00:14:27.880 | So I think at the core, it is basically two things.
00:14:30.440 | One is, you generate scripts,
00:14:33.000 | and that's optional, obviously,
00:14:34.280 | if you want to just write the script yourself,
00:14:36.680 | you can write the script yourself,
00:14:37.640 | but I think most of your users would generate a script.
00:14:41.080 | And then two is, from that script,
00:14:42.400 | you use AI voices, currently using 11 apps.
00:14:47.280 | Is that the rough flow?
00:14:48.720 | That's the really core, basic--
00:14:50.320 | - That's the core, basic.
00:14:51.160 | Obviously, there's a lot of plumbing on top of it,
00:14:52.640 | but that's the core.
00:14:53.640 | - And then you offer video clips for YouTube.
00:14:56.360 | You offer 28 languages that you can produce.
00:15:00.080 | You offer show notes production and podcast hosting, too.
00:15:03.240 | So they don't have to host it on Anchor.
00:15:06.280 | Don't host it on Anchor, by the way,
00:15:07.520 | if people don't host it on Spotify.
00:15:09.400 | Don't host it on Apple Podcasts.
00:15:11.200 | These people don't respect the RSS feed.
00:15:12.800 | Anyway, I have very strong feelings
00:15:15.160 | about preserving the sanctity of the RSS feed
00:15:17.920 | for open podcasting, and all these Spotify's of the world
00:15:21.560 | want to close the podcasting ecosystem.
00:15:24.360 | So I have this tirade about them.
00:15:27.360 | But yeah, those are your top-level features
00:15:28.560 | on your landing page.
00:15:30.160 | Anything that you highlight to go deeper on?
00:15:33.320 | - Yeah, I think those are the top-level ones.
00:15:35.420 | There's also, it's basically just a lot of
00:15:37.880 | also ancillary tooling that goes around all of this
00:15:40.240 | to just make it easier.
00:15:41.680 | The goal is like, every time we speak to a customer
00:15:44.360 | or someone who's thinking about it,
00:15:45.660 | they're like, yeah, literally yesterday
00:15:47.880 | I was speaking to a potential customer,
00:15:49.720 | and they're like, yeah, I just, you know,
00:15:51.400 | I want to make sure this isn't a distraction
00:15:52.760 | 'cause we don't have that much time to do this.
00:15:54.480 | - Yeah.
00:15:55.320 | - And really the whole point is that
00:15:56.140 | this doesn't take time, right?
00:15:57.580 | The whole point is to provide all the rails
00:15:59.600 | that make this not take time.
00:16:01.880 | And this comes with a million different things, right?
00:16:03.400 | Like we, you know, sometimes the AI voices
00:16:06.260 | don't really know how to pronounce a word.
00:16:07.720 | So we have a pronunciation feature.
00:16:09.060 | Go and define how you want that word pronounced,
00:16:10.920 | and it'll take care of it.
00:16:13.000 | If you are, we obviously have that hierarchy
00:16:15.760 | of like a podcast, an episode,
00:16:17.360 | and then all of that gets published in RSS feed
00:16:19.200 | that you can just upload to Spotify,
00:16:20.560 | and we'll host that for you.
00:16:21.960 | But what you also have is just like, you know,
00:16:25.720 | maybe you want some defaults, right?
00:16:28.120 | Every podcast needs some defaults.
00:16:29.920 | - Intro, outros.
00:16:30.760 | - Intro, outros, the music, the speakers.
00:16:33.320 | - Yeah.
00:16:34.160 | - We're working on adding templates
00:16:35.160 | for the kind of podcast that you're doing.
00:16:36.600 | Instead of it just being this narration style,
00:16:38.360 | you can just do an interview style podcast.
00:16:40.940 | And a few more features,
00:16:42.260 | but basically there's a lot of like tooling
00:16:43.640 | that just makes this a very usable product for podcasts.
00:16:47.140 | - Yeah.
00:16:47.980 | And is it, for stitching these things,
00:16:49.940 | I've actually often thought about making my own,
00:16:52.760 | is it just FFmpeg and then like some flags
00:16:54.540 | and just stitch stuff together?
00:16:56.700 | - Yeah, I mean, obviously as a startup,
00:16:57.940 | you kind of want to just like do something quickly.
00:16:59.900 | So we're not like reinventing the wheel.
00:17:01.540 | - Yeah, like YouTube uses FFmpeg, you know?
00:17:03.700 | Like it's, FFmpeg is a workhorse of anything.
00:17:06.840 | - Yeah.
00:17:07.780 | - Multimedia.
00:17:08.620 | - Yeah, yeah.
00:17:09.660 | - Cool.
00:17:10.500 | You list some use cases that I just want to,
00:17:12.380 | you know, still stay high level.
00:17:15.060 | Businesses, newsletter.
00:17:16.500 | So businesses like Trader Joe's,
00:17:18.620 | and you seem to have done a big push there.
00:17:21.120 | - Well, so we didn't do this for them, right?
00:17:23.660 | It was just kind of a case study,
00:17:24.620 | but it's super interesting when,
00:17:26.220 | apparently Trader Joe's claims
00:17:28.180 | that one of their best marketing techniques
00:17:29.660 | was their podcast.
00:17:30.620 | - Yeah.
00:17:31.460 | - Right, so first of all,
00:17:32.340 | if you're just thinking about this as a use case,
00:17:33.780 | podcasts are really useful.
00:17:35.160 | - Yeah.
00:17:36.000 | - I think your follow-up question,
00:17:37.080 | which is like, you know,
00:17:39.120 | maybe I'm getting ahead of myself here,
00:17:41.520 | but you know, okay, companies will do podcasts,
00:17:45.320 | probably 99% of the time they'll be shit.
00:17:47.160 | - Yes.
00:17:48.000 | - So what's the trick?
00:17:49.240 | - Yeah.
00:17:50.060 | - Well, Trader Joe's did a super interesting one.
00:17:51.120 | - Yeah.
00:17:51.960 | - They didn't do this generic corporate speak of like,
00:17:53.720 | oh, Trader Joe's is so great,
00:17:54.940 | let's talk about apples or fruit or whatever.
00:17:57.040 | No, they actually just went into like,
00:17:59.560 | you know, the mechanics of the super interesting,
00:18:01.360 | niche stuff about their industry.
00:18:02.920 | And that's something that gets people listening, right?
00:18:04.680 | You want, I find it super interesting,
00:18:06.320 | just from a curiosity perspective.
00:18:08.120 | So yeah, like that's an example, right?
00:18:10.920 | You can, podcasts are actually really useful
00:18:12.660 | to just like get people, you know,
00:18:14.800 | it helps you with brand building, frankly.
00:18:16.720 | - Yeah, and Trader Joe's I think went really hard on it.
00:18:19.960 | I think they got some of the,
00:18:21.320 | quite the senior management involved.
00:18:22.640 | - Yeah.
00:18:23.480 | - They told some stories behind
00:18:25.160 | some very beloved food selections.
00:18:27.760 | I listened to the podcast, so it was pretty fun.
00:18:30.040 | And they also promoted it on their receipts.
00:18:32.520 | - Yeah.
00:18:33.360 | - So you guys featured that on your blog post,
00:18:34.920 | and I thought it was pretty cool.
00:18:36.080 | So I think it's well done by them,
00:18:37.640 | but also clearly not done by AI yet.
00:18:40.320 | - Yeah.
00:18:41.160 | - But still, I think that the use case is there.
00:18:43.960 | I'll just kind of go through the list.
00:18:45.120 | Newsletter creators, you said you have
00:18:46.920 | a hundred creators publishing with you?
00:18:48.720 | - Yeah, so, you know, the interesting thing is
00:18:50.680 | if you write a newsletter, I mean, I don't know,
00:18:54.120 | my email's flooded with newsletters at the moment.
00:18:56.280 | Sometimes I just want like the recap of it.
00:18:58.000 | Again, audio form is just, for some people, easier.
00:19:01.240 | If you're on, you know, commuting or whatever,
00:19:03.200 | you can just listen to it.
00:19:05.200 | So a lot of folks actually just convert their newsletter.
00:19:07.120 | It takes like two minutes.
00:19:08.120 | - Yeah.
00:19:08.960 | - Put the text in there, voila, you have an audio version
00:19:11.400 | of your newsletter that you just published to Spotify.
00:19:12.960 | - Yeah, yeah.
00:19:13.800 | And I am a newsletter writer, and I clicked around
00:19:17.680 | and wanted to basically just chuck my RSS feed in there.
00:19:21.280 | - Yeah.
00:19:22.120 | - And I think I gave that feedback exactly to you guys
00:19:23.920 | for like four months ago, or three months ago,
00:19:26.720 | and it looks like you've already shipped it.
00:19:28.480 | - Yep, well, I'm announcing it basically here today,
00:19:30.860 | which is we, part of generally what we're building
00:19:35.860 | is the ability to just, you know,
00:19:38.020 | we want to make this useful for creators of all kinds.
00:19:40.600 | So obviously, if we're, you know,
00:19:41.760 | we're helping you translate your content
00:19:43.040 | from one form to the other,
00:19:44.240 | I want to make this as seamless as possible.
00:19:45.320 | So as of today, we've actually built a Zapier integration,
00:19:48.740 | and we have a bunch of blogs on our website
00:19:50.600 | to kind of show you how to do this.
00:19:52.080 | But what you can now do is as soon as you publish
00:19:55.280 | a newsletter, it goes on your RSS feed,
00:19:57.800 | it will pick up the newsletter from your RSS feed
00:20:00.220 | automatically, and just publish an episode for you.
00:20:02.900 | - Yeah, question, what if I change something
00:20:05.180 | after I publish?
00:20:06.060 | - So you don't have to publish.
00:20:07.220 | It'll basically just generate, do all the work for you,
00:20:09.980 | and then you can go in and kind of modify it a little bit,
00:20:11.900 | and then publish.
00:20:12.740 | - Makes sense.
00:20:13.560 | - We also have scheduled publishing so that you can,
00:20:15.220 | I don't know, maybe you want to release it
00:20:16.980 | a few hours later.
00:20:17.820 | - Yeah, the professional podcasters that I've spoken to
00:20:21.020 | say that that is very important.
00:20:23.080 | I personally don't care.
00:20:24.260 | Like, it shows up in my feed or not.
00:20:25.900 | I don't care when it drops.
00:20:27.500 | Anyway, so you do want to basically time it,
00:20:30.240 | like if you're basically targeting a commute
00:20:33.120 | for the US time zone, you want to be like,
00:20:35.560 | oh, 8 a.m. Pacific for people driving into work.
00:20:39.620 | Then you show up at the top of the reverse chronological
00:20:42.340 | feed.
00:20:43.600 | I feel like that's too much tactics.
00:20:45.820 | - You know, and that's a good point.
00:20:47.240 | I think it depends a little bit on your audience
00:20:48.640 | and what you're building, but I do think,
00:20:50.680 | so I don't want to undermine the importance
00:20:53.240 | of consistency in podcasting.
00:20:56.760 | Whether that consistency literally translates
00:21:00.020 | into I publish at 8 a.m. every single day,
00:21:03.880 | or I just publish every single day,
00:21:05.520 | or whatever, there is a huge importance
00:21:08.080 | in just making sure that what you're publishing
00:21:10.160 | is always, it's there.
00:21:11.680 | People need to know that your brand
00:21:12.840 | is constantly pushing stuff.
00:21:14.580 | - Yeah, so a lot of people who talk to me
00:21:16.720 | are interested in what's my advice on content creation.
00:21:19.520 | Yeah, at least once a week.
00:21:20.800 | Whatever you do, and I don't care when you do it.
00:21:22.800 | Just do it once a week, put something out.
00:21:24.880 | But I do notice that, specifically in the podcasting field,
00:21:28.160 | and you talk about this in the next point,
00:21:30.480 | daily podcasting is the meta game
00:21:32.760 | that is, I think, doing extremely well.
00:21:34.520 | Especially because I think the Apple podcast list
00:21:36.880 | biases for daily.
00:21:38.960 | Because, obviously, the downloads will be higher.
00:21:41.560 | So daily podcasters kind of rank higher more,
00:21:44.480 | and obviously, because you're daily,
00:21:45.920 | you also do shorter podcasts, which guarantees
00:21:48.080 | that more people listen to you.
00:21:50.440 | - Yeah, yeah, I think the fact that,
00:21:52.400 | so obviously we do the Hacker News recap.
00:21:54.680 | The fact that we did that, and that it is daily,
00:21:57.280 | actually just helped us reach
00:21:58.640 | that top 30 type of podcast on Spotify.
00:22:01.600 | - Yeah, that was mostly 'cause you were on HN, right?
00:22:03.760 | - We did launch, but obviously the fact
00:22:05.280 | that you just publish a lot of content,
00:22:07.080 | you're just gonna get a lot more,
00:22:08.080 | like it's a statistics thing, right?
00:22:09.520 | Obviously, I think they do it
00:22:10.360 | by total time listened as well.
00:22:12.000 | - Yeah, yeah.
00:22:12.840 | - But the fact that it's daily, it's just not overwhelming.
00:22:15.240 | Again, we don't have that much of an attention span anymore.
00:22:18.400 | - Yeah, yeah, that's true, that's true.
00:22:19.760 | Yeah, I love it, I listen to it every day.
00:22:22.720 | Okay, so internal podcasts for daily memos.
00:22:26.080 | I think Spotify might have internal podcasts.
00:22:29.080 | I don't know of any other famous company
00:22:30.480 | that has internal podcasts, but this is interesting.
00:22:33.180 | - It's interesting, 'cause I think here we're basically
00:22:36.200 | just like, again, we're overloading the term podcast.
00:22:39.440 | What is a podcast, right?
00:22:40.720 | It's like, I think the origin of the word
00:22:42.520 | is like iPod broadcast.
00:22:43.640 | - Yes.
00:22:44.480 | - Right?
00:22:45.840 | I don't know what that means, right?
00:22:47.040 | It's just like, again, it's on-demand radio
00:22:48.720 | if we just take that word, but really the idea is like,
00:22:51.640 | again, if we wanna enable access to content,
00:22:54.280 | and I think part of that is just like
00:22:56.760 | efficient information delivery.
00:22:59.160 | So communication in a company is like,
00:23:01.880 | we all know the perils of how hard it is sometimes
00:23:04.960 | if you're a big company to communicate,
00:23:06.760 | and I think that kind of stuff is super interesting.
00:23:08.560 | I think one particular use case,
00:23:09.800 | which is like the internal,
00:23:11.400 | you use that just to share information quickly,
00:23:14.640 | and I don't know if you have the next one kind of on there,
00:23:17.000 | or maybe I'm--
00:23:17.840 | - It's okay, we can jump around.
00:23:19.160 | - There's like a slightly different twist to it,
00:23:22.520 | which I think we can enable with the Zapier integration
00:23:24.720 | that we've built, which is this, I don't know,
00:23:28.120 | I imagine sometimes, for example, I'm a PM at a tech firm,
00:23:32.640 | a lot of things is going on on my team.
00:23:34.920 | I want to just do a quick recap, right?
00:23:36.880 | I don't know, some people will do that.
00:23:39.560 | If you're really good at keeping state
00:23:41.480 | on some of the different SaaS products that exist,
00:23:43.440 | yeah, good, you can just get a recap there.
00:23:45.480 | I think something that might be interesting is like,
00:23:47.280 | can we look at your GitHub issues, PRs, whatever,
00:23:51.640 | and whenever there's a status change,
00:23:52.800 | kind of bake that into something
00:23:54.880 | that you then consume as an audio?
00:23:56.520 | - Yeah.
00:23:57.520 | - We have like, you can do something similar like that
00:23:59.600 | with Zapier, it's actually super interesting
00:24:01.120 | because it just saves you time.
00:24:03.520 | You just get on your commute to work,
00:24:05.080 | be like, these are the things that happened today.
00:24:06.720 | This PR was closed, we added this feature,
00:24:08.560 | blah, blah, blah, blah, blah, right?
00:24:10.360 | And similarly, you can think of something like that
00:24:13.520 | for email, obviously the privacy implications there
00:24:15.280 | are just a little bit more work,
00:24:16.560 | but like, there's actually all of a sudden like,
00:24:19.000 | yeah, cool, I can actually consume a lot of the content
00:24:21.600 | that I historically have never been able to listen to
00:24:24.200 | now by audio.
00:24:25.080 | - Yeah.
00:24:26.080 | Yeah, it opens up different forms of learning,
00:24:29.400 | as you mentioned.
00:24:30.680 | And then finally, I'm skipping ahead a little bit.
00:24:33.840 | Podcast studios are using you.
00:24:35.280 | So you said you onboarded 10 studios.
00:24:37.920 | What do studios use you for?
00:24:39.960 | - A couple of interesting things, actually.
00:24:41.080 | So studios, like, they, podcasting is like,
00:24:46.400 | book publishers in the sense that there's a million genres
00:24:48.400 | at this point.
00:24:49.240 | - Yeah.
00:24:50.840 | - When we think of podcasts, I think we typically think
00:24:52.720 | of like, the interview style one.
00:24:55.040 | So they're not necessarily using it for that,
00:24:56.360 | but it's a lot more for like, two things generally.
00:25:00.920 | One is actually just like, producing the content out there
00:25:03.480 | that is more narration style format.
00:25:05.040 | So you can think of something like a kid's podcast
00:25:07.000 | that teaches them about, you know, I don't know,
00:25:09.560 | countries or whatever it is, or you know,
00:25:11.800 | basic science topics.
00:25:13.920 | We do that just completely through us, right?
00:25:17.160 | Obviously we produce the sound, we produce the music.
00:25:18.920 | If they need a little bit more editing,
00:25:20.040 | they can just take that to their editors and do it.
00:25:23.120 | The other thing that's really interesting is it actually
00:25:25.080 | lets them also build out a demo spec
00:25:29.040 | of what a more involved podcast can look like.
00:25:32.200 | - I see.
00:25:33.040 | - Right, so, you know, you were coming up with the idea
00:25:35.400 | for this podcast, you kind of want to show people
00:25:37.480 | or some of your guests what it's going to sound like,
00:25:39.360 | and whip something up.
00:25:40.760 | And I think for them, it's obviously really important
00:25:42.520 | because the way they operate, at least,
00:25:44.080 | is like there's, you know, people are pushing
00:25:45.680 | for a different podcast.
00:25:47.000 | And then you get sign-off to pursue it.
00:25:49.200 | So you need to kind of show just like,
00:25:50.320 | what's this going to sound like?
00:25:51.440 | - Yeah, excellent.
00:25:52.280 | Awesome.
00:25:54.200 | I think that's a really good overview.
00:25:56.480 | Then you also produce three in-house podcasts.
00:25:59.960 | - Yeah.
00:26:01.080 | - Hacker News Recap, Product Hunt Daily, and PGSA.
00:26:04.120 | - So we dropped the Product Hunt Daily.
00:26:05.720 | - Oh, okay, it's still on your landing page.
00:26:08.120 | - Yeah, we launched it, but we like,
00:26:09.520 | we don't push anything new there.
00:26:11.560 | - No interest in Product Hunts?
00:26:12.840 | - We didn't see too much traction.
00:26:13.760 | I think Product Hunts--
00:26:14.600 | - RAP Product Hunts, Alec.
00:26:15.760 | (laughs)
00:26:17.680 | No, so yeah, tell me--
00:26:19.280 | - So we do the Hacker News Recap and the PGSA,
00:26:23.000 | I think are the two most popular ones.
00:26:24.600 | - Yeah.
00:26:25.600 | - We're constantly experimenting with new internal,
00:26:27.240 | or like podcasts that we publish.
00:26:28.800 | - Yeah, yeah.
00:26:29.640 | What are your other, you can tease a little bit.
00:26:32.160 | What are you thinking about?
00:26:33.000 | - Tease a little bit?
00:26:33.840 | Well, I really like Reddit.
00:26:35.720 | - Yeah.
00:26:36.560 | - I'd love to listen to some of the Reddit things
00:26:38.640 | going on there, but instead of like reading them.
00:26:40.760 | - Yeah.
00:26:41.600 | - It's always just a notification that I get,
00:26:42.440 | I'm like, "Ooh, this sounds interesting."
00:26:43.560 | But I don't know, you can do it like per subreddit
00:26:45.560 | that you care about.
00:26:46.480 | - Yeah.
00:26:47.320 | - A few things like that.
00:26:48.160 | I love Life Pro Tips.
00:26:48.980 | - Yeah, I see.
00:26:50.400 | - Like super interesting things, or Wall Street Bets,
00:26:52.480 | or whatever you're into.
00:26:53.720 | - Yeah, well, the problem with these things
00:26:56.240 | is that a lot of them could involve images and memes,
00:26:59.400 | which you cannot consume.
00:27:01.600 | - Well, yes, we cannot consume.
00:27:03.640 | This is like a simple, we can't consume that at the moment,
00:27:06.600 | but maybe a few weeks down the line
00:27:11.160 | when that video feature of ours gets a little better,
00:27:13.200 | you can actually start shipping it like that.
00:27:14.720 | - Okay.
00:27:15.560 | - Anyway.
00:27:16.400 | - Yeah, yeah, yeah.
00:27:17.220 | And so, and I'll just feed you an idea.
00:27:19.440 | To keep up on AI, a lot of stuff actually happens
00:27:21.560 | in Discord, and there's way too many Discords.
00:27:24.080 | - Way too many, and they're way too active.
00:27:25.800 | - Yes, so I've actually built a little feed for myself
00:27:29.400 | that scrapes a bunch of Discords.
00:27:30.880 | - Yeah.
00:27:31.720 | - And creates a daily newsletter for myself.
00:27:33.120 | - Amazing.
00:27:33.960 | - I have thought about turning it into an audio feed,
00:27:37.160 | but, and this is the problem for Wondercraft,
00:27:40.880 | I read better, I read faster.
00:27:42.160 | I scan up and down faster than I listen, right?
00:27:44.880 | And there's just too much noise in Discord
00:27:47.760 | for me to listen as audio format.
00:27:49.640 | Your Hacker News stuff is very high signal
00:27:52.000 | because, obviously, you're folding, right?
00:27:53.880 | - 'Cause we haven't done the curation like Hacker News did.
00:27:56.280 | - Exactly, that's why it's guaranteed to be good,
00:27:58.920 | whereas for Discord, it's a bunch of junk.
00:28:02.820 | - Yeah. (laughs)
00:28:04.420 | But I do think there's something similar,
00:28:05.700 | like Reddit also does the curation.
00:28:08.460 | It's not us who's doing it, right?
00:28:09.940 | - Yes, yes.
00:28:10.780 | - If you're only the hot posts.
00:28:11.660 | - Still a little bit noisier, so I don't know if you know,
00:28:14.300 | I was a moderator of the React to Reddit for four years.
00:28:17.420 | - Nice.
00:28:18.620 | - So I've seen a bunch of stuff,
00:28:20.460 | and I know it's noisier than a Hacker News,
00:28:23.100 | but still pretty good.
00:28:24.580 | - Yeah, yeah.
00:28:25.820 | So we do Hacker News, we do PJ Essays.
00:28:27.340 | I think PJ Essays are also super interesting.
00:28:28.860 | I listen to them all the time 'cause I,
00:28:30.380 | well, first of all, I actually think
00:28:31.420 | they're pretty well produced.
00:28:32.580 | We do a good job, like, I don't know,
00:28:34.060 | if we're quoting someone, we'll use a different voice.
00:28:36.540 | - Yeah, yeah, yeah.
00:28:38.020 | - I think it's just well produced.
00:28:39.660 | And I also think, you know, the essays are so seminal
00:28:42.700 | to like, everyone in startups reads them.
00:28:45.180 | - Yeah, it's actually got me to read more PJ Essays
00:28:48.380 | than I would have otherwise, so mission accomplished.
00:28:50.860 | - I don't know if it was the last one,
00:28:52.300 | well, the time we published this.
00:28:53.140 | - No, How to Do Great Work.
00:28:53.980 | - How to Do Great, that was a one-hour podcast.
00:28:56.140 | - Yeah, oh my God.
00:28:57.100 | - No one, I could, I did not read it,
00:28:58.780 | I just had to listen to it.
00:28:59.620 | - Yeah, yeah, yeah.
00:29:00.460 | - And I think actually, if I'm being honest,
00:29:01.940 | I think the motivation for PJ Essays
00:29:03.300 | was just like, I need this.
00:29:04.900 | - Yeah, yeah, yeah.
00:29:06.700 | For that one, if it's like one hour,
00:29:08.540 | I would have actually appreciated like a segmentation,
00:29:12.860 | or like, hey, high level, you know,
00:29:14.940 | I know this is about to be an hour,
00:29:16.500 | but like, there are three main high level things,
00:29:18.380 | and then keep that in your mind,
00:29:19.540 | and then go like, part one, blah, blah, blah.
00:29:21.760 | Part two, blah, blah, blah.
00:29:22.600 | - I think we, so we do that to some extent,
00:29:24.220 | and like, we produce like, chapters, I guess.
00:29:26.700 | - Yeah.
00:29:27.540 | - So you can just look at them.
00:29:29.300 | Probably could do a better job like, introducing it,
00:29:31.220 | but we do try to like, not play around with the PJ Essays.
00:29:33.980 | - For sure, yeah, I mean, it's,
00:29:35.620 | you know how much work he puts into those things.
00:29:37.060 | - So we just kind of ship it as is.
00:29:38.780 | - Yeah, yeah, yeah.
00:29:39.840 | Actually, I'll tell you about one more that,
00:29:41.140 | one more daily, not daily,
00:29:43.140 | but frequent AI-generated podcast
00:29:44.900 | that I listen to, apart from you guys,
00:29:46.060 | which is PapersRed.ai,
00:29:47.860 | and I'll recommend it to anyone listening as well.
00:29:49.640 | - Super interesting, actually.
00:29:50.480 | I've come across it. - You've come across that.
00:29:51.940 | It's by this guy, Rob, and I've tried to look him down.
00:29:54.900 | He doesn't want to be found.
00:29:56.860 | Anyway, but the selections are very good.
00:29:58.560 | I think you guys could do a better job than him.
00:30:00.300 | - Yeah.
00:30:01.140 | - It's-
00:30:02.100 | - A job, Rob.
00:30:03.700 | - Well, as you can see, it converts PDFs to podcasts, right?
00:30:06.420 | And the problem with academic PDFs is a lot of references.
00:30:10.420 | You know, like, "Buy et al. 2022,"
00:30:12.340 | and then like, headers, and then a table,
00:30:14.260 | and then you know, like, reads the table
00:30:15.880 | when you don't need to read the table, you know?
00:30:18.200 | That kind of stuff.
00:30:19.040 | I think better engineering there from you guys
00:30:21.660 | would beat him, and I need that, so feature request.
00:30:26.460 | (laughs)
00:30:27.300 | - Working on it, working on it.
00:30:29.040 | - Okay, the final feature-related thing
00:30:31.020 | is the thing that we're announcing today as we release it.
00:30:33.660 | - Which we're super, super excited about,
00:30:35.300 | but Wondercraft now does video translation.
00:30:37.780 | - Okay. - And dubbing.
00:30:38.620 | - Okay, why do people want that?
00:30:41.060 | - Again, let's go back to our mission.
00:30:42.260 | We're trying to expand access to content.
00:30:44.580 | I think, don't quote me on this again,
00:30:46.980 | like, I don't know who actually knows the internet,
00:30:48.480 | but like, 60% of the internet is in English.
00:30:51.340 | You know, what if you don't speak English?
00:30:52.620 | You're automatically disbarred or kind of excluded
00:30:55.660 | from all the content that's produced.
00:30:58.140 | And thanks to all the advances
00:31:00.820 | that have just recently been made,
00:31:01.940 | we can actually make this,
00:31:03.140 | it's super easy to dub this in other languages.
00:31:05.540 | So we're super happy to announce this feature.
00:31:09.840 | We're super excited, we've been working on it
00:31:11.140 | for a really long time.
00:31:12.780 | But now basically, everyone, go on our platform,
00:31:15.340 | upload your podcast episode, and see the dub for yourself.
00:31:18.220 | We'll use your voices, we'll completely convert it,
00:31:20.820 | make sure it's aligned with the,
00:31:22.180 | if you have video, we'll make sure it's aligned.
00:31:23.660 | - Yeah.
00:31:24.700 | - And voila, just publish it.
00:31:25.940 | - And specifically for video,
00:31:27.740 | so like, obviously the thing that's going around
00:31:30.660 | is the hey, Jen thing, which changes your lips.
00:31:33.980 | - So we don't do lip sync at the moment.
00:31:35.820 | Could be another feature that we work on.
00:31:37.260 | - Yeah, because you're primarily podcasting,
00:31:39.040 | which is no video.
00:31:40.100 | - Well, primarily no video,
00:31:41.380 | but I think we still basically,
00:31:42.580 | if you do have a video, we'll still align it
00:31:44.660 | to the chunks that--
00:31:45.500 | - You still align it.
00:31:46.340 | Okay, so the hard problem is the aligning.
00:31:47.740 | - The alignment is the difficult bit, you're right.
00:31:50.420 | The actual, like--
00:31:51.260 | - It's with all things in AI.
00:31:52.980 | - Yeah, again, overloading the word alignment.
00:31:55.860 | - Um, we don't do the lip sync.
00:31:57.980 | - Yeah.
00:31:58.820 | - Yet.
00:31:59.660 | - It's kind of a gimmick, yeah.
00:32:00.480 | - It's like, it's not super necessary.
00:32:01.940 | If you're really just like listening to a podcast
00:32:03.380 | and you actively want to listen to it
00:32:04.540 | in a different language,
00:32:05.600 | then you're more interested in the content.
00:32:06.440 | - So what are you aligning to?
00:32:07.900 | - Basically the chunks where the speakers are speaking.
00:32:11.420 | So you won't have an instance
00:32:12.820 | where you'll have me as the dub speaking
00:32:17.180 | while the camera is on you.
00:32:19.320 | - Ah.
00:32:20.160 | - Right, so it's basically just like,
00:32:21.060 | the speaker turns around the audio.
00:32:21.900 | - So you will like, if it happens to be
00:32:23.580 | a little bit longer, you'll speed it up a little bit.
00:32:25.600 | - We do a little bit of, you know, trickery there.
00:32:27.940 | - Yeah.
00:32:28.780 | - But we get it aligned so that when you're speaking,
00:32:29.940 | it's you, and when I'm speaking, it's me.
00:32:31.680 | - Okay, cool.
00:32:32.860 | And then, one more thing before we go into like,
00:32:34.860 | the building Wondercraft side,
00:32:36.220 | building a company and being a founder, all that stuff.
00:32:38.820 | You said, you want to ride this wave.
00:32:43.820 | You said this dubbing thing was very important to you.
00:32:47.380 | And I think this is a core stress of a lot of AI founders,
00:32:52.380 | which is, how do I keep up?
00:32:54.180 | How do I pick what trend to ride
00:32:56.940 | and be like, very, very quick on?
00:32:59.420 | And what do I ignore?
00:33:00.900 | So just bring me through your thought process
00:33:02.380 | on like, why this is the one to really pressure
00:33:05.420 | and build quickly.
00:33:06.980 | - It's a really good question,
00:33:10.580 | 'cause I don't think it has a,
00:33:12.100 | I think you'll get 19 different answers.
00:33:15.100 | I'll ramble a little bit.
00:33:16.220 | - Yeah, chain of thought.
00:33:17.700 | - Dubbing, you know,
00:33:22.600 | it's something that I want.
00:33:26.460 | There is content that just exists that I can't consume
00:33:29.860 | just because it's not there.
00:33:32.060 | I'm lucky enough to speak, you know, three, four languages,
00:33:35.780 | so I can listen to some subset of podcasts
00:33:38.100 | that exists all around the world.
00:33:39.560 | Let's not even say podcasts, let's just say content.
00:33:42.060 | But the rest of it is kind of excluded for me.
00:33:45.340 | I have no context on what's going for it.
00:33:47.420 | And I think that problem is just multiplied a million times
00:33:51.300 | when you consider that if you don't speak English,
00:33:55.020 | you really are excluded from a lot of things.
00:33:57.780 | So the ability, it's just motivating to be able
00:34:00.100 | to actually give access to content
00:34:03.900 | that historically would be completely obscured
00:34:06.780 | for most people.
00:34:08.220 | I think your question though has like a,
00:34:10.300 | and the dubbing market we think is huge.
00:34:11.720 | I think like, you know, it's easy enough to dub these days,
00:34:15.700 | especially with our product.
00:34:16.780 | So like, it's doable.
00:34:17.900 | I think, you know, people are gonna,
00:34:19.780 | it basically was like a priority that we were tracking
00:34:21.740 | to push this out.
00:34:22.700 | We got a lot of requests for it, and so on.
00:34:24.900 | That's like the tactical answer on this dubbing thing.
00:34:28.700 | Obviously, you know, in general every week we sit down,
00:34:30.700 | we prioritize all the different features,
00:34:32.660 | and then we build.
00:34:33.800 | I think your question probably has a slightly higher level.
00:34:38.040 | - Yes, the meta strategy of--
00:34:40.500 | - The meta strategy is like, how do you just pick things?
00:34:42.560 | - Yes.
00:34:43.400 | - And do them, and what do you say yes to
00:34:45.180 | and what do you say no to?
00:34:46.440 | And frankly, it's a tough one,
00:34:49.620 | 'cause you don't know.
00:34:51.580 | You don't know what to pick.
00:34:53.260 | Obviously, I can tell you things like,
00:34:54.620 | yeah, look at metrics, see.
00:34:56.140 | There's simple metrics.
00:34:56.980 | - There's no metrics for this.
00:34:57.900 | - There are simple metrics in the sense like,
00:34:59.180 | how many times are customers asking me for this?
00:35:00.900 | - Sure.
00:35:01.740 | - And then you can go one level deeper,
00:35:03.340 | which is like, how much money am I gonna make
00:35:06.660 | from the customers who are asking me for this?
00:35:08.440 | So you can start to quantify from a revenue perspective.
00:35:11.620 | - In that case, was there something for that?
00:35:13.180 | - There's a lot of requests,
00:35:14.220 | and obviously that's part of the product.
00:35:16.460 | So the second you dub a podcast, we're making money.
00:35:20.580 | - Okay.
00:35:21.420 | - The thing that's slightly tougher is
00:35:25.660 | there is no better answer that I can give
00:35:30.620 | than try to do as much as you can.
00:35:33.140 | I think we pride ourselves in pushing at least
00:35:38.460 | two super highly demanded features
00:35:41.340 | from our customers a week.
00:35:42.620 | - Two, wow.
00:35:44.660 | - Yeah, so we just move super fast.
00:35:46.500 | And part of that is 'cause we're frankly just a small team.
00:35:49.320 | We're pretty like--
00:35:50.160 | - How big is your team?
00:35:50.980 | - Four people.
00:35:51.820 | - Okay.
00:35:52.660 | - Total, the whole company is four people.
00:35:53.500 | - Yeah.
00:35:54.780 | - We just move very independently and very quickly.
00:35:57.160 | So we're just able to push out a lot of things.
00:35:59.740 | But I think that's frankly,
00:36:01.740 | that's just the requirement of what we're doing.
00:36:03.500 | It's not meant to be a chill game.
00:36:05.300 | - Yeah.
00:36:06.140 | - And this is kind of a,
00:36:07.760 | don't wanna use that term, but for lack of a better word,
00:36:10.820 | like it's a race.
00:36:12.220 | Like you have to keep building
00:36:13.620 | or else someone else is gonna build.
00:36:15.060 | - It's almost, but it's like a race without an end.
00:36:17.020 | - It's a race without an end.
00:36:18.060 | So I guess all I'm trying to say is like,
00:36:20.160 | yeah, it's tough,
00:36:22.340 | but you kind of just like have to be prolific.
00:36:24.420 | - Okay, fair enough.
00:36:26.180 | Yeah, you mentioned that you were listening
00:36:27.660 | to the "Line Chain" episode.
00:36:29.140 | And "Line Chain" is also similarly like,
00:36:31.500 | in my mind, very high tension,
00:36:32.860 | but also obviously important for them
00:36:34.640 | to stay relevant on top of all the trends.
00:36:36.340 | - Yeah.
00:36:37.180 | - I just, it's very stressful.
00:36:41.540 | - But it's overwhelming, I agree.
00:36:42.380 | - Yeah, it's overwhelming.
00:36:43.220 | - Every day, my to-do list grows by like 10 times
00:36:46.140 | and I do less, like, you know what I mean?
00:36:48.300 | - Append only, yeah.
00:36:49.320 | - It's just, basically it's append only.
00:36:50.820 | So it's tough, but the reality is you just like,
00:36:53.960 | as a, I'm not even saying as an AI startup,
00:36:57.420 | as a startup, you just have to be prolific.
00:37:00.060 | - Okay, yeah, of course.
00:37:01.620 | - Yeah.
00:37:03.020 | - Do you think that you have,
00:37:06.420 | so often people will say they will build, build, build
00:37:10.260 | until they find quote unquote PMF.
00:37:12.560 | Can we talk about what you think PMF is?
00:37:14.760 | And then, like, those core features, right?
00:37:16.940 | Like, because there's such a thing
00:37:17.980 | as also a feature factory.
00:37:19.180 | I have worked at startups that were feature factory.
00:37:21.700 | Build, build, build, and then it's a whole bunch of junk
00:37:23.740 | that nobody, that doesn't have a coherent story.
00:37:25.980 | - Yeah, I think that's a really tough one, actually,
00:37:27.900 | to just like build, make sure that the things
00:37:29.480 | that you're building all actually like,
00:37:31.420 | from a very business perspective,
00:37:33.820 | either an upsell or they're somehow related
00:37:35.820 | to what you're building.
00:37:36.660 | 'Cause otherwise, it's really difficult to,
00:37:39.180 | I think, okay, let's just start from the baseline
00:37:41.340 | that you build a product for a customer.
00:37:44.600 | - Yeah.
00:37:45.520 | - It's okay to then build another product
00:37:48.140 | that caters to the same customer,
00:37:49.860 | or sell the same product to a different customer.
00:37:52.140 | - Yeah.
00:37:52.980 | - What's really difficult is to build a different product
00:37:54.140 | for a different customer,
00:37:54.980 | 'cause that's just a new company.
00:37:55.860 | - Yeah, yeah, yeah.
00:37:56.740 | - Right, so--
00:37:57.620 | - Have you seen the superhuman PMF score thing?
00:38:00.100 | - I have read that blog post ages ago,
00:38:02.720 | so I don't remember it.
00:38:03.560 | I think he said like, you know,
00:38:04.380 | 40% of people recommend it.
00:38:06.220 | - Yeah.
00:38:07.740 | - So, I don't know, 'cause--
00:38:08.580 | - His advice is like spend 60% of your time
00:38:10.180 | making your existing customers more happy.
00:38:11.740 | - Yeah.
00:38:12.580 | - And then 40% the adjacent guys,
00:38:13.680 | and then the remainder for R&D for future.
00:38:17.540 | - Yeah, yeah, I think that's pretty good advice,
00:38:19.420 | to be honest, 'cause I do think like,
00:38:21.460 | never neglect your current customers.
00:38:23.220 | - Of course.
00:38:24.060 | - Right, 'cause they're gonna be your biggest promoters.
00:38:25.100 | - Totally, yeah.
00:38:25.940 | - So, we definitely wanna make them feel loved.
00:38:27.220 | We, again, try to really, like,
00:38:29.100 | I think one thing we also have going for us
00:38:30.700 | is our customer support.
00:38:32.180 | - Yeah.
00:38:33.000 | - Which is a one-person show at this point.
00:38:34.900 | But it's really good, it's really responsive,
00:38:36.580 | and we really try to make sure that the customer
00:38:38.220 | is like, knows what to do, and if it's something we,
00:38:41.220 | it's also just really rewarding to, I think,
00:38:42.620 | be like, at such a small team,
00:38:44.620 | and this is a bit of a tangent,
00:38:45.740 | but like, you get an email from a customer being like,
00:38:48.620 | "Hey, this doesn't work."
00:38:49.660 | And I think people are typically primed
00:38:51.740 | to like, be a little bit aggressive with companies.
00:38:54.740 | Right, like, "Hey, this doesn't work, I need this."
00:38:57.500 | And then they're super surprised when within an hour,
00:38:59.620 | you're like, "Hey, we pushed a fix, this should work now,
00:39:01.660 | "I hope all's good."
00:39:02.500 | And they're like, "Oh my God, thank you."
00:39:03.700 | - Yeah, yeah, yeah.
00:39:04.780 | Yeah, that's the magic that's only possible
00:39:06.380 | at an early stage.
00:39:07.220 | - Yeah. - Yeah, that's great.
00:39:08.860 | Okay, since you mentioned pricing,
00:39:10.420 | maybe just a few minutes or two
00:39:12.620 | on just how you think about pricing.
00:39:14.700 | You are probably a lot of usage-based stuff,
00:39:19.580 | and I always have this struggle with usage-based,
00:39:23.580 | because you're basically charging your heaviest users more,
00:39:25.900 | and discouraging them from using you more.
00:39:27.980 | - I think of it slightly differently.
00:39:29.940 | We were not, so at the moment, okay.
00:39:33.220 | - Your tiers.
00:39:34.060 | - Let's just promise by saying,
00:39:35.180 | we haven't figured out what the pricing is.
00:39:36.020 | - Sure. - I think that's like
00:39:37.180 | the eternal struggle. - Yeah, what's in your mind?
00:39:38.420 | - We don't know how to price this thing.
00:39:39.260 | - Yeah, what's in your mind? - So far.
00:39:40.660 | We don't actually, at the moment,
00:39:45.900 | charge you if you consume above your quota.
00:39:48.060 | So we have tiers.
00:39:49.580 | - Ooh, hot take. (laughs)
00:39:51.180 | - As in, if your quota, you just have to upgrade
00:39:54.260 | to a different one, right? - Oh, I see, I see.
00:39:56.540 | - We don't have this additional consumption.
00:39:59.300 | - Meter billing, yeah.
00:40:00.340 | - Are we charging our biggest users more?
00:40:05.140 | - We actually just charge them per unit less.
00:40:07.780 | So it does work on, if you just produce more content
00:40:10.340 | than the unit, the unit, the per unit price is just lower.
00:40:14.780 | - Yeah.
00:40:15.620 | - I don't know, I have to say,
00:40:19.420 | pricing is probably pretty rudimentary at this point.
00:40:21.340 | - Yeah, it's not as important.
00:40:23.500 | - It is important, 'cause fundamentally it drives revenue,
00:40:25.540 | but there needs to be a bit of a science and an art,
00:40:29.780 | it's a bit science, a bit art,
00:40:30.740 | to figure out what the right pricing strategy is.
00:40:32.860 | I think you see like a million different things,
00:40:35.420 | and a million different things that could work for you,
00:40:36.860 | but you frankly just have to test it,
00:40:38.460 | and like, again, there's a million things to do,
00:40:42.340 | like four people to do them,
00:40:43.460 | so we haven't been moving as quickly as you want on that.
00:40:46.260 | I think we're probably gonna release something new soon
00:40:48.540 | in terms of pricing.
00:40:49.460 | - Yeah, cool.
00:40:51.500 | - But it's a tricky one.
00:40:52.540 | - Hey, nobody has it figured out.
00:40:54.140 | - No. (laughs)
00:40:55.900 | - I've been through many, many pricing changes in my career.
00:40:59.660 | Cool, I wanted to talk a little bit about the origin story,
00:41:02.380 | 'cause you flagged that Mooncraft was actually
00:41:05.260 | a big part of-- - Moonshot.
00:41:06.100 | - Moonshot was a big part of your arriving at this idea.
00:41:10.740 | - Yeah, let me give a little bit of background.
00:41:15.500 | So, Demi and I, we're both apparents here,
00:41:17.460 | we actually were flatmates right before we started this,
00:41:22.020 | and then we both had this idea that kind of came about
00:41:25.740 | from our respective upbringings,
00:41:27.460 | which was, there's a lot of people,
00:41:30.700 | and we're not the first to have this idea,
00:41:31.980 | by any stretch of imagination,
00:41:33.580 | but there's a lot of people out there in the world
00:41:35.740 | who are really good, really talented,
00:41:37.340 | who just don't have the financing to do the things
00:41:39.220 | that they're talented in, right?
00:41:40.580 | Like, you need to develop yourself a little bit.
00:41:43.500 | So, we were like, cool, how do we solve this?
00:41:45.500 | How do we get people funding to do the things
00:41:49.060 | that they're good at?
00:41:50.020 | We're like, cool, let's align incentives,
00:41:51.380 | let's get people investing in other people.
00:41:54.100 | Cool, that was the original idea.
00:41:55.460 | Applied to YC, we got in, long story short,
00:41:58.060 | we got in, and we're like, cool, let's see who's done this.
00:42:01.900 | We've obviously done that before,
00:42:02.940 | but let's really figure out
00:42:04.340 | what our unique edge is gonna be.
00:42:06.340 | So, we looked at all the other companies that had done that,
00:42:08.460 | and I think one particular one that comes to mind
00:42:10.500 | is Upstart, which is now a multi-billion dollar company
00:42:13.260 | listed, and we called them moonshots,
00:42:17.940 | that was kind of the idea,
00:42:18.780 | like people who have the moonshot potential,
00:42:20.740 | they called them upstarts,
00:42:22.180 | and we both had this idea that you were a backer.
00:42:24.740 | Backers back moonshots or upstarts.
00:42:27.020 | They spent two years on this,
00:42:28.380 | and then they pivoted to, I think, a lending platform,
00:42:31.220 | some lending platform that they do now.
00:42:33.180 | So, we were like, we spoke to a couple folks there,
00:42:37.020 | and we're like, so what didn't work?
00:42:38.380 | And the thing they said is there is,
00:42:40.740 | I'm also giving a lot of detail right now on some moonshots,
00:42:43.500 | so maybe that's not super relevant,
00:42:44.500 | but just added context.
00:42:46.220 | - It's fun to hear.
00:42:47.060 | - They're like, it's really hard to get your average Joe
00:42:49.700 | interested in backing your other average Joe, right?
00:42:52.040 | People like, it's not really interesting
00:42:54.060 | to back this other guy who wants to go
00:42:55.540 | get a better education somewhere.
00:42:57.300 | You don't, you know what I mean?
00:42:58.940 | There's no driver, 'cause A, the economics
00:43:00.500 | typically aren't so good,
00:43:01.780 | and historically, they just found that people don't do this.
00:43:06.880 | So, we were like, cool, if that's the problem,
00:43:08.980 | then let's focus instead on a group of talent
00:43:12.860 | that has this exponential style trajectory.
00:43:15.840 | So, we picked athletes.
00:43:18.320 | People love sports, and people really feel
00:43:20.740 | what we call the emotional dividend of this investment.
00:43:23.260 | So, we actually stopped calling it investments,
00:43:24.900 | 'cause the investment wasn't great.
00:43:26.660 | Financially speaking, your investment,
00:43:28.780 | you're better off putting it in the stock market
00:43:30.380 | or real estate or whatever,
00:43:31.660 | just the risk return did not make sense.
00:43:33.420 | - So, you might trigger some regulation.
00:43:35.800 | - We were fully triggering regulation.
00:43:37.620 | We were compliant with the SEC.
00:43:39.180 | It was all regulated.
00:43:40.460 | We weren't doing this NFT style.
00:43:42.180 | And then, so we were selling this.
00:43:46.780 | So, you get a financial dividend.
00:43:47.860 | You still get part of their payback,
00:43:49.940 | or if tennis player X makes that much, you get 10%,
00:43:53.940 | but we also started calling it the emotional dividend.
00:43:58.220 | So, actually, what we're selling a lot
00:43:59.580 | is also the experience.
00:44:00.460 | You get to be part of the front scene of this,
00:44:04.120 | of this person going on this journey.
00:44:05.340 | You're with them.
00:44:06.820 | We actually found a lot of people
00:44:07.660 | that kind of do this already.
00:44:09.100 | Now, what we found out after that
00:44:10.380 | is this wasn't a big enough market.
00:44:11.940 | So, long story short, we tried really hard.
00:44:14.340 | I frankly think, there's a comment here
00:44:16.400 | about founder market fit.
00:44:17.540 | We were not the right founders for this,
00:44:19.060 | and when we were raising money,
00:44:20.380 | VCs would ask us, what are two software engineers
00:44:23.860 | from Panther doing?
00:44:25.100 | Like, why are you funding athletes?
00:44:26.700 | What?
00:44:27.660 | And we'd sell this 'cause, you know,
00:44:28.820 | I played a little bit of tennis.
00:44:30.440 | They may actually play professional basketball,
00:44:32.140 | so we'd sell it like that.
00:44:33.460 | But, you know, for all the hate that VCs get,
00:44:36.960 | sometimes they ask good questions, right?
00:44:38.900 | And we were not the right people for this.
00:44:40.980 | Our day-to-day was dealing with agents
00:44:42.620 | who are not nice people.
00:44:44.260 | - 'Cause you had to sign them up.
00:44:45.460 | - Well, you had to sign players up.
00:44:46.660 | - And you had to sell their earnings.
00:44:48.860 | - Yeah, so, A, just the cycle took so long.
00:44:51.700 | There was one guy, we were doing this as charity.
00:44:53.980 | We were like, don't give us back anything.
00:44:55.940 | - You're just gonna give money away?
00:44:56.980 | - We're gonna give you money.
00:44:58.060 | We still didn't sign them after six months.
00:44:59.580 | - Oh my God.
00:45:00.420 | - Like, I don't understand at this point.
00:45:01.620 | Like, it was taking too long.
00:45:02.940 | We weren't the right people.
00:45:03.780 | - It's your OPMF.
00:45:04.620 | (laughs)
00:45:05.440 | - And the most important thing,
00:45:06.280 | which actually leads us to WannaCraft,
00:45:07.100 | is this was not a tech product.
00:45:09.380 | And Moonshot was not a tech product.
00:45:11.500 | It was a legal product.
00:45:12.340 | It was a regulation.
00:45:13.180 | It was like, hey, now you can invest in people.
00:45:14.860 | The reality is, like, I think we just found out
00:45:17.000 | the hard way that we were building something
00:45:18.980 | people did not want.
00:45:21.200 | And the thing that, like...
00:45:24.000 | So we realized, like, we're building something
00:45:25.400 | that isn't our strengths.
00:45:26.280 | Let's find something else.
00:45:27.580 | Looking back, I can also kind of, like,
00:45:30.440 | there were pointers when we were at Moonshot
00:45:32.260 | about, like, yeah, we should build something technical.
00:45:35.480 | I was listening to the "Lang Chain" episode
00:45:37.120 | that you guys did, and...
00:45:38.720 | Harrison, right?
00:45:40.560 | - Yes.
00:45:41.380 | - I remember you saying he was a big sports fan.
00:45:43.080 | - Yes.
00:45:43.920 | - And, you know, he built this, like,
00:45:44.880 | basketball scraper or whatever.
00:45:46.920 | Literally, my biggest, like,
00:45:48.340 | the thing I was most excited about with Moonshot
00:45:50.640 | was, like, okay, let's build out this infrastructure
00:45:52.040 | so that people can invest and blah, blah, blah, blah, blah.
00:45:53.640 | And then we get to the point where we can do, like,
00:45:55.160 | data analytics on them and start predicting.
00:45:57.000 | I'm like, this is the most exciting part of this.
00:45:59.200 | And I was like, this is definitely not
00:46:01.160 | the core part of the business.
00:46:02.520 | So I'm excited about something that's gonna take
00:46:04.200 | either years to get to or something else.
00:46:06.040 | But the core, you know, we designed our landing page
00:46:08.920 | and built it.
00:46:09.760 | That was the funnest thing I've done in Moonshot.
00:46:12.000 | So we're like, when we decided to pivot,
00:46:13.840 | we were like, we need to do something technical.
00:46:16.240 | The story from there just became, like, okay, cool.
00:46:19.280 | Let's list some ideas.
00:46:20.400 | What are we gonna track?
00:46:21.240 | Which one do we have the most conviction in?
00:46:22.400 | We rank them.
00:46:23.680 | And Monocraft was the one we had the most conviction in.
00:46:25.660 | It was this idea, again, expanding access
00:46:27.420 | and just translating your ability to produce content.
00:46:30.760 | So producing content in one format
00:46:33.760 | and then taking that to all the other formats.
00:46:35.940 | So we built that.
00:46:36.780 | We built the podcast builder.
00:46:38.920 | Super quick prototype because I think at this point,
00:46:42.080 | to anyone pivoting or hard pivoting or considering it,
00:46:44.780 | the name of the game isn't, like,
00:46:47.520 | to get attached to your idea.
00:46:48.520 | Just, like, actually, you should be trying
00:46:50.160 | to invalidate this idea as quickly as possible.
00:46:52.160 | So get it out there and let people tell you
00:46:53.640 | it's a piece of shit.
00:46:54.620 | - Okay, so what did you do to get it out there?
00:46:55.460 | - So we built it out.
00:46:56.520 | We built out a little, like, UI.
00:46:58.760 | Literally no authentication.
00:47:00.040 | It was a form where what you guys see now
00:47:02.400 | on our platform, which is, like,
00:47:03.240 | the content script page, blah, blah, blah, blah, blah.
00:47:05.120 | It was one page.
00:47:06.080 | Like, you click and it was the most janky React stuff
00:47:09.280 | that we had.
00:47:10.120 | Zero authentication.
00:47:10.940 | So in theory, if people found it,
00:47:12.280 | they could just produce as much audio as they wanted.
00:47:15.380 | - I'm quickly developing an opinion, by the way,
00:47:16.820 | that I think all early-stage startups
00:47:18.920 | should not have a marketing site.
00:47:20.880 | Your .com should just be your app.
00:47:23.200 | - That's an interesting one.
00:47:25.120 | I mean, we, I don't think we have one too much.
00:47:27.240 | I think you still need, like, discoverability.
00:47:30.040 | - Okay.
00:47:30.860 | - Whether that's your marketing site
00:47:31.700 | or a lot of artifacts.
00:47:32.520 | I think we are moving to--
00:47:33.360 | - Or, like, embed your app into the marketing site
00:47:35.320 | or whatever.
00:47:36.160 | Just, like, have, get people using it
00:47:38.760 | as much as possible, right?
00:47:39.760 | - Yeah, I think that you're, like,
00:47:41.480 | the most convincing thing is, like,
00:47:43.840 | so we have a feature on our app,
00:47:45.040 | which is, like, test with example.
00:47:46.120 | As soon as you log in, it says, "Test with example."
00:47:47.600 | And the whole point of that was, like,
00:47:49.280 | to, between login and audio generated,
00:47:53.560 | how few clicks does this take?
00:47:55.080 | - Yeah, yeah.
00:47:55.920 | And yours was?
00:47:56.760 | - Like, two or three clicks.
00:47:57.580 | It's, like, test, create podcast.
00:47:59.080 | - You, like, pre-filled everything.
00:48:00.080 | - Generate script.
00:48:00.920 | - Yeah.
00:48:01.740 | - Generate audio.
00:48:02.580 | - Yeah, yeah, yeah.
00:48:03.420 | - Right?
00:48:04.240 | So, and arguably, that was still too many clicks.
00:48:05.120 | We should probably put something
00:48:05.960 | on the actual landing page,
00:48:07.000 | so that, like, you can see--
00:48:07.840 | - Yeah, now you have just the player there, right?
00:48:09.320 | You can just kind of listen
00:48:10.160 | to the Hacker News Daily thing.
00:48:12.160 | But what signals did you get from doing that?
00:48:14.760 | - The signal that we got is that someone picked it up
00:48:16.200 | on Twitter, and just, like, you know,
00:48:17.680 | all these, like, AI--
00:48:18.800 | - Influencer voice.
00:48:20.120 | - So, someone picked it up, posted it.
00:48:21.800 | We started getting a ton of inbound.
00:48:24.040 | So, we were like, "Holy shit, let's just, like,
00:48:26.080 | "paywall this."
00:48:27.600 | So, we just, like, again, the jankiest Stripe integration,
00:48:30.880 | which was basically, like,
00:48:32.200 | we have an app with a Stripe integration.
00:48:34.080 | This was just to ship it within the hour.
00:48:35.440 | We have an app with a Stripe integration
00:48:37.480 | that, once you click, then takes you to a different app,
00:48:40.720 | hosted somewhere else.
00:48:41.560 | So, that one was still unauthenticated.
00:48:44.320 | - It was hilarious.
00:48:45.160 | - Yeah, security by obscurity.
00:48:46.000 | - It was hilarious, right?
00:48:47.440 | But we basically just made that, like, 3K in one day.
00:48:52.440 | - One day, yeah.
00:48:53.680 | - We charged around, like, 50 bucks.
00:48:55.120 | We didn't even, literally didn't think about it.
00:48:56.400 | We just charged 50 bucks, and people paid.
00:48:58.000 | And we're like, "Okay, well, there's something there."
00:48:59.920 | - Yeah.
00:49:00.760 | - So, then--
00:49:01.600 | - 50 bucks for one--
00:49:02.840 | - For a month.
00:49:03.680 | We were just charging 50 bucks a month, nothing.
00:49:06.720 | Like, just, like, will someone pay for this?
00:49:08.120 | - And you were just, like, on one, like, VPS somewhere.
00:49:10.080 | - Yeah, will someone pay for this?
00:49:11.080 | We were on, like, on one EC2 instance.
00:49:12.480 | - EC2 instance, yeah.
00:49:13.320 | - You know what I mean?
00:49:14.160 | Like, it was janky.
00:49:15.000 | We were like, "Just someone needs to pay for this
00:49:16.320 | "before we move further."
00:49:17.440 | Someone did.
00:49:18.280 | People did.
00:49:19.120 | So, then we were like, "Okay, cool.
00:49:20.160 | "This is interesting."
00:49:21.120 | - Interesting.
00:49:22.120 | I'm gonna move up the question that we said
00:49:24.080 | was gonna be the meatiest question of this interview.
00:49:27.440 | So, you chose this out of your list of ideas.
00:49:31.040 | And this is one of the things
00:49:32.960 | that a lot of AI founders worry about, right?
00:49:34.560 | So, the framing of this is,
00:49:36.520 | are you worried that you're a thin wrapper around 11 labs?
00:49:39.040 | What is your mode?
00:49:40.440 | And how do you think about that
00:49:43.440 | when you're evaluating your idea?
00:49:45.280 | - I think you can, that's a seminal question.
00:49:48.400 | I think anyone in an AI startup is,
00:49:50.600 | I think, frankly, everyone in an AI startup
00:49:53.360 | should convince themselves of this.
00:49:55.240 | Don't listen to me.
00:49:56.080 | And, like, just make sure from first principles
00:49:57.480 | that you can derive this.
00:49:58.920 | But I guess I would start by first saying,
00:50:03.080 | what is a mode?
00:50:04.520 | Well, like, or what is defensible?
00:50:07.600 | In theory, if we're just taking it,
00:50:09.120 | and this is trivially true,
00:50:10.920 | but the fact that someone built it
00:50:12.400 | means someone else can build it, right?
00:50:14.480 | So, modes tend to just be around,
00:50:16.480 | built around, like, you have a lot of network effects,
00:50:19.000 | or you have a really good product for this use case,
00:50:21.880 | or, you know, something like that.
00:50:23.560 | And I think, typically, when people ask this question
00:50:28.080 | in the AI context, they're thinking of, like,
00:50:30.040 | okay, you're a thin wrapper,
00:50:31.920 | you're an application layer thing,
00:50:37.280 | as opposed to you're one of the, like,
00:50:38.840 | underlying technologies or APIs that people use.
00:50:41.760 | Cool, I think that's fair.
00:50:45.960 | But I think the reality is that, like,
00:50:47.560 | yeah, these APIs exist,
00:50:49.080 | and they probably do serve a million different use cases,
00:50:52.440 | but they're not built to serve
00:50:54.040 | these million different use cases.
00:50:54.920 | So, whenever you ask the question of modes,
00:50:56.320 | it always has to be with the perspective
00:50:59.840 | of who is the user I'm building this for, right?
00:51:02.640 | I can use chat GPT to do half of my writing,
00:51:07.040 | but, you know, I don't know.
00:51:08.040 | Jasper claims that they do this much better for marketing,
00:51:10.560 | so it's tailored.
00:51:11.480 | Actually, don't quote me on how well they're doing
00:51:13.720 | after chat GPT came out,
00:51:14.720 | 'cause they were really big before.
00:51:16.000 | - Yeah, there's some negative data points,
00:51:17.600 | but I'm sure they--
00:51:18.640 | - I know, but the point is, like,
00:51:20.120 | you're making this easier.
00:51:21.840 | We make creating a podcast easier,
00:51:24.640 | and there is tooling there.
00:51:25.960 | We help you, we can post it directly through us.
00:51:28.720 | We have the tooling around, you know,
00:51:30.160 | setting the intros and the outros.
00:51:31.920 | We have the music, we have an editor.
00:51:33.480 | All these things are also getting
00:51:34.440 | just much more and more developed.
00:51:36.600 | We're building templates
00:51:37.640 | so that you can do different style of podcast.
00:51:39.720 | So the idea is, if you're trying to start a podcast,
00:51:42.120 | yeah, don't go to a generic text-to-speech engine.
00:51:44.920 | Come to us.
00:51:46.280 | And the reality is that we then can,
00:51:48.440 | in a very opinionated way,
00:51:49.680 | actually select which text-to-speech engine we want, right?
00:51:52.840 | So we actually have just, like,
00:51:55.040 | in my mind, it's the application layer,
00:51:56.360 | fundamentally, that, you know, people use,
00:51:59.000 | and then all these API layers
00:52:00.680 | are what developers use to build products on top of them,
00:52:03.600 | right? - Right.
00:52:04.680 | - I appreciate that it is, like, a seminal
00:52:07.240 | and really hard thing to wrap your head around,
00:52:09.880 | especially if you're, like, about to invest in a company.
00:52:11.360 | It's like, will they actually just be defensible
00:52:13.400 | and be able to grow? - Yeah.
00:52:14.760 | - And yes, there's no doubt that companies can do this.
00:52:19.320 | The question is just, like,
00:52:20.160 | are you building the right product for the right use case?
00:52:23.600 | I think, particularly, if you're, like,
00:52:24.440 | always framing your company as an AI company,
00:52:26.840 | then you're putting the carriage before the horse
00:52:30.440 | in the sense that you focused on the implementation
00:52:32.160 | rather than the use case.
00:52:33.000 | Focus on the use case.
00:52:34.400 | And then build a product for it.
00:52:35.640 | - Yeah. - Right?
00:52:36.480 | 'Cause fundamentally, you know,
00:52:37.960 | any of the SaaS's that exist,
00:52:40.240 | think, like, more traditional SaaS.
00:52:41.920 | What's their mode?
00:52:42.760 | The technology, everyone has access to it,
00:52:44.240 | so they just pick the thing
00:52:45.280 | that does it better than the other.
00:52:48.600 | Now, that mode question is super interesting
00:52:50.600 | because I think you should actually
00:52:53.080 | - Flip it around. - Flip it around,
00:52:54.000 | which is, what is your mode as an API?
00:52:55.880 | - Yeah. - Right?
00:52:57.040 | So, Chad GPT, like, yeah, fine,
00:52:59.880 | they had a first-mover advantage,
00:53:01.200 | and I think, you know, by no means,
00:53:03.520 | this is my opinion, but by no means
00:53:05.480 | was Google, like, caught off-guard with this, right?
00:53:07.800 | It just, Google has some,
00:53:10.200 | half the technologies that Google invented
00:53:12.160 | are actually what's used to power all these transformers.
00:53:15.060 | But, you know, it went against Google's strategy, maybe,
00:53:19.480 | to, like, be the first-mover in this
00:53:21.080 | 'cause they cannibalized their own market.
00:53:22.200 | Whatever it was, I'm not sure.
00:53:24.120 | But, yeah, open AI's mode
00:53:26.760 | is that they paid for the training bill.
00:53:28.840 | - Yeah. - So, they just have
00:53:29.960 | a good model.
00:53:30.800 | Cool, people now know that that's valuable.
00:53:31.880 | - And, they hired, like, very top-notch.
00:53:33.520 | - Superstar, sorry, obviously.
00:53:34.480 | Obviously, not taking that for granted.
00:53:35.720 | But, like, they, you know, assuming everyone
00:53:38.160 | can do the hiring and that these people exist,
00:53:40.720 | they paid the bill, and they were the first to launch this.
00:53:43.120 | But now people know it's a thing,
00:53:44.160 | so people are gonna launch similar APIs.
00:53:46.200 | So, what is your mode as an API?
00:53:49.160 | So, it's just an, it's an existential question.
00:53:52.160 | It's like, how do you do, how do we defend any of this?
00:53:54.960 | - Yeah. - And you do this, frankly,
00:53:56.520 | by being probably better just as a product.
00:54:00.140 | Again, the product is always with the perspective
00:54:01.760 | of who's your customer that you're selling it to.
00:54:04.160 | And the other thing is, frankly,
00:54:05.560 | that, like, let's not forget, the market is huge.
00:54:08.760 | There is space for everyone, right?
00:54:10.720 | If you manage to, like, if there's four good products
00:54:12.800 | out there in any specific thing, the market is huge,
00:54:14.700 | and they're all gonna be able to, you know,
00:54:16.940 | make a living out of it.
00:54:17.840 | - Sometimes people justify markets.
00:54:21.060 | By the way, that was a really good answer.
00:54:22.080 | Thanks for taking that hit on.
00:54:24.200 | - I've had to answer that question way too many times.
00:54:26.200 | - Yeah, I know, it's not the first time.
00:54:28.920 | But, I think it's actually, you know,
00:54:30.520 | having been an investor, it is more important for you
00:54:32.520 | to answer that question authentically for yourself,
00:54:34.920 | 'cause you're the one spending your time on this.
00:54:36.840 | We're just giving you money, it's not that big of a deal.
00:54:39.840 | Well, so, ah, I forgot what I was gonna say there.
00:54:45.320 | Yeah, I think 11 Labs, you know,
00:54:46.400 | I think the thing that you were mentioning is, you know,
00:54:50.000 | there is not that much loyalty.
00:54:52.120 | Obviously, you and Matti are friends.
00:54:54.360 | But there's not much loyalty if your main goal
00:54:57.280 | is to serve API, because you just need to wait
00:55:00.440 | until the next best one comes along.
00:55:03.120 | - The same thing as Uber, like, why is Uber defensible?
00:55:05.640 | - Yeah, yeah, yeah.
00:55:07.240 | My favorite quote, actually, I went to an early,
00:55:11.480 | like, pre-HRGBT forum with Sam Altman,
00:55:15.120 | and I had this video advice from Sam
00:55:17.640 | that said, like, Facebook had no moat,
00:55:20.000 | and they just built and got the network.
00:55:22.040 | - But frankly, also, Facebook was building something
00:55:24.280 | back then, which is kind of ludicrous, like,
00:55:26.600 | cool, you're building a social media app,
00:55:27.800 | okay, how big can it be, right?
00:55:29.160 | - Like, how big can the internet be?
00:55:30.840 | You know what I mean?
00:55:31.680 | All of a sudden, it's this behemoth.
00:55:32.520 | So it's like, yeah, the fact that it was built,
00:55:35.240 | again, trivially true, but the fact that it was built
00:55:37.280 | means it can be built by anyone else.
00:55:39.480 | So you, there is no such thing as, like,
00:55:42.360 | a absolute true moat.
00:55:44.880 | The question is how well, how quickly,
00:55:47.120 | how much earlier than everyone else did you get there,
00:55:50.440 | and a million other things as well.
00:55:51.640 | - Yeah, cool.
00:55:53.200 | Two more sort of product-building questions,
00:55:55.280 | and then I want to turn the mic over to you
00:55:57.600 | to ask me anything as a podcaster.
00:55:59.740 | So one is script generation, which, you know,
00:56:01.960 | sort of the two core activities of Wondercraft.
00:56:05.440 | I think you guys do a really good job
00:56:06.880 | of summarizing the Hacker News comments and the content.
00:56:11.880 | In my mind, the hardest thing is just managing context,
00:56:15.160 | because, obviously, it is unbounded for the context window.
00:56:18.960 | Do you use LangChain?
00:56:20.360 | Do you build your own LangChain?
00:56:21.720 | - I like LangChain, I've never used it.
00:56:23.760 | I've never, ever used it.
00:56:24.600 | - So you just build your own recursive summarization thing.
00:56:27.360 | - Yeah, so--
00:56:28.200 | - Do you use a vector database?
00:56:29.040 | Which one?
00:56:29.880 | (both laughing)
00:56:31.640 | - I'm not gonna say any further,
00:56:32.520 | unless you realize how unsophisticated we are.
00:56:34.880 | - It's okay, no, I think it's okay.
00:56:36.480 | People, builders and investors want to know
00:56:39.480 | what's actually useful--
00:56:40.320 | - I want to say the thing, the Hacker News recap,
00:56:42.600 | I think, I don't know, but we can look on the date
00:56:45.440 | of the first episode when we launched it.
00:56:47.320 | I think it was like the 8th of April or something like that.
00:56:50.040 | We decided on Wondercraft probably on the 5th of April.
00:56:55.400 | So this is how quickly it took us--
00:56:57.080 | - The name of Wondercraft?
00:56:57.920 | - As in the company, this is what we should pursue.
00:56:59.840 | - I see, so you took three days.
00:57:01.000 | - And this is before people started paying.
00:57:02.200 | We were just like, Hacker News started off
00:57:04.480 | as a proof of concept.
00:57:05.320 | It's like, do people even want to listen
00:57:06.560 | to AI-generated content?
00:57:08.100 | And it's been going on since.
00:57:09.440 | So it wasn't actually that hard
00:57:10.920 | to build the Hacker News recap.
00:57:12.520 | Obviously, now it became a little bit more sophisticated,
00:57:14.400 | so we summarized comments and blah, blah, blah, blah, blah.
00:57:16.240 | - Yeah, that's great, by the way.
00:57:17.760 | I love that.
00:57:18.600 | - But actually, it's like, especially with our API now,
00:57:20.780 | it's super easy to do something like that.
00:57:22.800 | We, Hacker News, frankly, just doesn't need
00:57:25.420 | a vector database unless we're trying to do something.
00:57:28.980 | At the moment, we just summarize the posts.
00:57:30.900 | And actually, for anyone listening
00:57:31.860 | who listens to the Hacker News recap,
00:57:33.020 | just tell us some feedback on how to improve it.
00:57:35.580 | At the moment, it's just like, it's one long narration.
00:57:38.020 | But one thing we're thinking of doing, for example,
00:57:40.060 | is sometimes you have multiple posts
00:57:42.580 | talking about the same topic.
00:57:43.700 | So actually, an idea is do top 10 topics
00:57:45.760 | instead of top 10 posts.
00:57:48.080 | That starts to need, I don't think that needs a database,
00:57:50.860 | a vector database.
00:57:52.200 | I think it needs something, you know,
00:57:53.700 | it needs to do some embedding, manipulation.
00:57:55.400 | - You need MP.A?
00:57:56.420 | - Yeah, exactly.
00:57:58.500 | But something like that.
00:58:00.020 | I think for when you start to really, really scale up
00:58:03.740 | the content that you're pulling from,
00:58:06.560 | then you start needing a vector database.
00:58:09.300 | - Okay, so what are the actual hard problems
00:58:11.140 | for script generation?
00:58:12.580 | Prompt engineering?
00:58:14.340 | - Prompt engineering is one of them,
00:58:15.940 | especially when you're trying to tune things,
00:58:18.060 | like tone, or how long it should be, and all of that.
00:58:21.460 | I actually think, so the way our app works,
00:58:24.100 | and maybe that's leading to your next question,
00:58:25.660 | but the way our app works is twofold.
00:58:27.500 | You have the script generation,
00:58:28.740 | and then you have the audio generation.
00:58:30.540 | In our minds, we imagined it as a really linear flow.
00:58:33.820 | That, of course, I want to generate the script,
00:58:36.460 | and then I want to generate the audio for it.
00:58:39.700 | Where in reality, I don't think it's that linear,
00:58:41.580 | to be honest.
00:58:42.420 | I think it's very circular in how you do this, right?
00:58:44.400 | So you generate the script,
00:58:46.220 | you listen to it, it doesn't sound quite right,
00:58:48.220 | and you want to modify the tone a little bit, and so on.
00:58:50.140 | So I think our UI doesn't get that wrong,
00:58:51.940 | unless you're exactly what you're going for,
00:58:54.180 | is this recap, summarization style.
00:58:57.080 | So I think what we need to do a much better job
00:58:59.780 | is this templatization, and this on-the-go, on-the-run,
00:59:04.700 | kind of like, okay, let's play around
00:59:06.260 | with this text via some prompt.
00:59:07.840 | So I think it's definitely an improvement
00:59:09.820 | we're looking to make, but yeah,
00:59:12.860 | it's a prompt engineering question, fundamentally.
00:59:15.060 | It's like how, prompt engineering on one level,
00:59:17.840 | the next level's like how good, how fine-tuned
00:59:19.900 | is your model to specific tasks,
00:59:21.300 | and then maybe you can use your own model if you wanted to.
00:59:23.460 | - Would you look at that, or is that not--
00:59:24.940 | - We'll definitely look at that,
00:59:25.820 | but you look at that when it's a problem.
00:59:28.800 | It's not a problem right now.
00:59:30.820 | - I would say, as a listener, daily listener,
00:59:33.680 | I think your show notes are really good.
00:59:36.100 | Most daily podcasters are horrible at show notes.
00:59:38.260 | They don't tell you anything.
00:59:40.100 | Here, if I want to click in, I can click in there,
00:59:42.460 | so definitely keep that.
00:59:43.880 | But the top thing that annoys me about you,
00:59:46.740 | and I'm just complaining now,
00:59:48.380 | is you don't write the post title.
00:59:51.200 | You don't spell out the--
00:59:52.660 | - Really, we don't write the post title?
00:59:53.940 | - Yeah. - I think we used to.
00:59:55.540 | - Well, it's gone. - That's horrible.
00:59:56.700 | I'm sorry.
00:59:57.540 | - So the voice, by the way,
01:00:00.040 | I wonder if there's any story behind picking the voice.
01:00:02.320 | The voice would just say, "This post does,"
01:00:04.400 | and it doesn't tell me the title of the post.
01:00:05.240 | - Oh, in the audio, in the audio, we don't say that.
01:00:06.940 | Ah, sorry, sorry, got it, got it, got it.
01:00:08.260 | - And when I'm running--
01:00:09.620 | - Did you email us about this?
01:00:10.820 | - No, no, I've kept it to myself.
01:00:13.060 | - Someone did email us, so it's like,
01:00:14.220 | "Please say the post." - I'm like,
01:00:15.040 | "This is so obvious."
01:00:15.880 | I would actually just fork "WonderCraft"
01:00:17.700 | just to add that one feature,
01:00:19.620 | because it's so annoying when I don't know
01:00:21.140 | what I'm listening to.
01:00:22.340 | - We'll fix it, we'll fix it, easy peasy.
01:00:24.720 | - Okay, cool.
01:00:26.820 | The audio generation, you use 11Labs.
01:00:30.540 | What makes a good podcast voice?
01:00:32.300 | You have a bunch of options that I clicked,
01:00:34.580 | and in my mind, I like a deep voice.
01:00:36.940 | I like the Morgan Freemans.
01:00:38.300 | You don't have that many deep voices.
01:00:41.060 | Do we want, is there such a thing as a high-energy voice?
01:00:44.200 | You also insert breaths.
01:00:46.540 | 11Labs has also advertised that they have an AI
01:00:48.860 | that can laugh, which I think is fun, important.
01:00:51.780 | Basically, what makes a good audio?
01:00:53.540 | Yeah, generate audio.
01:00:55.220 | - Yeah, it's, it depends, again, on the perspective.
01:00:59.020 | Everything is kind of answered
01:01:00.140 | with the frame of reference that you're looking at.
01:01:02.820 | If you like a deep voice,
01:01:04.140 | A, that's kind of a personal preference,
01:01:05.980 | and B, it just kind of depends on the thing.
01:01:07.660 | So if you, I don't know, let's do,
01:01:09.300 | say you're doing something like meditative
01:01:10.780 | or kind of affirmations or something
01:01:13.100 | that encourages people every day.
01:01:14.720 | You probably do want a slow, deep voice,
01:01:16.320 | something relaxing.
01:01:17.280 | If you're doing the Hacker News recap--
01:01:19.840 | - Cheerful. - We picked Anna,
01:01:21.280 | who's our default voice. - Yes, Anna, yeah.
01:01:23.120 | - Because-- - I have an attachment to Anna.
01:01:25.680 | - Yeah, we all do. (laughs)
01:01:28.640 | She's just news anchory style, very professional,
01:01:31.200 | very formal, very neutral. - Very pleasant, yeah.
01:01:32.480 | - Very neutral.
01:01:33.320 | So it depends, really, on what makes a good voice.
01:01:38.240 | It depends on what you're doing.
01:01:39.840 | There's a few things, but if you're doing an interview,
01:01:42.680 | I think it also just, frankly,
01:01:45.000 | then you get into the question
01:01:45.840 | of what makes a good podcast.
01:01:46.800 | Well, the good podcast is like,
01:01:48.360 | I think it's also kind of a personal question,
01:01:50.560 | which I haven't, or probably there's a general trend
01:01:52.820 | that I'm yet to decipher.
01:01:54.560 | But yeah, you probably do want
01:01:57.140 | a little bit of humanity in there.
01:01:58.360 | You want a stutter. - Yeah.
01:02:00.200 | - You want some pauses, right?
01:02:01.520 | I'm speaking, I don't speak in complete utterances.
01:02:03.520 | I have an utterance, and then I pause a little,
01:02:05.120 | and then I speak again, and so on.
01:02:06.820 | Laughs and something to make it human.
01:02:09.600 | - Yes. - It's kind of overlaying
01:02:10.840 | of the two, if you have two speakers,
01:02:12.320 | it's like exchange, right?
01:02:14.600 | I will be speaking.
01:02:15.500 | If you look at the transcript of this episode,
01:02:17.240 | we probably overlap in when we're speaking.
01:02:19.360 | - And that's fun.
01:02:20.200 | - And that's actually interesting, right?
01:02:21.520 | 'Cause it is a conversation.
01:02:22.360 | - Shows the sign of excitement,
01:02:23.880 | especially in our studio when we're three people,
01:02:26.220 | and we're all talking at once, you know it's good.
01:02:28.080 | - Yeah, yeah.
01:02:29.400 | I don't like this zoomification style
01:02:30.800 | where if you're gonna big zoop and big zoom,
01:02:33.600 | only two people can speak the second
01:02:35.040 | more than two people try to speak.
01:02:36.400 | - Yeah, yeah. - It's a disaster.
01:02:38.040 | So I think it frankly just depends on what you're doing.
01:02:40.720 | We are, yeah, at the moment we're really good
01:02:42.480 | at doing this narration stuff,
01:02:43.680 | but I think we are building a lot of functionality
01:02:46.160 | and tooling to just make this multi-host thing
01:02:50.240 | a more of a reality.
01:02:51.320 | - Okay, okay.
01:02:52.960 | I would say objectively, if it was a friend and company,
01:02:56.720 | not that important.
01:02:57.680 | So this comes down to how human should your users try to be?
01:03:03.960 | Because I'm fine with Hacker News Daily making mistakes,
01:03:09.680 | because I know it's AI generated, right?
01:03:12.840 | I would be less fine if you were not up front,
01:03:16.080 | and Alec here actually thought it was a human
01:03:19.360 | when he first listened to it,
01:03:20.440 | 'cause Anna's just really good.
01:03:22.320 | But then you'll make mistakes, like pronunciation mistakes.
01:03:25.320 | So I actually have a clip that I wanted to play you.
01:03:28.840 | Anna takes this, on September 8th,
01:03:30.960 | Anna was taking a lot of breaths.
01:03:32.160 | She was very out of breath.
01:03:33.600 | I was very worried about her.
01:03:35.280 | She was hyperventilating.
01:03:36.520 | I was like, "Inner UK?"
01:03:39.040 | Anyway, so basically, I think if you disclose up front
01:03:41.880 | that you're an AI podcast,
01:03:43.240 | then people will be like, "Oh, okay, I tolerate that mistake
01:03:45.240 | "and I use you for information,
01:03:46.560 | "and not for believing that there's some human
01:03:48.840 | "on the other side that I might meet someday."
01:03:51.920 | But if you're investing so much effort into being real,
01:03:57.160 | then your end goal is you have to lie to your users.
01:04:00.960 | - I don't think the investing and being real
01:04:02.760 | is for the purpose of deception
01:04:04.600 | as much as it is for the purpose
01:04:05.720 | of making it slightly pleasant to read about.
01:04:07.680 | I think on Hacker News, we do say on our Spotify page
01:04:11.800 | that this is an AI-generated podcast.
01:04:13.760 | - For now, but--
01:04:15.600 | - As in, yeah, yeah, so there's two things.
01:04:18.280 | I think if you wanna be smart about this,
01:04:20.480 | you should say that this is AI-generated content.
01:04:22.320 | The second people find out that it's not you,
01:04:24.400 | the backlash is gonna be big,
01:04:26.360 | 'cause it will be interpreted as deception.
01:04:29.160 | So you should do this just to be smart.
01:04:30.520 | I don't think there's a point in lying,
01:04:32.640 | especially if the content that you're putting out there
01:04:34.440 | is just like, this is informational for you,
01:04:36.480 | so consume it, this was efficient,
01:04:38.160 | this helped us put it out there.
01:04:40.240 | The second thing is, frankly, I don't think it's up to you
01:04:42.960 | whether you tell them or not.
01:04:44.120 | Very, very soon-- - They'll know.
01:04:45.800 | - Google is just gonna mark things as AI-generated.
01:04:49.280 | So I think there's a new thing.
01:04:51.240 | I saw a quick YouTube video about it,
01:04:53.760 | so I don't know what the exact terms and conditions are,
01:04:56.640 | but YouTube has, I think, released a new monetization rule,
01:04:59.640 | and it does mention something about AI-generated content.
01:05:02.960 | So there is, it's not up to you anymore.
01:05:06.080 | People are gonna know that this is AI-generated,
01:05:07.840 | so I think it's just in your interest
01:05:09.040 | to say that you're AI-generated.
01:05:10.400 | Ain't no shame. - Yeah, no shame at all.
01:05:12.880 | - 'Cause fundamentally what we do
01:05:14.680 | relies on the premise that you have done some content.
01:05:17.480 | We don't generate our own content.
01:05:19.520 | We don't synthesize our information.
01:05:21.840 | It assumes that you've written a blog post,
01:05:25.000 | done an actual podcast, or have some artifact
01:05:27.640 | on which you wanna base
01:05:29.040 | what you're feeding through Wondercraft.
01:05:31.520 | - And you said in some of your material
01:05:34.360 | that I've seen before that you are interested
01:05:36.040 | in watermarking all your stuff.
01:05:38.200 | You haven't done it yet,
01:05:39.040 | but whenever there's a standard for doing that,
01:05:41.040 | you will do it. - Yeah.
01:05:41.880 | - Okay. - I think the thing
01:05:43.000 | that this is blocking on is the standard.
01:05:45.000 | I'm not super up-to-date on what the work on this is.
01:05:47.760 | - I think OpenAI will probably--
01:05:49.200 | - But there just needs to be a standard
01:05:50.500 | so everyone can interpret it. - Yeah, yeah, yeah.
01:05:52.920 | Yeah, cool, awesome.
01:05:54.440 | Great, I wanted to dive in a little bit on tech options,
01:06:00.720 | and then zoom out to just you asking me questions.
01:06:05.360 | So TTS Options, we talked a little bit about 11Labs.
01:06:09.440 | I would also say as a podcaster,
01:06:10.920 | the leading competition to you guys,
01:06:13.360 | I know it's not exact competition, but it's Descript,
01:06:16.100 | because they have Overdub.
01:06:17.800 | - Yeah, I think Descript is really good,
01:06:19.680 | and they're definitely a solid company.
01:06:23.360 | I've used their video editor before.
01:06:24.840 | It's great.
01:06:25.680 | The Overdub thing is super useful.
01:06:27.080 | I think it's really creative to have edit videos
01:06:30.280 | by editing the transcript.
01:06:31.600 | - Yes.
01:06:32.440 | - Super, super creative, super user-friendly.
01:06:33.680 | - Would you build that?
01:06:34.940 | - I think, again, it's like, we're not building
01:06:37.600 | for the sake of building.
01:06:38.440 | We're building more for the purpose of the user.
01:06:42.000 | - Yeah.
01:06:42.840 | - Whatever users find more interesting.
01:06:43.760 | I think what we're doing is we,
01:06:46.800 | the use cases are slightly different, right?
01:06:48.600 | And I think the people that they're targeting
01:06:50.680 | are slightly different.
01:06:51.600 | - Yeah.
01:06:52.540 | - We do wanna have a lot of automation on the script side
01:06:55.120 | to also just help out with the way you formulate
01:06:59.320 | your content or the way you pull your content,
01:07:01.280 | much more so than just the editing process.
01:07:02.720 | - The ingest, yeah.
01:07:03.560 | - Yeah.
01:07:04.380 | - Okay, got it.
01:07:05.220 | And I just wanna map out,
01:07:06.440 | here's how I think about TTS, text-to-speech.
01:07:09.320 | There's the big cloud options,
01:07:10.320 | Amazon Polly, Google Text-to-Speech,
01:07:12.480 | and Microsoft Cognitive Services.
01:07:14.860 | - Mm-hmm.
01:07:16.360 | - As someone who is ex-Amazon,
01:07:18.440 | I'm very embarrassed by Polly.
01:07:19.600 | It sucks.
01:07:20.480 | - Yeah, Google's isn't great, and Microsoft--
01:07:22.040 | - I'm sure you investigated all these things,
01:07:23.360 | and you're like, "Okay, this is not serious."
01:07:25.800 | There's Play.ht, which is probably the other big YC--
01:07:29.040 | - Yeah.
01:07:29.880 | - Alum.
01:07:30.960 | Just a quick two seconds on your thoughts on Play.ht.
01:07:33.560 | - Sounds good.
01:07:36.380 | I think it doesn't sound super,
01:07:37.200 | as good as Alum Labs, my opinion.
01:07:38.800 | - Yeah.
01:07:40.000 | - But I think it sounds good.
01:07:40.840 | - But I have heard other founders tell me this as well.
01:07:43.140 | - Yeah, they're--
01:07:43.980 | - And I don't know why.
01:07:44.800 | - I think they have a, what's it called?
01:07:48.680 | A more comprehensive platform.
01:07:50.600 | - Okay.
01:07:51.440 | - As in they let you do this pronunciation.
01:07:53.480 | They just have a lot of tooling around it.
01:07:56.320 | - So different features.
01:07:57.160 | - I think quality, in terms of the voice,
01:08:00.080 | 11 Labs is still better.
01:08:01.260 | - Yeah.
01:08:02.100 | - I think they are releasing a new model.
01:08:03.080 | I don't know if they've released it already or not.
01:08:03.920 | - Yes, they did, yeah.
01:08:05.500 | - Could be better.
01:08:06.340 | I don't know.
01:08:07.160 | But they do have some functionality out there.
01:08:09.160 | - Yeah.
01:08:10.000 | They also released the viral
01:08:11.420 | Joe Rogan, Steve Jobs interview from last year.
01:08:13.400 | - Yeah.
01:08:14.240 | - And on your landing page, you were like,
01:08:15.560 | "This is something that WonderGraph will never do.
01:08:17.800 | "AI content speaking to each other."
01:08:22.080 | - Yeah, who wants to listen to that?
01:08:24.200 | Like, it's fun--
01:08:25.040 | - Apparently a lot of people.
01:08:25.880 | - It's fun 'cause it's a cool gimmick.
01:08:27.160 | I think it's like nice viral material.
01:08:29.600 | I would never listen to like a synthetic Joe Rogan.
01:08:33.720 | - Yeah.
01:08:34.560 | - This brings us on to a little bit
01:08:35.640 | about the whole like content question
01:08:37.720 | or the proliferation of AI, which is like,
01:08:39.800 | okay, if it's this easy for me to create content
01:08:43.120 | that's like, you know, somewhat engaging,
01:08:45.120 | like all these AI songs, the Drake song.
01:08:49.520 | Well, okay, so if it's this easy and it's just like,
01:08:52.000 | if it's this easy to generate content,
01:08:54.560 | well, why will I listen to it?
01:08:56.640 | Like, I think we already suffer from the problem
01:08:58.440 | that there's an oversaturation of content.
01:09:00.660 | So I think, you know, I'll give a Netflix example.
01:09:05.080 | When I go on Netflix, there's too much content.
01:09:06.320 | I don't like discovering new things.
01:09:08.040 | I either know exactly what I'm about to go and watch
01:09:10.240 | or I will re-watch one of the seven things
01:09:12.000 | that I constantly keep re-watching, right?
01:09:14.040 | So I'm not like, their discoverability isn't the best
01:09:17.240 | and every now and then I maybe like look at
01:09:19.360 | some of the other things,
01:09:20.180 | but I'm not trying to look at Netflix.
01:09:21.720 | - It's quite funny 'cause they used to invest a lot
01:09:23.520 | in their recommendation systems.
01:09:25.000 | And like, nobody I know uses that.
01:09:25.840 | - But now it's just too big.
01:09:27.160 | It's just too big.
01:09:28.040 | And they're like, it's overwhelming.
01:09:29.160 | So many tiles that you're looking at on the screen.
01:09:31.080 | So, cool.
01:09:32.520 | Now, what if I told you that there's probably
01:09:34.120 | is gonna 1,000X and half of it's gonna be AI generated?
01:09:37.640 | So what I just said is like,
01:09:39.120 | we all have this internal proxy
01:09:41.240 | of what a thing I'm likely to watch is,
01:09:43.220 | which is it was recommended by a friend
01:09:44.860 | or it's one of the things I previously watched.
01:09:46.720 | I'm not doing this discoverability.
01:09:48.320 | So imagine that's the case.
01:09:49.960 | I think another layer of proxy will just be like,
01:09:53.280 | from the Netflix category of the shows and movies,
01:09:56.280 | was this human made or is this AI made?
01:09:57.720 | If it's AI made, I'm not watching it.
01:09:58.960 | I want real human stuff.
01:10:00.420 | But I think there's gonna be a fine equilibrium
01:10:02.360 | that reaches, which is like,
01:10:03.260 | it's not even worth generating that much content with AI.
01:10:05.880 | I'm talking from entertainment purposes.
01:10:07.280 | I'm not talking so much for the social media propaganda,
01:10:11.240 | blah, blah, blah, not getting into it.
01:10:13.120 | But at least from an entertainment perspective,
01:10:15.920 | if I can just always have an on-demand Drake song
01:10:18.480 | that I wanna listen to,
01:10:20.120 | well, at some point I'm just gonna listen to Drake.
01:10:21.640 | 'Cause like, you know what I mean?
01:10:22.480 | - This is the greatest hits.
01:10:24.120 | - Yeah, it's like, I'm just gonna listen to a curated set.
01:10:26.120 | The fundamental thing is curation.
01:10:27.800 | It's just like a parallel example of the Discord thing.
01:10:30.760 | It's like full of noise.
01:10:31.880 | So I need to like figure out
01:10:34.120 | what is the right thing to consume.
01:10:36.600 | - Yeah, awesome, awesome.
01:10:38.680 | Great answer.
01:10:39.900 | I just wanna bring through,
01:10:41.960 | here's my map of the market, right?
01:10:43.760 | There's Speechify.com, which focuses on celebrity voices.
01:10:46.520 | I noticed that you don't have celebrity voices,
01:10:48.080 | probably 'cause of licensing issues, right?
01:10:49.640 | - Yeah, and it's also--
01:10:50.480 | - Like, they don't usually use their voice.
01:10:51.320 | - It could at some point,
01:10:52.240 | but like not a priority at the moment.
01:10:54.000 | - Yeah, I really want a Morgan Freeman one.
01:10:55.960 | (laughs)
01:10:57.160 | - That's gonna cost.
01:10:58.360 | - I know, I know.
01:11:00.040 | Mycroft.ai, privacy focus, run offline.
01:11:03.640 | Probably not--
01:11:04.480 | - Haven't heard of them, actually.
01:11:05.320 | - Not important for you.
01:11:06.480 | There is some interest in virtual characters for games.
01:11:10.560 | So Conv.ai is the one that I had listed here.
01:11:13.200 | Did you look at the gaming market?
01:11:15.320 | - Not deeply, to be honest.
01:11:16.480 | - Yeah.
01:11:17.320 | - But it could be an interesting one.
01:11:18.140 | - Yeah, people are exploring that.
01:11:19.280 | There's obviously Heijian now,
01:11:21.320 | and that is it for as far as I can scope out the landscape.
01:11:25.000 | And then there's the open source systems.
01:11:26.320 | So Tortoise TTS, as far as I can tell,
01:11:28.040 | is kind of market leader in open source.
01:11:29.840 | - Yeah, I would actually say probably a lot of these models
01:11:31.800 | were built on top of it.
01:11:33.600 | - What models?
01:11:34.440 | - Maybe.
01:11:35.280 | Some of the--
01:11:36.100 | - These other companies.
01:11:36.940 | - The companies, yeah.
01:11:37.780 | - Yeah, there's PyTSX, Koki, used to be Mozilla,
01:11:41.140 | and then Larynx.
01:11:42.880 | - Yeah.
01:11:43.720 | - Anyway, all these things.
01:11:44.540 | And then there's also sort of research-grade stuff
01:11:46.480 | coming out of the major big tech companies.
01:11:48.880 | You talked about Google Sandstorm.
01:11:50.480 | - Probably the one I'm most excited about.
01:11:51.800 | - Okay, why don't you--
01:11:52.880 | - Well, it's really good.
01:11:54.440 | You can check out the paper.
01:11:55.520 | - Yeah, we'll play a clip.
01:11:56.800 | - Yeah, they haven't, I think all you need
01:11:58.600 | is like three seconds.
01:11:59.600 | - Yeah.
01:12:00.440 | - And it'll just, it's like a three-second sample,
01:12:03.280 | and it'll play the audio in your tone.
01:12:05.000 | It sounds really human as well.
01:12:05.960 | Like it has utterances, it laughs.
01:12:08.100 | It's pretty accurate to like, it sounds human.
01:12:10.840 | - Yeah.
01:12:12.440 | - So very interested in that.
01:12:13.920 | They haven't open sourced it,
01:12:14.760 | and I assume for good reason.
01:12:16.000 | - Yeah.
01:12:16.840 | Google never launches anything.
01:12:17.800 | You have to wait for somebody to,
01:12:19.360 | or you guys could re-implement it yourself.
01:12:21.400 | - Yeah, yeah.
01:12:22.560 | GPUs after PMF.
01:12:25.200 | - Ah, that's a nice quote.
01:12:27.200 | How strongly do you believe that?
01:12:28.680 | - GPUs after PMF?
01:12:29.560 | - Yeah.
01:12:30.400 | - Well, I believe, I think this was another question, yeah.
01:12:31.880 | Which is like--
01:12:32.700 | - What is PMF?
01:12:33.540 | - No, what is your favorite like PG advice
01:12:36.280 | in building a company, right?
01:12:37.320 | I think that my favorite thing is just like,
01:12:38.680 | don't spend your money, obviously.
01:12:39.920 | - Ah, okay.
01:12:40.760 | - Everyone and their mother is trying to get a GPU
01:12:44.080 | at the moment, so I don't think it's,
01:12:46.920 | we're definitely substantially reducing our runway
01:12:49.680 | by doing that.
01:12:50.920 | - Yeah.
01:12:51.760 | - Obviously you do that when you believe
01:12:52.600 | the investment is worth it.
01:12:53.440 | - Yeah.
01:12:54.680 | - And again, you have to pick the time at which you do that.
01:12:57.080 | - I mean, there's other companies,
01:12:58.880 | I think this is somewhat consensus.
01:13:03.180 | I think the non-consensus thing is to spend a shit ton.
01:13:05.980 | So like inflection raising a few hundred million dollars,
01:13:09.200 | and then spending 95% of it on GPUs.
01:13:11.200 | - Same with Mistral.
01:13:12.040 | I think it depends on the company you're launching.
01:13:13.960 | I think if you're like, you know,
01:13:15.560 | maybe you're a brand new TTS company,
01:13:17.480 | maybe it is worth just doing that.
01:13:19.320 | - Yeah.
01:13:20.160 | - I don't know.
01:13:21.000 | - Okay.
01:13:21.920 | There's also AudioLM, also out of Google,
01:13:24.400 | VAL-E from Microsoft, and MetaVoiceBox.
01:13:27.120 | Are you just watching any of these?
01:13:28.560 | - Watching any of these, obviously paying close attention.
01:13:30.760 | - You try them all out, yeah?
01:13:31.800 | - Try them all out.
01:13:33.520 | - What are you looking for?
01:13:34.360 | What is like the holy grail?
01:13:36.040 | What are you looking for?
01:13:37.080 | - There's also how human it sounds,
01:13:38.360 | and like how likely I'd be to listen to this if I did it.
01:13:40.720 | Also how like customizable it is.
01:13:42.360 | - Yeah.
01:13:43.180 | - I think the problem with all these voice things,
01:13:44.120 | and generally a lot of the AS stuff is somewhat random,
01:13:47.520 | but you're using it in production applications
01:13:50.640 | that require certainty, right?
01:13:53.060 | Just as an example, if I promise my users
01:13:55.200 | this podcast or this segment will be 30 seconds,
01:13:57.360 | it needs to be 30 seconds.
01:13:59.200 | Or, you know, given some SLA.
01:14:00.600 | - Discontolerance.
01:14:01.680 | - Some SLA around like, you know, it's 95% of that.
01:14:04.280 | - Yeah.
01:14:05.120 | - But I think a lot of these things
01:14:07.680 | just tend to be a little random at the moment.
01:14:10.080 | So like how, can I literally specify a tone
01:14:13.320 | that I'd like this to be and be certain that it's doing it
01:14:15.520 | and it's not some weird like attempt at sounding surprised.
01:14:19.800 | - Yeah.
01:14:20.640 | - It's just like, yeah.
01:14:21.480 | Basically how controllable and how realistic they sound.
01:14:23.500 | - Yeah.
01:14:24.340 | And then final question around just the landscape of TTS.
01:14:28.560 | What are the unique challenges for non-English TTS?
01:14:31.120 | And I'll tell you, right?
01:14:32.240 | So I'm interested in having 28 languages of latent space.
01:14:35.440 | Right, that's only good things for me.
01:14:37.280 | Except if it sucks.
01:14:38.840 | - Mm-hmm.
01:14:39.720 | - And I obviously, I have no way to validate.
01:14:41.840 | - I think that's the problem with--
01:14:42.840 | - Latent space Ukraine.
01:14:44.520 | - Yeah, and I think that's the problem with dubbing,
01:14:46.000 | I think.
01:14:46.840 | So the reason, one thing we're gradually building out,
01:14:49.400 | but we already have as part of our dubbing product
01:14:50.920 | is that we have QA as part of that.
01:14:52.840 | So we actually work with professional translators
01:14:54.920 | to just make sure that the things that we publish--
01:14:56.440 | - Oh, nice.
01:14:57.280 | Oh, you should put that up front.
01:14:58.120 | - Yeah.
01:14:58.960 | So that's really one of the, like,
01:15:01.840 | fundamentally the problem with dubbing,
01:15:03.040 | if you ask anyone who's ever tried to dub,
01:15:04.400 | is you don't know what good sounds like
01:15:07.480 | in these other languages.
01:15:08.400 | You're like, I can tell you I dub,
01:15:09.720 | but I'm gonna tell you that--
01:15:10.800 | - Yeah.
01:15:11.640 | - I think there's a lot of big podcast studios
01:15:12.680 | who have tried this before.
01:15:13.520 | There's one I can think of that's tried this
01:15:14.760 | maybe five times with five different companies
01:15:16.560 | in the last five years.
01:15:18.200 | Their fundamental problem is that you just cannot,
01:15:21.400 | yeah, fine, Spanish sounds good to me
01:15:22.720 | as a person who doesn't speak Spanish,
01:15:23.840 | but like, it doesn't sound good to a Spanish person
01:15:26.160 | or an Argentine person who have totally different accents.
01:15:30.120 | Right?
01:15:30.960 | So I think the tricky thing as well
01:15:34.000 | is getting that straight is,
01:15:36.440 | in English, it's very clear,
01:15:37.840 | and there's a huge proliferation of content
01:15:39.920 | in every single accent, if you will,
01:15:42.680 | Australian, English, British, American, Irish, whatever.
01:15:46.920 | In other languages, especially if you consider the niche ones
01:15:50.200 | I don't know, I'll pick like German.
01:15:54.040 | There's like a few German accents.
01:15:56.440 | I don't know how much content Austria puts out
01:15:59.000 | 'cause they have a specific accent, right?
01:16:00.800 | Germany itself, probably every--
01:16:02.160 | - Swiss-German.
01:16:03.320 | - Yeah, just record Roger Federer speaking, right?
01:16:07.320 | So there's just a lot of nuance
01:16:09.560 | that isn't as widely available.
01:16:11.440 | The English market is just the biggest.
01:16:12.760 | - Yeah, yeah.
01:16:13.600 | - And I say this as someone like,
01:16:15.160 | so I say someone who speaks German,
01:16:16.320 | but like, think of all the other language,
01:16:18.760 | I have no idea what Spanish, Spanish-Spain,
01:16:22.000 | sorry, Spain-Spanish sounds like
01:16:24.280 | and Argentina-Spanish sounds like.
01:16:25.600 | There's a difference.
01:16:26.440 | I know that.
01:16:27.260 | I just don't know what it sounds like.
01:16:29.000 | - Cool.
01:16:29.840 | Well, if you ever need Chinese validation,
01:16:31.720 | I know I have some very fanatical Chinese listeners
01:16:36.040 | who translate every podcast.
01:16:37.440 | - Oh, that's amazing.
01:16:38.360 | - So we can use that as QA.
01:16:39.960 | - Yeah, I would definitely love that.
01:16:41.800 | - Shout out to the Chinese army.
01:16:43.560 | Great, great, awesome.
01:16:45.680 | What do you want to ask me as a podcaster?
01:16:48.360 | - So tell me about your setup process
01:16:50.720 | for setting up a podcast.
01:16:51.760 | - Oh my God.
01:16:52.600 | - And I think it goes far back as like,
01:16:54.800 | cool, you know what guests you're having on,
01:16:56.200 | you're just like, no, no.
01:16:57.680 | - Oh, you mean starting--
01:16:58.520 | - Even further back.
01:16:59.340 | Even further back.
01:17:00.180 | It's like, I want to talk about someone in this space.
01:17:03.280 | - So like for a new episode
01:17:04.600 | or for a completely new podcast?
01:17:05.880 | - New episode.
01:17:06.720 | - A new episode is pretty easy, right?
01:17:07.760 | Like the way I set it up with you
01:17:08.760 | was like, hey, I'm a fan.
01:17:09.960 | Do you want to chat?
01:17:10.800 | And you said yes.
01:17:11.840 | And we scheduled--
01:17:13.400 | - And how long did it take to get the scheduling right?
01:17:15.100 | - Scheduling, you guys are very responsive,
01:17:19.080 | very easy to work with.
01:17:20.440 | Some others are very hard.
01:17:22.040 | Like we took a few months for Harrison to come on.
01:17:24.320 | - Yeah, and you said that.
01:17:26.240 | - For you guys, it only took two weeks.
01:17:27.720 | - Okay.
01:17:28.880 | - And then, yeah, we got this studio.
01:17:32.040 | We came here and then, you know,
01:17:34.480 | all three of us spent 50 minutes
01:17:36.480 | setting up the audio for ourselves.
01:17:38.680 | In San Francisco, we pay someone to do it.
01:17:41.280 | And so we just kind of walk in and start recording.
01:17:44.160 | - Yeah, and how do you come up with the questions
01:17:45.840 | that you want to ask?
01:17:47.200 | - I do prep and research on you and the business.
01:17:50.320 | And usually it has a format of
01:17:52.000 | find information about the speaker
01:17:54.680 | and then get to know you on a personal basis,
01:17:57.080 | put you at ease, so that you'll be more open
01:17:59.360 | for the other stuff.
01:18:01.440 | 'Cause you know I care about you as a human being,
01:18:04.680 | not just as an interview subject.
01:18:06.640 | - Yeah.
01:18:08.000 | - Then talk about the business.
01:18:09.240 | I do emphasize the what is question.
01:18:11.840 | They're very straightforward.
01:18:12.680 | Like what the hell it is you're doing,
01:18:14.720 | because people want a clear answer
01:18:16.800 | and some people cannot give a clear answer.
01:18:18.920 | And that's also information.
01:18:20.320 | Then I always am interested about company building
01:18:23.880 | just 'cause I'm a founder myself, you know,
01:18:26.680 | and I want to learn about what it's like on your side
01:18:30.080 | and ask the business side of things as well.
01:18:32.640 | Some people don't have that side
01:18:34.400 | 'cause they're academics.
01:18:35.480 | So then you go more into their research
01:18:36.960 | and try to keep up with any academic in theory
01:18:41.760 | and math that they might bring up.
01:18:44.160 | And then maybe try to, at the end,
01:18:47.320 | and this is the hardest bit,
01:18:48.160 | and this is the stuff that I have
01:18:50.200 | the most second guessing about,
01:18:51.960 | try to have a segment at the end
01:18:53.720 | where you ask a consistent set of questions
01:18:55.760 | for essentially norming across guests.
01:18:59.880 | So I have a system prompt for our guests
01:19:04.560 | and at the end of it we basically end every episode
01:19:06.640 | with a lightning round, right?
01:19:07.480 | Acceleration, exploration, take away.
01:19:09.960 | I don't love the questions that we ask.
01:19:12.160 | We're still iterating on that.
01:19:13.800 | - You gotta experiment to do it.
01:19:15.040 | - Yeah, I think just people like Lean Space
01:19:18.080 | mostly because the kind of questions that I ask
01:19:21.960 | are the stuff that is on everybody's minds
01:19:24.040 | because I just talk to a lot of people.
01:19:26.080 | And so I just force you to answer them
01:19:28.120 | because you're right in front of me.
01:19:29.680 | - It's great.
01:19:30.520 | - It's great.
01:19:31.340 | And so you have, this is a whole interesting conversation
01:19:33.720 | 'cause you're a human podcaster, AI podcaster.
01:19:35.680 | - Yes.
01:19:36.720 | - So as a human podcaster and someone
01:19:39.160 | who was with a really popular show,
01:19:41.380 | and also someone who can actually implement
01:19:44.560 | this stuff himself,
01:19:46.280 | what is some of the AI tooling recently
01:19:49.880 | that you've baked into your processes?
01:19:52.880 | - I only use the script.
01:19:54.080 | And we built for editing.
01:19:58.320 | So for example, and by the way,
01:20:01.160 | this goes into a theory of content,
01:20:03.320 | which as a content creator myself,
01:20:04.680 | professionally and as an advisor, I have,
01:20:07.440 | which is that we developed a few show formats.
01:20:10.520 | Lean Space is a channel, it's kind of like a TV channel,
01:20:13.240 | and channels need different formats.
01:20:15.600 | So you have the reality TV show, you have the news show,
01:20:17.800 | you have the cooking show, whatever.
01:20:20.840 | For us, we have the founder interview, straightforward.
01:20:23.880 | Everyone has them.
01:20:25.000 | We have the breaking news, Twitter space.
01:20:27.480 | And that is, we want to be the day one first podcast
01:20:32.480 | to come up with the most in-depth breakdown
01:20:35.280 | of something that is that everybody needs to know.
01:20:38.200 | And that has high value to people, right?
01:20:40.160 | Because if you're a week delayed, one month delayed,
01:20:42.760 | then no one cares anymore.
01:20:44.840 | And then finally, we have the fundamentals,
01:20:46.480 | like the one-on-one evergreen episodes
01:20:48.160 | that are less time-bound.
01:20:49.520 | So this one is relatively time-bound
01:20:51.040 | because it's a snapshot of who you are right now.
01:20:53.240 | But we want to have evergreen episodes
01:20:55.320 | that people can go back two, three years in the backlog
01:20:58.800 | and still get value from.
01:21:00.280 | - Yeah, and these more fundamental ones.
01:21:01.800 | - Yes.
01:21:02.640 | So we have three show formats right now.
01:21:04.160 | And I would say we have different tooling for each, right?
01:21:08.680 | So the one that I don't need any tooling for, essentially,
01:21:13.680 | is the fundamentals one,
01:21:16.360 | because we plan basically every minute of that show.
01:21:19.760 | It is a lot of work.
01:21:21.000 | But it's high quality because people love it.
01:21:24.040 | It's got the longest tail by design, right?
01:21:26.280 | The Twitter spaces require Descript
01:21:33.320 | because a lot of silences and a lot of ums.
01:21:36.320 | And that's not good podcast audio.
01:21:38.400 | So you got to cut it out.
01:21:39.440 | - So you literally just go in,
01:21:41.240 | like you edit out the Twitter space that you did.
01:21:44.200 | You record it and then you edit it out.
01:21:45.440 | - Usually it's like two hours, we cut it down to one.
01:21:47.120 | - Okay.
01:21:48.720 | - And it's a lot of pain and a lot of work.
01:21:51.000 | But it's the only way that I get
01:21:52.400 | some pretty high-profile people onto my podcast
01:21:55.320 | without booking them.
01:21:56.280 | They just show up.
01:21:58.000 | And that has value to me, right?
01:21:59.440 | Simon Willison has been on my podcast three times
01:22:01.680 | and I never had to schedule him.
01:22:03.840 | And people love him.
01:22:04.920 | I mean, he's great.
01:22:05.920 | And then this one, basically we do,
01:22:11.920 | I don't need Descript, obviously,
01:22:13.680 | but we do use Small Podcaster,
01:22:16.480 | which is a 100-line Python script
01:22:20.320 | that throws the transcript into Anthropic
01:22:22.040 | and then generates show notes.
01:22:23.400 | - Nice.
01:22:24.240 | - So that's about it right now.
01:22:25.440 | - Nice, so it's interesting 'cause I think
01:22:28.080 | you're in a very nice position
01:22:29.440 | where you're able to do what a lot of these services
01:22:31.520 | charge you for, you can just do it yourself.
01:22:33.360 | - Yeah.
01:22:34.200 | - So it's an interesting one.
01:22:35.040 | - But obviously I'm interested in paying for things
01:22:37.000 | 'cause my time is valuable
01:22:38.400 | and if it does a good job, then I'll use it.
01:22:41.880 | For Wondercraft, the thing that I really wanted
01:22:44.760 | was the RSS to podcasting, right?
01:22:47.000 | Which you now have.
01:22:49.080 | So I'll try it out, but chances are
01:22:50.720 | I will not be happy with something.
01:22:52.360 | And so then the question is,
01:22:53.920 | how much customizability do you give me to do that?
01:22:57.760 | And we'll see.
01:22:59.680 | - Yeah, interesting, cool.
01:23:02.360 | These are all my questions.
01:23:03.200 | - I will say, well, you missed out one thing,
01:23:05.240 | which is marketing the podcast, which is a huge part.
01:23:08.880 | That is mostly my job.
01:23:10.080 | - So how do you market your podcast?
01:23:12.160 | - Twitter.
01:23:13.480 | - So you think Twitter's a good media for that?
01:23:14.760 | - Twitter and Happy News.
01:23:15.600 | - And threads, or you post clips, or what do you do?
01:23:20.080 | - I have tried posting clips.
01:23:22.120 | It's just too much work.
01:23:23.640 | So if you guys do a good job of clips,
01:23:25.020 | I will use your stuff, but it's just too much work.
01:23:27.440 | So mostly I just put a big post saying,
01:23:32.440 | so for our George Hots episode,
01:23:34.240 | we were like, "Latent Space is excited to present
01:23:36.080 | "George Hots on Tiny Corp and Commoditizing Petaflops."
01:23:40.240 | Something like that.
01:23:41.160 | And just sometimes the fame of the guest
01:23:44.280 | will just lead the episode.
01:23:46.500 | So the one I dropped yesterday was Chris Ladner.
01:23:50.300 | And people were like, "Chris Ladner's the boss.
01:23:52.060 | "I don't care about anything else.
01:23:53.000 | "Just I want to hear as many Chris Ladner tokens
01:23:55.980 | "as possible."
01:23:57.180 | Others who are less famous, I have to introduce
01:23:59.340 | who you are and why I care about you,
01:24:01.100 | why they should care about you,
01:24:02.620 | 'cause most people will not have heard about you
01:24:04.300 | as well if you've done.
01:24:06.220 | So then I need to make the case a little bit more.
01:24:08.740 | But that's fine, that's my job.
01:24:10.560 | I just think it takes a lot of work,
01:24:12.280 | and that's the part that will be hardest
01:24:14.880 | for me to hand over to AI,
01:24:16.580 | 'cause I have a very specific voice for myself.
01:24:20.660 | And apparently all AIs think that Twitter,
01:24:23.660 | to tweet you have to have emojis and hashtags,
01:24:26.300 | which is so dumb.
01:24:27.200 | It's so obviously dumb.
01:24:29.820 | The training data is very bad.
01:24:30.660 | - I think it depends, right?
01:24:31.500 | Sometimes you do, if you're trying to be more
01:24:32.820 | on the professional side, you don't.
01:24:34.740 | - Yeah.
01:24:35.580 | - It just depends on the tone and brand voice, I guess.
01:24:37.940 | - It's just cringe.
01:24:38.980 | And that's the problem with AI.
01:24:40.980 | AI is mid by design, right?
01:24:43.180 | And you do not want to be mid as a thought leader,
01:24:45.120 | as a content creator.
01:24:46.080 | People don't want to subscribe to mid things.
01:24:48.320 | They want to subscribe to a level
01:24:50.880 | slightly ahead of where they are.
01:24:52.440 | And so that's for better or worse where we target.
01:24:55.800 | - Which I guess the interesting thing for us
01:24:56.920 | is how do we really enable that you have your process
01:25:01.060 | and really bake that in to make sure
01:25:02.240 | that what you're presenting
01:25:03.080 | and what you're coming out with is unique.
01:25:04.720 | - Yeah.
01:25:05.540 | I'll tell you one thing, though.
01:25:06.960 | I'm happy to hand over my LinkedIn to you.
01:25:10.200 | I have a LinkedIn following, I don't use it.
01:25:12.520 | So completely money on the table there.
01:25:15.520 | So happy to give that to you.
01:25:17.820 | You don't do social media posting, right?
01:25:19.720 | So they have to go get like spot social or buffer,
01:25:22.480 | both of which are horrible.
01:25:24.240 | So if you have an adjacent product some day,
01:25:27.480 | you might want to get into,
01:25:29.000 | basically money on the table marketing
01:25:32.000 | is what I think about it.
01:25:33.780 | Basically, we would not do this by ourselves.
01:25:36.840 | It is still inferior to a human crafted thing,
01:25:40.120 | but it doesn't matter because I just,
01:25:41.760 | I'd have zero LinkedIn presence.
01:25:43.300 | - Makes sense.
01:25:45.240 | - Yeah.
01:25:46.080 | - Great answers.
01:25:46.920 | (laughs)
01:25:47.760 | - Obviously happy to offer any thoughts
01:25:50.080 | as you build out for podcasters.
01:25:53.120 | Okay, well, shall we go into the lightning round?
01:25:54.640 | - Let's do it.
01:25:55.480 | - Yeah, the one I just said that I wasn't confident about.
01:25:57.680 | (laughs)
01:25:58.520 | - Great intro.
01:25:59.360 | - Yeah, yeah, yeah.
01:26:00.180 | But these came from previous guests.
01:26:02.160 | So I just think we haven't found PMF yet on those things.
01:26:06.040 | So acceleration, what has already happened in AI
01:26:08.600 | that you thought would take much longer?
01:26:11.520 | - The GPT, the fact that basically you can do anything.
01:26:16.520 | Obviously, it might not be the smartest way to do it,
01:26:18.800 | but you can basically do anything with GPT.
01:26:21.560 | - And were you watching it while you were at Palantir?
01:26:23.760 | 'Cause Palantir has so many high stuff.
01:26:25.280 | - Yeah, I think,
01:26:26.360 | vaguely, I think it really blew up after.
01:26:32.240 | - Yeah.
01:26:33.080 | - I think Palantir really--
01:26:33.920 | - You just happened to be the right spot
01:26:35.120 | where Moonshot wasn't super working.
01:26:37.440 | - Yeah.
01:26:38.280 | - So you were open enough to actually pursue this,
01:26:40.600 | seriously.
01:26:41.440 | - Yeah, and I think Palantir also just shifted
01:26:43.400 | the whole strategy once it blew up the way it did.
01:26:45.600 | - Yeah, yeah, awesome.
01:26:47.120 | Exploration, if you weren't doing what you're doing,
01:26:49.480 | what do you think is the most interesting,
01:26:50.840 | unsolved question in AI?
01:26:52.080 | - Hmm.
01:26:54.560 | Great question.
01:27:02.760 | There's so many things.
01:27:04.240 | - Like what?
01:27:05.240 | - There's the context length, there's just the--
01:27:08.480 | - You want longer context length?
01:27:09.760 | - The latency and how to just make everything
01:27:12.280 | flow through much quicker,
01:27:13.240 | 'cause I think you can then enable
01:27:14.480 | some really interesting real-time applications.
01:27:16.400 | - Are you interested in real-time?
01:27:17.360 | Actually, I was meaning to think about that.
01:27:19.360 | - Not necessarily.
01:27:20.680 | - Right, because your whole thing is async, right?
01:27:22.500 | - Maybe, yeah, but maybe in the future
01:27:24.080 | for some other product.
01:27:25.120 | But as a person, you know, it's interesting.
01:27:28.020 | And also just the ability to,
01:27:31.600 | really the interesting stuff is the validation, right?
01:27:36.600 | I actually think it's the QA that we were talking about.
01:27:38.680 | - We need evals, yeah, yeah.
01:27:39.840 | - You still need human QA.
01:27:43.080 | - Yeah, especially yours is the hardest to QA.
01:27:45.280 | - Yeah.
01:27:46.120 | - I would say.
01:27:46.940 | Okay, very cool.
01:27:47.780 | - It's very time-consuming.
01:27:48.620 | - Takeaway, what is one message you want
01:27:50.520 | all of our listeners to remember?
01:27:52.280 | Take away with them.
01:27:53.280 | - Okay, if you would like to start a podcast, start.
01:27:59.880 | We're here to help.
01:28:01.240 | Super easy.
01:28:02.160 | If you have a podcast, we wanna help you
01:28:04.100 | make it more accessible by dubbing it.
01:28:08.140 | On the other side, if you are a founder, an AI engineer,
01:28:11.440 | I think it's really important to convince yourself
01:28:13.220 | that what you're building is valuable.
01:28:15.220 | Don't listen to people saying, I have a motor,
01:28:17.760 | you don't have a motor.
01:28:18.600 | Convince yourself of what that is and launch.
01:28:22.320 | Launch and don't burn that much money.
01:28:24.240 | - Frequently and often, don't spend your money.
01:28:26.000 | - Yeah, yeah, be smart about it.
01:28:28.320 | - Yeah, I think you are one of the most successful cases
01:28:31.940 | of AI engineers so far.
01:28:34.440 | I'm really glad to spend time with you in person
01:28:37.520 | and excited to see what comes next.
01:28:39.300 | - Yeah, it was great coming here.
01:28:40.620 | Great meeting you guys in London.
01:28:41.880 | - Yeah.
01:28:42.720 | - And see you soon.
01:28:43.760 | - All right, see you.
01:28:45.120 | - Bye.
01:28:46.040 | (laughs)
01:28:47.360 | - How was that?
01:28:48.200 | - Great, man.
01:28:49.040 | Hopefully you enjoyed it.
01:28:51.880 | - Very much, very much.
01:28:52.720 | It was a good conversation.