53: AI Profile Pic - Rony Fadel

Transcript

Autogenerated by MacWhisper

00:00:00.000 --> 00:00:06.920

Welcome to Launched. I'm Charlie Chapman, and today I'm excited to bring you the developer

00:00:06.920 --> 00:00:19.520

behind the AI profile pick app, Ronnie Fidel.

00:00:19.520 --> 00:00:25.680

Ronnie, welcome to the show.

00:00:25.680 --> 00:00:27.760

Hi, Charlie. Thank you for having me.

00:00:27.760 --> 00:00:28.760

Thanks for coming on.

00:00:28.760 --> 00:00:32.940

This is a we actually we talked a month or so ago when we were talking about doing the

00:00:32.940 --> 00:00:37.420

show but scheduling wise a little peek behind the curtain.

00:00:37.420 --> 00:00:41.120

We're like sort of dancing around holidays and vacations and stuff.

00:00:41.120 --> 00:00:44.580

This AI category is that what you call it?

00:00:44.580 --> 00:00:48.660

Like you know the chat GBT as well as all the stable diffusion stuff.

00:00:48.660 --> 00:00:49.660

Do you just call that AI?

00:00:49.660 --> 00:00:51.020

Is there a name for this?

00:00:51.020 --> 00:00:53.900

I guess you'd want to call it maybe generative AI.

00:00:53.900 --> 00:00:54.900

Generative.

00:00:54.900 --> 00:01:00.020

AI is like, it feels like an older and more broad category than these things.

00:01:00.020 --> 00:01:01.620

Mm-hmm. Yeah.

00:01:01.620 --> 00:01:07.540

But like this space is moving so ridiculously fast that I didn't want to record early like

00:01:07.540 --> 00:01:11.140

I often do for this show. So, we're kind of, this is the first one after break.

00:01:11.140 --> 00:01:13.380

So, I'm excited to have you on.

00:01:13.380 --> 00:01:18.180

Yeah, it's been wild, right? The behind the curtains thing is that we talked about this

00:01:18.180 --> 00:01:23.460

a month ago and it's still going fast. It's still changing on a weekly basis.

00:01:24.500 --> 00:01:30.980

I feel like everyone's just keeps on catching up to whatever the new thing is.

00:01:30.980 --> 00:01:37.060

Yeah, it's been wild. And I think it will continue on being wild for the foreseeable future.

00:01:37.060 --> 00:01:41.780

Before we get into that, and I have like a million questions in this area,

00:01:41.780 --> 00:01:45.140

I want everyone to get a little bit of a background on who you are. So

00:01:45.140 --> 00:01:50.500

the questions I asked to start this off is, where are you from? Do you have a formal education

00:01:50.500 --> 00:01:55.860

related to what you do? And then what was your career like that led up to this sort of generative

00:01:55.860 --> 00:02:01.140

AI space that you've been playing in? I'm from Lebanon. I'm Lebanese, born and raised.

00:02:01.140 --> 00:02:10.900

And then I went to France for higher education, right for uni. So I studied software engineering

00:02:10.900 --> 00:02:19.300

there. And right out of school, I joined Apple, first on Mac OS, and then on iOS. So I worked

00:02:19.300 --> 00:02:24.660

at Apple for about six years as part of the application frameworks team.

00:02:24.660 --> 00:02:33.060

So I worked on Spotlight, sharing iCloud, iCloud Sync and iCloud Drive,

00:02:33.060 --> 00:02:38.740

so which became the Files app. So that's kind of my career path at Apple.

00:02:38.740 --> 00:02:44.180

And at some point, I was like, it would be great to, you know, explore the world

00:02:45.060 --> 00:02:52.180

and kind of switch it up career-wise. I knew that I wanted to leave the corporate space

00:02:52.180 --> 00:02:57.300

for a little while just to kind of experiment with my own things, develop my own products,

00:02:57.300 --> 00:03:04.500

really own a product from A to Z, be it development, design, marketing, and just

00:03:04.500 --> 00:03:12.580

learn as much as possible. That's what I had done. I left Paris in 2019, so that was three,

00:03:12.580 --> 00:03:21.300

three and a half years ago and set out to just kind of YOLO, like, you know, figure it out,

00:03:21.300 --> 00:03:28.900

right? And I started developing smaller apps. So there's Mission Control Plus, an app that helps

00:03:28.900 --> 00:03:34.980

you just take control of all of the windows that show up in Mission Control, being able to... so

00:03:34.980 --> 00:03:41.140

it's kind of an OS tweak. And the Batteries app, which tracks all of your device's batteries. And

00:03:41.140 --> 00:03:46.980

And these were kind of hits, even though they were kind of very small, targeted, very specific

00:03:46.980 --> 00:03:47.980

apps.

00:03:47.980 --> 00:03:48.980

Yeah.

00:03:48.980 --> 00:03:50.820

And on the Mac too, right?

00:03:50.820 --> 00:03:57.060

Were those in the Mac App Store or they like separately deployed or distributed?

00:03:57.060 --> 00:04:00.860

They both use private API.

00:04:00.860 --> 00:04:01.980

Ah, okay.

00:04:01.980 --> 00:04:04.260

The choice is made for you then.

00:04:04.260 --> 00:04:05.260

Choice was made for me.

00:04:05.260 --> 00:04:07.980

So I can't sell them on the App Store.

00:04:07.980 --> 00:04:11.260

So yeah, I sell them off of my own website.

00:04:11.260 --> 00:04:13.500

How did you, how did those end up being hits?

00:04:13.500 --> 00:04:15.820

They weren't through, you know, the Mac App Store.

00:04:15.820 --> 00:04:18.300

What was your sort of marketing?

00:04:18.300 --> 00:04:23.180

Even before talking about marketing, I, why did I want to create these apps?

00:04:23.180 --> 00:04:24.180

Right.

00:04:24.180 --> 00:04:25.660

It kind of started off that way.

00:04:25.660 --> 00:04:30.340

It's like it had been these two problems solved by Mission Control Plus and batteries had been

00:04:30.340 --> 00:04:32.540

nagging me for four years.

00:04:32.540 --> 00:04:34.220

And so, so they were on my list.

00:04:34.220 --> 00:04:37.460

It's like, I would love Mac OS to do this someday.

00:04:37.460 --> 00:04:42.820

And it became a nagging issue in the sense that like on almost, you know, this is annoying.

00:04:42.820 --> 00:04:43.820

I needed them.

00:04:43.820 --> 00:04:45.740

I needed them both of them.

00:04:45.740 --> 00:04:49.100

And so that's how I, you know, thought of creating them.

00:04:49.100 --> 00:04:50.740

That's how I developed them.

00:04:50.740 --> 00:04:55.580

And it just turned out that there were so many people that had these same two nagging

00:04:55.580 --> 00:04:56.580

issues.

00:04:56.580 --> 00:05:00.900

So I imagine I kind of lucked out on it in the sense that there were so many other people

00:05:00.900 --> 00:05:03.540

that had that same nagging issue.

00:05:03.540 --> 00:05:05.440

And that kind of helped marketing.

00:05:05.440 --> 00:05:06.900

It helped it quite a bit.

00:05:06.900 --> 00:05:12.700

I only launched on Product Hunt, I think, and I didn't have that many Twitter

00:05:12.700 --> 00:05:17.340

followers or Twitter audience, but it just seemed like these issues were

00:05:17.340 --> 00:05:20.860

nagging enough so that people, it clicked to people.

00:05:20.860 --> 00:05:21.660

Nice.

00:05:21.660 --> 00:05:22.420

Well, that's cool.

00:05:22.420 --> 00:05:25.580

And those are still, those are still in the store too, or not in the store, but

00:05:25.580 --> 00:05:27.220

those are still available on your website, right?

00:05:27.220 --> 00:05:28.580

They're still available.

00:05:28.580 --> 00:05:30.220

They're, they're still up to date.

00:05:30.220 --> 00:05:30.620

Yeah.

00:05:30.620 --> 00:05:31.340

Okay.

00:05:31.340 --> 00:05:35.100

So you have, you have some Mac utility apps.

00:05:35.700 --> 00:05:43.700

How did we get from that to not a Mac utility app, an iOS app built around this generative

00:05:43.700 --> 00:05:47.060

AI art sort of space?

00:05:47.060 --> 00:05:48.540

What's the path there?

00:05:48.540 --> 00:05:55.780

After batteries came out, I was sort of in exploratory mode and wanting to play around

00:05:55.780 --> 00:06:02.820

with different technologies, things that I hadn't touched before, doing some web dev,

00:06:02.820 --> 00:06:10.320

Electron, right, just to kind of try something new and trying to make some kind of product

00:06:10.320 --> 00:06:12.260

or business out of it.

00:06:12.260 --> 00:06:14.740

And that didn't really pan out.

00:06:14.740 --> 00:06:19.680

Quite interestingly, if you're ideating and trying to launch a new product, do stick to

00:06:19.680 --> 00:06:20.680

what you know.

00:06:20.680 --> 00:06:22.940

Do not try to learn a new technology at the same time.

00:06:22.940 --> 00:06:24.580

Like that will slow you down.

00:06:24.580 --> 00:06:26.540

And then the pandemic hit.

00:06:26.540 --> 00:06:33.720

And at that point I started contracting for Snap, Snapchat, for a while and that kind

00:06:33.720 --> 00:06:37.040

of dove me back into the iOS world.

00:06:37.040 --> 00:06:40.520

So kind of steered away from Mac OS, went back into iOS.

00:06:40.520 --> 00:06:42.980

And I did that for about one year full time.

00:06:42.980 --> 00:06:47.400

So stopped focusing on the indie side of things.

00:06:47.400 --> 00:06:52.680

After that one year I had decided, okay, like I'm going to dive back into the indie world,

00:06:52.680 --> 00:06:57.240

But this time it's going to be kind of the more straightforward, streamlined way of

00:06:57.240 --> 00:07:03.760

developing, meaning back into the App Store, focusing on developing something for iOS.

00:07:03.760 --> 00:07:04.880

Because the market is there.

00:07:04.880 --> 00:07:10.560

The Mac App Store market felt a bit smaller, and it just made more sense to develop something

00:07:10.560 --> 00:07:11.840

for iOS.

00:07:11.840 --> 00:07:18.140

And so I set out to develop a kind of a small portfolio of apps, of editing apps.

00:07:18.140 --> 00:07:28.340

So a interactive audio editor where, so PodBuddy is this app that transforms your audio content,

00:07:28.340 --> 00:07:34.220

let's say a podcast clip or a music clip, into kind of an Instagram story-like UI that

00:07:34.220 --> 00:07:39.380

you can share on a video that you can share on social media.

00:07:39.380 --> 00:07:44.980

And after that, I developed VidCap, which is a captioning app for videos.

00:07:44.980 --> 00:07:50.580

I had set out to develop these because I had always been interested in designing and making

00:07:50.580 --> 00:07:58.980

editing UI or editing tools. So anything that you can add media to sort of canvas.

00:07:58.980 --> 00:08:04.580

For some reason, I had always been interested as a kind of experiment to just develop something

00:08:04.580 --> 00:08:11.300

like that. So I set out to make these two apps, put them on the iOS App Store. And

00:08:12.420 --> 00:08:19.620

I started also, you know, I got in contact with iOS marketing on the App Store.

00:08:19.620 --> 00:08:26.820

I got in contact with marketing on the App Store in the sense that, you know, ASO, knowing how to

00:08:26.820 --> 00:08:38.740

monetize any an app on the App Store and kind of just becoming more in sync with how to develop an

00:08:38.740 --> 00:08:39.740

an app store business.

00:08:39.740 --> 00:08:43.300

Like you were digging, you were digging into those and becoming interested in

00:08:43.300 --> 00:08:45.020

like growing that skill. Is that what you're saying?

00:08:45.020 --> 00:08:51.020

Exactly. So it's kind of a how to create products that delight me and others,

00:08:51.020 --> 00:08:53.140

but also how to create a business out of it.

00:08:53.140 --> 00:08:53.580

Yeah.

00:08:53.580 --> 00:08:57.740

And so I became more tied in with the with the app store business in general,

00:08:57.740 --> 00:09:01.900

right? It is its own ecosystem. And I thought that there was opportunity there.

00:09:01.900 --> 00:09:07.140

And so I just got into that. At that point, I started contracting for Mighty,

00:09:08.020 --> 00:09:12.820

which was a web browser that is based in the cloud,

00:09:12.820 --> 00:09:15.780

that is streamed onto your machine.

00:09:15.780 --> 00:09:20.020

So all the rendering, all the heavy lifting is done on the cloud for you.

00:09:20.020 --> 00:09:24.660

And then what you receive is just a video stream of that rendered content.

00:09:24.660 --> 00:09:30.020

I did a lot of prototyping and kind of ideating for what the browser of the future

00:09:30.020 --> 00:09:32.260

could look like at Mighty.

00:09:32.260 --> 00:09:34.500

And at some point, there was kind of a,

00:09:35.300 --> 00:09:41.300

you know, we had finished all of the prototyping tasks, we still had a few, but this generative AI

00:09:41.300 --> 00:09:47.540

thing started becoming, you know, more, let's say, present, right? We started seeing more of it on

00:09:47.540 --> 00:09:54.020

Twitter and tech circles, etc. And just playing around with these tools at the time, the CEO was

00:09:54.020 --> 00:10:00.740

like, it will be interesting to explore something in this space, because this tech is, it's something

00:10:00.740 --> 00:10:04.820

else. We haven't seen anything like that in such a long time, like this could be something big.

00:10:04.820 --> 00:10:12.240

And so from prototyping web browsers, we started prototyping, testing out these

00:10:12.240 --> 00:10:17.980

generative AI art APIs, call them, right? We started playing around with stable

00:10:17.980 --> 00:10:23.740

diffusion and thinking how can we make interesting tools based on that. And that

00:10:23.740 --> 00:10:30.420

became Playground AI, which you can visit at playgroundai.com. It is a bit of a

00:10:30.420 --> 00:10:35.980

generative AI tool slash social network where you can see other people's

00:10:35.980 --> 00:10:41.420

creations, can remix other people's generated AI work, etc.

00:10:41.420 --> 00:10:44.420

That's right. Okay, I have seen this before.

00:10:44.420 --> 00:10:48.020

And at that point I had felt like I had swallowed the red pill.

00:10:48.020 --> 00:10:52.820

I was bitten by the AI bug. I was all in on this. I was like,

00:10:52.820 --> 00:10:55.620

this is going to change everything.

00:10:55.620 --> 00:11:00.620

And at that point, I had finished or stopped my contract with Mighty.

00:11:00.620 --> 00:11:06.180

Just to kind of explore how I can solve problems using AI.

00:11:06.180 --> 00:11:09.780

It's kind of like everything became a nail

00:11:09.780 --> 00:11:13.540

and AI was the hammer that I was wielding.

00:11:13.540 --> 00:11:15.740

It kind of did.

00:11:15.740 --> 00:11:18.900

But I did feel like, oh my God, this is so transformative.

00:11:18.900 --> 00:11:23.460

This will change so many things that I better jump on the bandwagon.

00:11:23.460 --> 00:11:28.060

There is a hype around this, like, obviously, I won't hide this fact.

00:11:28.060 --> 00:11:30.860

But I was, yeah, so

00:11:30.860 --> 00:11:35.100

deep into this, I was like, OK, let's let's start making products around this.

00:11:35.100 --> 00:11:41.300

We've been living through a decade, I guess, of Web3 and crypto

00:11:41.300 --> 00:11:44.100

kind of being the.

00:11:44.100 --> 00:11:47.540

The red pill, like this is

00:11:47.540 --> 00:11:51.020

everything's a nail because I'm holding this hammer and everybody was searching

00:11:51.020 --> 00:11:54.980

for how do I use this tech that's cool on something and everybody's just

00:11:54.980 --> 00:12:00.220

trying to come up with products and the AI thing or generative AI thing feels

00:12:00.220 --> 00:12:06.120

like it started with obvious products, but then you can imagine a thousand

00:12:06.120 --> 00:12:11.720

more, like it just seems like there's a much more obvious fertile ground for

00:12:11.720 --> 00:12:15.580

actual products that people would want in this space.

00:12:15.580 --> 00:12:19.820

And I def I share a similar feeling as you, where I can just imagine like a

00:12:19.820 --> 00:12:21.100

thousand possibilities.

00:12:21.100 --> 00:12:23.820

Like if the tech gets a little bit better, imagine this and imagine that.

00:12:23.820 --> 00:12:29.260

I feel the interesting thing about the current state of generative AI is that

00:12:29.260 --> 00:12:37.180

what we got with DALI APIs, such as DALI stable diffusion, is like going from

00:12:37.180 --> 00:12:40.620

from nothing to something so fast.

00:12:40.620 --> 00:12:41.020

Yeah.

00:12:41.020 --> 00:12:42.340

It just felt like a shock.

00:12:42.340 --> 00:12:44.820

We immediately got good enough.

00:12:45.820 --> 00:12:50.020

Uh, instead of, and maybe that's why the whole world is still reacting to this.

00:12:50.020 --> 00:12:53.900

Like we got from zero to good enough so fast to compare it with.

00:12:53.900 --> 00:12:59.720

Web3 I kind of agree is like, I was never bitten by that bug.

00:12:59.720 --> 00:13:04.960

I, it just always felt like maybe it's my lack of imagination, but I could not

00:13:04.960 --> 00:13:10.660

conceive of an interesting application to make out of all of these web3 technologies.

00:13:10.660 --> 00:13:11.800

Yeah.

00:13:12.020 --> 00:13:18.100

somebody who doesn't live in the world of finance or investment vehicles or whatever,

00:13:18.100 --> 00:13:22.660

yeah, it wasn't like I looked at that and thought, "Oh my gosh, the possibilities."

00:13:22.660 --> 00:13:26.740

And so admittedly, I'm in the category of people that it's easy to dunk on.

00:13:26.740 --> 00:13:32.260

I know I'm in the space where I don't know it well enough to really make fun of it as much

00:13:32.260 --> 00:13:40.420

as maybe I do. But I do live in being bad at writing copy and being okay at illustrating

00:13:40.420 --> 00:13:46.340

things. And so these generative AI, you know, ideas that have come out so far,

00:13:46.340 --> 00:13:52.580

it's like, yes, I imagine this tool built into Photoshop and yes, I can

00:13:52.580 --> 00:13:55.300

immediately see where I would utilize this.

00:13:55.300 --> 00:13:58.020

And, you know, people are already doing that all over the place.

00:13:58.020 --> 00:14:04.220

A hundred percent. It's just so amazing that these tools, just one tool, right?

00:14:04.220 --> 00:14:08.500

Taking just stable diffusion, for example, it's like there's this one tool that

00:14:08.500 --> 00:14:15.700

you can prompt that has so many applications, whereas back in the day, you had to form a

00:14:15.700 --> 00:14:20.000

team to solve each of the problems that this thing could solve.

00:14:20.000 --> 00:14:26.060

For example, I don't know, remove a background or outpaint an image, like add, you know,

00:14:26.060 --> 00:14:29.140

around the borders, add more content, etc.

00:14:29.140 --> 00:14:32.980

Adobe had, you know, teams working around the clock just for each of these features.

00:14:32.980 --> 00:14:38.420

And then we, you know, one API lands that can do all of these and do them quite well.

00:14:38.420 --> 00:14:45.220

a good way of phrasing it. Because every Adobe Max has been somebody in basically a white lab

00:14:45.220 --> 00:14:52.980

coat coming up and showing off the latest content background removal that you can run in After

00:14:52.980 --> 00:14:57.620

Effects. And look, it can work on video. But it's like they're opening up a script and running a

00:14:57.620 --> 00:15:01.380

thing and you're just like, "Wow, in 10 years or 5 years, this is going to be amazing."

00:15:01.940 --> 00:15:09.700

And everything with stable diffusion in DALI, it feels like, "Wow, next week, we can now do XYZ."

00:15:09.700 --> 00:15:18.260

Like the pace of improvement has been ridiculous, which is I think part of what adds to that feeling

00:15:18.260 --> 00:15:25.300

of excitement. I feel like every time excitement, worldwide excitement starts waning, some new

00:15:25.300 --> 00:15:31.460

advancement in AI just lands and it just renews, gives everyone that dopamine rush.

00:15:31.460 --> 00:15:33.460

And it keeps on going.

00:15:33.460 --> 00:15:36.460

Yeah, that's a good comparison.

00:15:36.460 --> 00:15:41.460

Before we get into the product that you then built through this excitement,

00:15:41.460 --> 00:15:46.460

I think we should address some of the controversy around it in terms of

00:15:46.460 --> 00:15:50.460

legally, everything about this is still murky.

00:15:50.460 --> 00:15:54.460

And I don't mean that in a "is this shady or not?"

00:15:54.460 --> 00:15:57.460

I think everybody's pretty upfront with exactly what's going on.

00:15:57.460 --> 00:16:03.420

on. It's just that the laws in every country are still catching up because of how fast

00:16:03.420 --> 00:16:06.540

this is going and like what your take on that is. I know we've talked about this a little

00:16:06.540 --> 00:16:10.180

bit, but hey, that was a couple of weeks ago and like we talked about then everything will

00:16:10.180 --> 00:16:14.020

probably change by the time we record. So I don't want to like not address that though.

00:16:14.020 --> 00:16:21.060

Oh, for sure. The funny thing is like a month ago we had briefly discussed this and what

00:16:21.060 --> 00:16:26.840

I had said last month is like my view on this. And again, I am not a lawyer, so maybe I'd

00:16:26.840 --> 00:16:31.760

like to more talk about the ethics of things more than more than the legality.

00:16:31.760 --> 00:16:33.120

Right. Yes. Agreed.

00:16:33.120 --> 00:16:37.320

Yeah, it was back then it was still evolving on a weekly basis just to kind

00:16:37.320 --> 00:16:41.180

of also get as many data points as possible from smart people around the

00:16:41.180 --> 00:16:43.160

world. And it's still ongoing now.

00:16:43.160 --> 00:16:47.480

My viewpoint is really is still evolving.

00:16:47.480 --> 00:16:53.280

But one thing that I'm kind of still fixated on, to be honest, is

00:16:54.200 --> 00:16:57.840

How does a human being get inspiration?

00:16:57.840 --> 00:17:01.120

How does a human being learn how to create art?

00:17:01.120 --> 00:17:05.440

It is never created in a vac or in a bubble.

00:17:05.440 --> 00:17:10.320

It's always born out of inspiration from other artists.

00:17:10.320 --> 00:17:12.040

Obviously, and again,

00:17:12.040 --> 00:17:14.160

if we're just taking the human being example,

00:17:14.160 --> 00:17:16.960

like a human artist, if it looks copied,

00:17:16.960 --> 00:17:19.980

if it seems, if it's obvious that it's plagiarized,

00:17:19.980 --> 00:17:23.420

then that is considered unethical.

00:17:23.420 --> 00:17:28.820

But there's just no way that you can create art without having seen art.

00:17:28.820 --> 00:17:34.040

There's no way you can just, you know, create amazing masterpieces without

00:17:34.040 --> 00:17:38.180

seeing, without learning how art works from other great artists.

00:17:38.180 --> 00:17:38.760

Right.

00:17:38.760 --> 00:17:41.620

I still haven't fully formalized that because what, how do you

00:17:41.620 --> 00:17:43.040

translate that to a machine?

00:17:43.040 --> 00:17:49.640

Like how do you teach a machine to just be inspired by art, teach it what art is.

00:17:49.780 --> 00:17:54.220

and actually tell it, this is great art and this is not,

00:17:54.220 --> 00:17:57.500

and then expect it to produce some kind of art.

00:17:57.500 --> 00:17:59.860

My viewpoint has not fully matured on that.

00:17:59.860 --> 00:18:02.300

It's not black and white.

00:18:02.300 --> 00:18:04.780

If it were easy, then it would be more obvious

00:18:04.780 --> 00:18:07.580

in the sense that we do want,

00:18:07.580 --> 00:18:09.180

like technology is moving forward,

00:18:09.180 --> 00:18:10.900

we do want machines to be able

00:18:10.900 --> 00:18:13.180

to create illustrations and art,

00:18:13.180 --> 00:18:15.020

but we don't want machines

00:18:15.020 --> 00:18:18.980

to basically steal art from artists.

00:18:18.980 --> 00:18:20.580

And what is the healthy middle?

00:18:20.580 --> 00:18:22.480

I'm still not sure.

00:18:22.480 --> 00:18:23.540

This is evolving.

00:18:23.540 --> 00:18:28.040

This, this whole thing did not exist at least to me six months ago.

00:18:28.040 --> 00:18:31.540

Uh, so yeah, it's, it's, it's not an easy one.

00:18:31.540 --> 00:18:38.300

There's sort of explicitly not a fine line between like plagiarism and

00:18:38.300 --> 00:18:40.760

stealing something and inspiration.

00:18:40.760 --> 00:18:46.700

It is like very fuzzy and everybody's going to have a different impression

00:18:46.700 --> 00:18:48.060

of that, of where that line is.

00:18:48.060 --> 00:18:53.980

yeah, how do you define that at a machine level is difficult. The comparison I keep running to

00:18:53.980 --> 00:19:02.860

is it feels like we're in the same beginning era as the Napster days with MP3s.

00:19:02.860 --> 00:19:07.420

Another example of one that went the other way is the VHS days. When VHS first came out,

00:19:07.420 --> 00:19:11.580

it was like, "Yeah, people should be able to record whatever's on their TV." And it's like,

00:19:11.580 --> 00:19:17.260

You can imagine during that time, that seeming obviously wrong. Like, "Well, you don't own

00:19:17.260 --> 00:19:22.700

that. The corporation owns that." And now nobody gives a second thought to like,

00:19:22.700 --> 00:19:26.380

"Yes, of course, you should be able to record things with a VHS."

00:19:26.380 --> 00:19:29.900

Well, I guess maybe now that's a bad example because with Hulu and all those,

00:19:29.900 --> 00:19:34.540

it kind of reversed on us. But you get the idea. We kind of went a certain direction and the whole

00:19:34.540 --> 00:19:40.940

world and industry adapted to that. And then Napster with MP3s was the same thing. It was like,

00:19:40.940 --> 00:19:46.620

nobody knew this is illegal because it wasn't illegal. It was being figured out.

00:19:46.620 --> 00:19:51.900

And it sort of feels like we're in that phase again, where we have a tool that is obviously

00:19:51.900 --> 00:19:59.900

has an extremely high utility. And how do we manage that? And it's going to have to be figured

00:19:59.900 --> 00:20:06.460

out. At least that's where I'm landing with it. It could be interesting to give back to artists

00:20:07.660 --> 00:20:17.900

whose works have made it basically into the model, right? So, I'm guessing there's a few artists

00:20:17.900 --> 00:20:24.700

definitely gonna botch the name but I think it's Greg Rutkowski. That name has been seen in like

00:20:24.700 --> 00:20:33.740

every stable diffusion prompt available basically because he is an artist on ArtStation and his

00:20:33.740 --> 00:20:43.340

style is somehow, I don't know, sexy for stable diffusion, it just works. And so, "Buy Greg

00:20:43.340 --> 00:20:50.300

Rutkowski" just appears in so many prompts. And it would almost be unfair to have Greg Rutkowski

00:20:50.300 --> 00:20:55.180

producing art for the world being reused by every application based on stable diffusion

00:20:55.180 --> 00:21:01.980

and not having the artist, I'm guessing, maybe paid back, but at least getting some sort form

00:21:01.980 --> 00:21:06.220

of attribution, right? That's the minimum. And there are some platforms that already give

00:21:06.220 --> 00:21:12.300

attribution. If the prompt has the artist's name, they get attribution, they get a link back. I'm

00:21:12.300 --> 00:21:17.180

not saying this is enough, but this is a start. I don't know enough about it, but that's where

00:21:17.180 --> 00:21:22.860

it starts feeling weird is when it's like, it is explicitly from this one person that you're

00:21:22.860 --> 00:21:30.460

generating something, right? And I know, I don't want to go too far into this because I'm pretty

00:21:30.460 --> 00:21:35.900

ignorant of it, but like stable, was it stable diffusion to the added the, the feature where

00:21:35.900 --> 00:21:40.780

you can basically opt out of being included in the model, or is that the next model that

00:21:40.780 --> 00:21:47.500

they're going to build? I believe it was art station that added an option to be up to opt

00:21:47.500 --> 00:21:54.860

out of being included and stable diffusion. Oh, like it's specific, uh, databases basically.

00:21:54.860 --> 00:21:59.660

And artists are also angry at, uh, at this option because I think it's opt out. So I

00:21:59.660 --> 00:22:03.420

I think it's opt in by default and you have to opt out manually.

00:22:03.420 --> 00:22:09.920

It's also about understanding what solution kind of serves everyone equally,

00:22:09.920 --> 00:22:15.820

what feels fair. Is it a Spotify like model where if you actually use artwork

00:22:15.820 --> 00:22:21.420

from a specific artist, they get paid back somehow, they get some form of royalty?

00:22:21.420 --> 00:22:25.940

Interestingly, that would change the whole economics out of it.

00:22:25.940 --> 00:22:32.380

Because currently you might as well treat it as stolen art and it's all baked into this

00:22:32.380 --> 00:22:40.780

brain that is the stable diffusion V1.5 model and you can hash and rehash this art as much

00:22:40.780 --> 00:22:44.940

as you want and just make it your own, right?

00:22:44.940 --> 00:22:47.540

Into your own creations out of a prompt.

00:22:47.540 --> 00:22:52.140

It could feel unfair for sure, but currently it's almost for free, right?

00:22:52.140 --> 00:22:55.640

I mean, it's just a GPU cost.

00:22:55.640 --> 00:23:05.520

What would it become if you start having to pay a small fraction of royalty to every artist

00:23:05.520 --> 00:23:07.140

whenever you generate the prompt?

00:23:07.140 --> 00:23:10.280

And how do you distribute that fairly?

00:23:10.280 --> 00:23:12.560

Could be an interesting problem to solve.

00:23:12.560 --> 00:23:16.380

It's really, I don't know, the whole space, both on the product side, like we were sort

00:23:16.380 --> 00:23:21.960

of fawning over a second ago, and then from a sort of ethics legal side, I find it all

00:23:21.960 --> 00:23:24.120

really fascinating as an outside observer.

00:23:24.120 --> 00:23:31.040

I think we're in that phase where both technologically and I guess legally, everything's getting

00:23:31.040 --> 00:23:34.800

hashed out that feels like it's going to drive the next 10 years of what this space looks

00:23:34.800 --> 00:23:37.600

like, which is really fascinating too.

00:23:37.600 --> 00:23:41.980

I'll pull us off of this train because there's probably people way deeper in this than either

00:23:41.980 --> 00:23:50.220

of us having more maybe accurate conversations about this on other podcasts.

00:23:50.220 --> 00:23:54.460

This episode of Launched is sponsored by, well, maybe you.

00:23:54.460 --> 00:23:57.260

If you listened to the Christmas special, you'll know that I've committed to growing

00:23:57.260 --> 00:24:02.000

the podcast this year and that starts today with two new initiatives.

00:24:02.000 --> 00:24:06.500

The first is sponsorships, which will sound, well, a lot like this.

00:24:06.500 --> 00:24:09.740

If you're a company and interested in sponsoring the show, please reach out.

00:24:09.740 --> 00:24:13.540

There will be a link in the show notes to our new sponsor page, which will have more

00:24:13.540 --> 00:24:16.840

information and metrics about the show, all that kind of stuff.

00:24:16.840 --> 00:24:21.640

If you don't want to hear ads on this show, though, the second initiative is for you.

00:24:21.640 --> 00:24:23.840

Launched is now on Patreon.

00:24:23.840 --> 00:24:28.680

There are a few tiers with some different perks, but all of them will come with an ad-free

00:24:28.680 --> 00:24:30.360

feed of the show.

00:24:30.360 --> 00:24:33.760

I've got some ideas for special content I could add in the future as well, but for now

00:24:33.760 --> 00:24:37.620

I'm keeping it simple while providing a way for folks to support the show directly if

00:24:37.620 --> 00:24:39.080

they'd like to.

00:24:39.080 --> 00:24:42.440

Please don't feel any pressure to join in with this, though.

00:24:42.440 --> 00:24:47.160

This show is a passion project and the fact that so many of you all listen and give me

00:24:47.160 --> 00:24:51.660

the opportunity to talk to all these inspiring creators that I get to talk to on the show

00:24:51.660 --> 00:24:54.620

is the biggest support that you can provide at all.

00:24:54.620 --> 00:24:59.580

These two initiatives will allow me to pay my excellent editor Jonathan and to be completely

00:24:59.580 --> 00:25:03.980

transparent I'd love to make some money off the show if there's some extra to spare.

00:25:03.980 --> 00:25:09.800

So thank you all so much for listening and new patrons especially for the direct support.

00:25:09.800 --> 00:25:12.520

Now back to the show.

00:25:12.520 --> 00:25:16.160

Let's pull back out to the app you ended up building.

00:25:16.160 --> 00:25:20.640

So you're playing with this as part of the contract, that job that you're doing, the

00:25:20.640 --> 00:25:23.080

generative art, you know, sort of world.

00:25:23.080 --> 00:25:26.480

What led to the AI profile pick app?

00:25:26.480 --> 00:25:27.760

Was that like the first idea you had?

00:25:27.760 --> 00:25:31.400

Or were you trying to figure out, you know, what's a product that you can build around

00:25:31.400 --> 00:25:32.400

this?

00:25:32.400 --> 00:25:37.920

I was indeed trying to figure out what kind of product I could create in the space.

00:25:37.920 --> 00:25:47.600

And around that time, I was watching a YouTube channel that just tracks advances in AI.

00:25:47.600 --> 00:25:53.120

And it's this guy, Medved Pro, that just plays around with all of the new technologies coming

00:25:53.120 --> 00:25:54.120

out.

00:25:54.120 --> 00:25:58.800

If I remember correctly, he was using this tool called Astria back then, which is an

00:25:58.800 --> 00:26:04.640

API that is based on Dreambooth and stable diffusion.

00:26:04.640 --> 00:26:09.520

So what Dreambooth is, is you could consider it as a layer on top of Stable Diffusion,

00:26:09.520 --> 00:26:15.000

where you can fine tune the model based on some kind of input images.

00:26:15.000 --> 00:26:20.960

For my own understanding here, so Stable Diffusion is a giant model with, you know, millions

00:26:20.960 --> 00:26:28.020

of images and whatever magic sauce that they're doing to take text that you input, and it

00:26:28.020 --> 00:26:33.760

turns that into an image based on this massive library of images that it understands.

00:26:33.760 --> 00:26:40.840

And then Dreambooth, you're saying, is a layer on top of that that will let you combine your

00:26:40.840 --> 00:26:43.920

own images into that model, essentially?

00:26:43.920 --> 00:26:44.920

Yes.

00:26:44.920 --> 00:26:54.200

So Dreambooth initially was a paper written at Google that basically takes diffusion-based

00:26:54.200 --> 00:26:59.840

models and fine-tunes them into being able to add a new subject.

00:26:59.840 --> 00:27:06.600

So stable diffusion currently does understand, if you will, what a chair means.

00:27:06.600 --> 00:27:13.100

So if you write in, type in a prompt with the word chair in it, it will pop out an image

00:27:13.100 --> 00:27:15.400

containing a chair, right?

00:27:15.400 --> 00:27:19.920

But it doesn't know who Charlie Chapman is.

00:27:19.920 --> 00:27:21.400

You're still not famous enough, Charlie.

00:27:21.400 --> 00:27:26.040

So it doesn't know who Brad Pitt is.

00:27:26.040 --> 00:27:30.440

As you know, BreadPit exists on the web and it's quite popular.

00:27:30.440 --> 00:27:32.700

It doesn't know who Charlie Chapman is yet.

00:27:32.700 --> 00:27:40.940

So what you can do using Dreambooth is fine-tune the model into teaching it a new subject.

00:27:40.940 --> 00:27:47.020

So you add in an identifier that Stable Diffusion doesn't know.

00:27:47.020 --> 00:27:52.300

So you can't use something like dog, cat, or chair, but some sort of random identifier.

00:27:52.300 --> 00:27:54.480

And then you teach it.

00:27:54.480 --> 00:27:57.660

These are the instances of this identifier.

00:27:57.660 --> 00:28:01.880

So it's pictures of you, let's say pictures of your face.

00:28:01.880 --> 00:28:10.280

So once you put in, let's say about 10 or 20 images of you with this instance name,

00:28:10.280 --> 00:28:16.340

that's enough for stable diffusion, the fine-tuned version of stable diffusion to learn what

00:28:16.340 --> 00:28:18.560

or who Charlie Chapman is.

00:28:18.560 --> 00:28:24.020

And so then I could say something like Charlie Chapman in a chair and it's combining, it's

00:28:24.020 --> 00:28:27.500

understanding of chair and understanding now new understanding of Charlie

00:28:27.500 --> 00:28:29.580

Chapman into a single image.

00:28:29.580 --> 00:28:30.220

Yes.

00:28:30.220 --> 00:28:34.860

So that's generally what that's what dream booth is.

00:28:34.860 --> 00:28:38.180

You said dream booth was just a white paper that somebody wrote or Google,

00:28:38.180 --> 00:28:41.340

somebody at Google wrote, and then it was productized, I guess.

00:28:41.340 --> 00:28:43.720

So dream booth has this paper written at Google.

00:28:43.720 --> 00:28:44.220

Yes.

00:28:44.220 --> 00:28:49.580

Um, and it's based on two parts, the fine tuning part and a diffusion based

00:28:49.580 --> 00:28:53.300

model that Google owns, uh, that is not public called image.

00:28:53.300 --> 00:29:05.300

So someone figured out how to re-implement Dreambooth, but not use Imogen as a way to generate images, but use Stable Diffusion.

00:29:05.300 --> 00:29:08.300

So that was released as open source.

00:29:08.300 --> 00:29:15.300

We haven't specified here, Stable Diffusion is one of these models, but importantly, it's open source, right?

00:29:15.300 --> 00:29:21.300

So anybody can download it, run it on their machines or on a virtual machine or whatever, right?

00:29:21.300 --> 00:29:29.540

right? Exactly. So the model itself can be used. The training data is not

00:29:29.540 --> 00:29:34.220

available and obviously, well maybe not obviously, but it would cost hundreds of

00:29:34.220 --> 00:29:39.620

thousands of dollars to train your own version of Sable Diffusion, meaning to

00:29:39.620 --> 00:29:44.360

train it from scratch. But the model is out there and the code that calls into

00:29:44.360 --> 00:29:48.140

that model is out there as well. So yes, you can run it on your machine to run

00:29:48.140 --> 00:29:52.440

on inferences, which means image generations based on a prompt.

00:29:52.440 --> 00:29:54.980

So that is out there.

00:29:54.980 --> 00:29:57.440

And the fine-tuning part of Dreambooth,

00:29:57.440 --> 00:30:01.320

off of that paper, was implemented and released as open source.

00:30:01.320 --> 00:30:06.920

And you just plug the two together, Dreambooth and Stable Diffusion,

00:30:06.920 --> 00:30:10.320

and you can start training any object or any person,

00:30:10.320 --> 00:30:16.600

have Stable Diffusion "understand" that concept,

00:30:16.660 --> 00:30:21.460

and then generate any variation of image

00:30:21.460 --> 00:30:23.440

containing that new subject.

00:30:23.440 --> 00:30:26.940

- We have stable diffusion, open source software,

00:30:26.940 --> 00:30:31.500

that, or at least the model itself is open source

00:30:31.500 --> 00:30:33.280

so that you could build software around it

00:30:33.280 --> 00:30:36.300

to generate images from a prompt.

00:30:36.300 --> 00:30:39.460

And now you have, I already forgot the name,

00:30:39.460 --> 00:30:43.460

but not Dream Booth, but an open source implementation

00:30:43.460 --> 00:30:46.580

of the Dream Booth paper that lets you

00:30:46.580 --> 00:30:52.980

basically add your own training to that stable diffusion model.

00:30:52.980 --> 00:30:54.940

And that is also open source.

00:30:54.940 --> 00:31:00.180

So now you have the ability to take, to add your own things to

00:31:00.180 --> 00:31:04.060

stable diffusions, understanding of the world, to make your prompts more specific.

00:31:04.060 --> 00:31:07.180

That's the sort of end result of these elements.

00:31:07.180 --> 00:31:08.460

That is correct.

00:31:08.460 --> 00:31:12.420

So you have these two, um, and even better.

00:31:12.420 --> 00:31:14.940

And that's why the, the, the bar was so low.

00:31:14.980 --> 00:31:17.340

And there was also an API.

00:31:17.340 --> 00:31:19.380

There was also someone, Astrea, who

00:31:19.380 --> 00:31:24.700

had made an API to do all of the ML work, et cetera, for you.

00:31:24.700 --> 00:31:26.180

It was implemented for you.

00:31:26.180 --> 00:31:28.780

And that's how-- and it was basically

00:31:28.780 --> 00:31:33.540

a service that cost $3 or $5, et cetera,

00:31:33.540 --> 00:31:38.620

to where you could just send off your own images,

00:31:38.620 --> 00:31:42.420

and you'd get a fine-tuned model ready-made for you.

00:31:42.420 --> 00:31:46.660

And then you can prompt it with whatever prompts you want, and it would generate

00:31:46.660 --> 00:31:47.740

these avatars for you.

00:31:47.740 --> 00:31:49.500

Okay, cool.

00:31:49.500 --> 00:31:54.340

So the bar, the bar, it was like the level of entry, it was, it was pretty easy.

00:31:54.340 --> 00:32:00.180

And at that point, it kind of started just clicking that there should be an app for

00:32:00.180 --> 00:32:00.540

that.

00:32:00.540 --> 00:32:06.100

Around the same point, like other indie devs had discovered Estrea as well.

00:32:06.100 --> 00:32:11.980

So, uh, Levels.io and Danny Postma started tweeting, you know, some, um, kind of

00:32:12.020 --> 00:32:17.700

thinly veiled tweets about a product that could leverage something like that or create

00:32:17.700 --> 00:32:19.780

your own maybe pictures or avatars.

00:32:19.780 --> 00:32:28.020

And around that point, knowing about Astria, I immediately knew like iOS is the place

00:32:28.020 --> 00:32:30.420

for this. This is where your photos live.

00:32:30.420 --> 00:32:36.620

Yeah. UI can be made very, very simple, very intuitive, and the result could feel

00:32:36.620 --> 00:32:41.540

magical. And at that point, I had decided like the web is not the way to go.

00:32:41.540 --> 00:32:45.500

And these two in the hackers or in the developers

00:32:45.500 --> 00:32:51.420

were primarily web developers and were making these hints

00:32:51.420 --> 00:32:54.340

that they were shipping something on the web.

00:32:54.340 --> 00:32:56.780

And yeah, I mean, I am an iOS dev.

00:32:56.780 --> 00:32:58.540

And I knew from the beginning, like, you'd

00:32:58.540 --> 00:33:01.340

get such a better user experience on iOS.

00:33:01.340 --> 00:33:04.420

And so I went like 100 miles per hour

00:33:04.420 --> 00:33:08.340

trying to implement this ASAP and release it on the App Store.

00:33:08.340 --> 00:33:14.020

Funnily enough, because of a lack of GPU computing scalability,

00:33:14.020 --> 00:33:18.140

Astrea had to move to another backend provider.

00:33:18.140 --> 00:33:20.580

So they were on AWS and they moved elsewhere.

00:33:20.580 --> 00:33:25.180

And overnight they moved from $3 per fine tune to $20.

00:33:25.180 --> 00:33:28.100

Was that while you were building this or after you released?

00:33:28.100 --> 00:33:31.180

It was the day I was releasing it, I think.

00:33:31.180 --> 00:33:32.260

Oh my gosh.

00:33:32.260 --> 00:33:32.780

Yeah.

00:33:32.780 --> 00:33:34.660

Yes.

00:33:34.660 --> 00:33:37.060

And that was a good lesson to learn, right?

00:33:37.060 --> 00:33:42.700

like own your back end, own your info now for obvious reasons.

00:33:42.700 --> 00:33:49.700

Because I knew that, again, this would be a winner on iOS,

00:33:49.700 --> 00:33:53.500

but the pricing had to be very fair, especially that not a lot of people were

00:33:53.500 --> 00:33:58.460

familiar with the technology. So might as well seem like a scam app,

00:33:58.460 --> 00:34:01.100

that is asking for so much money up front.

00:34:01.100 --> 00:34:02.140

Right.

00:34:02.140 --> 00:34:06.380

So five dollars could have been OK, but charging,

00:34:06.620 --> 00:34:09.680

after Apple's tax and how much it costs to run on the back end,

00:34:09.680 --> 00:34:14.620

charging like 25 or $30, that would seem like a big investment.

00:34:14.620 --> 00:34:19.860

And so that would kind of stifle the app's initial popularity or launch.

00:34:19.860 --> 00:34:20.400

Right.

00:34:20.400 --> 00:34:25.460

But I had just decided like I needed to launch and you had to launch ASAP.

00:34:25.460 --> 00:34:29.220

And, uh, and I knew that compute would get cheaper.

00:34:29.220 --> 00:34:30.660

Like that's how it always works.

00:34:30.660 --> 00:34:31.160

Right.

00:34:31.160 --> 00:34:31.580

Right.

00:34:31.580 --> 00:34:34.620

It's going to get cheaper with time, but now's the time to just

00:34:34.620 --> 00:34:36.140

launch and get something out there.

00:34:36.500 --> 00:34:44.260

So it took about 48 or 72 hours of not much sleep to really get everyone running.

00:34:44.260 --> 00:34:48.100

And then, yeah, I just put it out into the world.

00:34:48.100 --> 00:34:49.180

Wait, wait, wait, wait, wait.

00:34:49.180 --> 00:34:51.380

Like that was how long it took you to build the app?

00:34:51.380 --> 00:34:52.460

To build the app.

00:34:52.460 --> 00:34:53.100

Yes.

00:34:53.100 --> 00:34:55.660

Yeah, it was about 72 hours.

00:34:55.660 --> 00:34:57.460

It was a simple app.

00:34:57.460 --> 00:34:58.580

It was a simple app.

00:34:58.580 --> 00:35:03.420

Just calling into a backend and it was a B1 like bare bones.

00:35:03.420 --> 00:35:03.980

Right.

00:35:03.980 --> 00:35:04.700

Wow.

00:35:04.700 --> 00:35:05.380

Okay.

00:35:05.700 --> 00:35:09.460

So, so you had the idea and when you said you were off to the races, I mean, it's

00:35:09.460 --> 00:35:15.960

like zero to in the, or submitting to Apple in less than a week.

00:35:15.960 --> 00:35:17.560

Yes.

00:35:17.560 --> 00:35:19.180

Um, something I did help.

00:35:19.180 --> 00:35:23.820

Firebase is, is honestly amazing.

00:35:23.820 --> 00:35:26.520

Like that's, that's my take on Firebase.

00:35:26.520 --> 00:35:28.660

Uh, like kudos to them.

00:35:28.660 --> 00:35:29.900

These guys really did the job.

00:35:29.900 --> 00:35:35.020

They really do help you go from, from zero to a hundred quite fast.

00:35:35.580 --> 00:35:40.920

Um, and thanks to my previous apps, I already had my kind of standard library,

00:35:40.920 --> 00:35:41.220

right?

00:35:41.220 --> 00:35:41.380

Yeah.

00:35:41.380 --> 00:35:45.120

Like UI components, all these helper methods, et cetera.

00:35:45.120 --> 00:35:49.140

So that also helped a lot into getting something done.

00:35:49.140 --> 00:35:50.200

Wow.

00:35:50.200 --> 00:35:50.540

Okay.

00:35:50.540 --> 00:35:52.840

So that's pretty crazy.

00:35:52.840 --> 00:35:58.680

And I assume part of you rushing through this is, you know, that other people are

00:35:58.680 --> 00:35:59.660

going to do the same thing.

00:35:59.660 --> 00:36:05.660

And it's it's sort of like a land grab, you know, type situation where you want

00:36:05.660 --> 00:36:08.900

to get that early press, you want to get that early set of reviews.

00:36:08.900 --> 00:36:11.660

So you're sort of the established, you know, version of this.

00:36:11.660 --> 00:36:12.860

Yes.

00:36:12.860 --> 00:36:17.060

Were you, were you one of the first ones then to get out there with this?

00:36:17.060 --> 00:36:18.200

One of the first ones?

00:36:18.200 --> 00:36:18.780

Yes.

00:36:18.780 --> 00:36:24.100

Um, either the first or the second, and it was almost a month ahead of the big

00:36:24.100 --> 00:36:29.740

players such as Lensa, Prequel, so like these heavy hitting

00:36:29.740 --> 00:36:33.740

companies with an existing user base of millions of users that

00:36:33.740 --> 00:36:36.540

they can just simply send a push notification to saying, hey,

00:36:36.540 --> 00:36:39.420

there's this new feature. Regarding your point about

00:36:39.420 --> 00:36:44.660

needing to go very fast and chipping and putting something

00:36:44.660 --> 00:36:49.060

out there, that is also related to this new generation of AI,

00:36:49.380 --> 00:36:52.180

where it's very hard to build a moat.

00:36:52.180 --> 00:36:56.940

So since this technology is open and it's so magical,

00:36:56.940 --> 00:37:01.340

it's so powerful, but accessible at an API call.

00:37:01.340 --> 00:37:06.300

So technology like GPT-3 or Sable Diffusion,

00:37:06.300 --> 00:37:11.660

well, now anyone can build these magical apps in days.

00:37:11.660 --> 00:37:13.140

So there is no moat.

00:37:13.140 --> 00:37:18.180

So the new kind of advantage that you need

00:37:18.260 --> 00:37:26.300

is possibly better marketing or maybe niche down on something, but the bar of

00:37:26.300 --> 00:37:32.120

entry has just become solo that this magical effect is basically one prompt

00:37:32.120 --> 00:37:34.260

away or one API call away.

00:37:34.260 --> 00:37:35.280

Right.

00:37:35.280 --> 00:37:38.580

And so, yeah, anyone can build something like that nowadays.

00:37:38.580 --> 00:37:42.420

You spent a couple of days building this thing.

00:37:42.420 --> 00:37:47.840

Um, although, I mean, we're saying it like it wasn't, it was the small amount

00:37:47.840 --> 00:37:52.400

of work, but you're building off of this amount of time that you had spent

00:37:52.400 --> 00:37:55.400

learning and understanding the whole space, right?

00:37:55.400 --> 00:37:59.960

Like knowing how to use all these models and these APIs, part of that is the

00:37:59.960 --> 00:38:03.160

experience that you had, but either way you spent this time getting it out

00:38:03.160 --> 00:38:03.560

there.

00:38:03.560 --> 00:38:09.160

You had to update your pricing, presumably on the fly day of, but you got

00:38:09.160 --> 00:38:09.600

it out there.

00:38:09.600 --> 00:38:10.400

So how was that?

00:38:10.400 --> 00:38:12.160

How was that launch itself?

00:38:13.120 --> 00:38:19.320

The launch was exciting because even at a price point of, I think it was around $30,

00:38:19.320 --> 00:38:28.920

$28 or $30, you still have these early adopters that just want to try new technology.

00:38:28.920 --> 00:38:35.840

That kind of, you forget about the price point for a minute because they know that it's still raw, it's still fresh.

00:38:35.840 --> 00:38:40.800

But they would want to try any new technology no matter what.

00:38:41.160 --> 00:38:44.800

Right. And and they love giving feedback as well.

00:38:44.800 --> 00:38:46.600

Like I love early adopters.

00:38:46.600 --> 00:38:50.120

They kind of put blind trust into into a product.

00:38:50.120 --> 00:38:54.800

A lot more willing to accept bugginess and stuff because what you're comparing

00:38:54.800 --> 00:38:58.760

to is running through what all the blog posts were doing a week before where

00:38:58.760 --> 00:39:03.360

they're like, all right, you can download all these little, you know, scripts

00:39:03.360 --> 00:39:08.080

that you run locally and then you need to like rent this space on some Google

00:39:08.080 --> 00:39:10.240

server somewhere. And it's like, yes.

00:39:10.440 --> 00:39:15.640

What if instead I just hit a button to send, you know, $30 over to this person on the app store

00:39:15.640 --> 00:39:20.680

and then it just did all that for me? Like, that's the comparison. Exactly. And it's also that

00:39:20.680 --> 00:39:28.360

if they do that on, you know, Google Colab, then they would have to also research into writing

00:39:28.360 --> 00:39:35.880

their own prompts, getting the effect that they desire, etc. So it is quite like time investment.

00:39:35.880 --> 00:39:41.400

I mean, the whole thing on Google could take one hour just to train, fine-tune, etc.

00:39:41.400 --> 00:39:46.680

And then you get into another interface where you have to get some kind of prompting magic going to

00:39:46.680 --> 00:39:52.760

be able to create these avatars. So it is true that all of this was productized for you,

00:39:52.760 --> 00:39:59.800

even though it was at a price of $30, which is quite high, right? So even if you do compare it to

00:40:01.160 --> 00:40:06.440

you know, paying an illustrator for one illustration of you and the back and forth,

00:40:06.440 --> 00:40:13.160

or getting, you know, similar images of decent quality from a photographer, still $30

00:40:13.160 --> 00:40:20.600

with that blind trust could seem like a high price point, right? Yeah.

00:40:20.600 --> 00:40:26.440

So that became a priority very early on. I knew that I had to reduce the price point somehow.

00:40:26.440 --> 00:40:31.440

So that essentially translates into reducing backend costs.

00:40:31.440 --> 00:40:34.440

And that became the priority is rolling out my own backend.

00:40:34.440 --> 00:40:36.840

So that's what happened.

00:40:36.840 --> 00:40:40.840

Eventually fully focused on rolling out the backend

00:40:40.840 --> 00:40:44.640

because all of the players in the space

00:40:44.640 --> 00:40:46.840

were using that same API, were using Astria.

00:40:46.840 --> 00:40:51.040

So Astria was free to choose their price point

00:40:51.040 --> 00:40:54.240

and essentially drive the whole market to,

00:40:54.240 --> 00:40:59.040

You know, they capped it at a minimum of, let's say, $20 or $15.

00:40:59.040 --> 00:41:02.400

So whatever SGS price was, everyone else followed along.

00:41:02.400 --> 00:41:03.680

Yup, exactly.

00:41:03.680 --> 00:41:09.720

And that's, yeah, when I had to put kind of my ML hacker hat, knowing almost nothing

00:41:09.720 --> 00:41:16.320

about machine learning and starting to have to learn how to use and tweak PyTorch and

00:41:16.320 --> 00:41:20.920

Stable Diffusion and Dreambooth and understand all of the bits there.

00:41:22.640 --> 00:41:29.920

And yeah, so since so at that point, I, I deployed my own backend and the cost went from

00:41:29.920 --> 00:41:36.480

think between 15 and $20 on Astria to it was like around $4, I think, at the time.

00:41:36.480 --> 00:41:37.200

Wow.

00:41:37.200 --> 00:41:38.320

Right? Yeah.

00:41:38.320 --> 00:41:38.960

Yeah.

00:41:38.960 --> 00:41:42.640

And that was the new unlock, right? That was the how to get the next

00:41:42.640 --> 00:41:50.000

thousand or 10,000 users is by charging $5 and not $30.

00:41:51.040 --> 00:41:57.360

something that I guess a lot of people know, but people do not like to pay for apps. People do not

00:41:57.360 --> 00:42:04.480

like to pay for... like a small minority of people are okay with paying for things, especially on the

00:42:04.480 --> 00:42:11.440

App Store. And so charging any amount, even one dollar up front, instead of having a freemium

00:42:11.440 --> 00:42:19.680

kind of model, rubs people off the wrong way. And I saw that in the reviews, right? So there were so

00:42:19.680 --> 00:42:25.040

many, unfortunately, but there were so many one-star reviews. I'm like, "Oh, there's no free

00:42:25.040 --> 00:42:31.440

trial." Even though you try to explain that this costs money on the back end and it's just the

00:42:31.440 --> 00:42:37.440

economics of it just don't work out where I can offer a free trial first. Most people aren't

00:42:37.440 --> 00:42:44.880

interacting with software that has that much of a unit cost, right? Like unit economics aren't

00:42:44.880 --> 00:42:51.680

really a thing in our world. So pricing is bonkers and backwards in software because

00:42:51.680 --> 00:42:58.160

per unit cost is like zero almost. But the actual cost to build everything costs money.

00:42:58.160 --> 00:43:02.560

And so you have to do weird things. But you actually had real unit economics to deal with.

00:43:02.560 --> 00:43:10.000

Yes. And there was an FAQ in the app. There was an explainer that just lays the whole thing out,

00:43:10.000 --> 00:43:18.000

explains why it costs up front. And it's because, well, you know, we have a GPU on the back end

00:43:18.000 --> 00:43:23.680

that we have to park for one hour just to fine tune your photos. And even then, the reviews were

00:43:23.680 --> 00:43:28.480

like, I know that there's an explainer, but I am still, I still don't buy the argument.

00:43:28.480 --> 00:43:36.720

It's like, just people don't want to take it. Yeah, that's funny. Man. Yeah. And like,

00:43:37.280 --> 00:43:44.000

And everything about it is unproven, right? There's not any norms built around this in

00:43:44.000 --> 00:43:48.240

terms of people's expectations. Like weather apps, you can get away... Not get away with.

00:43:48.240 --> 00:43:52.400

That's the wrong word. But people are kind of trained to understand like it costs money to

00:43:52.400 --> 00:43:58.400

figure all this out or to pay APIs to figure this out. But generating an image is something that

00:43:58.400 --> 00:44:03.440

you're used to happening on your device itself. So it doesn't feel like that should be costing

00:44:03.440 --> 00:44:07.520

the company money, even though it obviously really is in this case.

00:44:07.520 --> 00:44:10.880

Especially when you don't get a sample.

00:44:10.880 --> 00:44:14.080

I wish the technology worked that way.

00:44:14.080 --> 00:44:20.240

I wish we could sample a few images, but it's a bit of all or nothing.

00:44:20.240 --> 00:44:25.680

We got to fine tune the model first, and that's where most of the cost goes in.

00:44:25.680 --> 00:44:29.040

And then it's not that costly afterwards.

00:44:29.040 --> 00:44:36.040

Right. The actual prompts in each image, the cost is pretty low relative to generating that actual model, right?

00:44:36.040 --> 00:44:42.940

Yes. Interestingly, I had record sales when when Lenza came out.

00:44:42.940 --> 00:44:51.340

Lenza having millions of users and Lenza having millions of sales really did and a huge boom on social media,

00:44:51.340 --> 00:44:56.400

really that teach people about the AI avatar thing, call it.

00:44:56.700 --> 00:45:00.060

So people started understanding that, oh, this thing exists.

00:45:00.060 --> 00:45:01.420

I've seen it on other people.

00:45:01.420 --> 00:45:05.620

So it added a lot of social proof in the sense that, well, all of my friends have one.

00:45:05.620 --> 00:45:08.900

So this thing kind of works and these avatars look cool.

00:45:08.900 --> 00:45:10.980

And so I'm going to try this thing out.

00:45:10.980 --> 00:45:16.620

And I understand because my friends told me as well that this thing costs money upfront.

00:45:16.620 --> 00:45:19.300

They paid first and they got these results.

00:45:19.300 --> 00:45:21.500

Yeah, that's true.

00:45:21.500 --> 00:45:24.660

Then the trust, it started building trust, right?

00:45:24.900 --> 00:45:28.860

Interestingly, competitors, bigger players

00:45:28.860 --> 00:45:32.180

on the market or in the field,

00:45:32.180 --> 00:45:35.420

brought this into the mainstream,

00:45:35.420 --> 00:45:37.360

removed so much friction,

00:45:37.360 --> 00:45:40.980

and resulted in record sales for AI Pro Fat Pick.

00:45:40.980 --> 00:45:45.540

- I feel like also, this is where your naming for the app

00:45:45.540 --> 00:45:48.260

probably helped a significant amount.

00:45:48.260 --> 00:45:49.620

I feel like in our indie circles,

00:45:49.620 --> 00:45:51.440

especially people listening to this,

00:45:51.440 --> 00:45:52.740

or people that I've had on,

00:45:53.780 --> 00:45:58.020

would want the name... Most products you make, you want it to have this sort of

00:45:58.020 --> 00:46:05.620

cute name that you can slap on stickers and it's like a whole... It's a very different thing.

00:46:05.620 --> 00:46:09.540

But in your case, when you're in this rat race where it's like,

00:46:09.540 --> 00:46:15.940

there's 1000 competitors instantly, you have to play the ASO game.

00:46:15.940 --> 00:46:21.860

Having a name like AI Profile Pick, establish yourself so that when one of these giant

00:46:21.860 --> 00:46:27.460

players with all the marketing money and user bases, when they come out with something that

00:46:27.460 --> 00:46:32.260

goes viral, if you don't remember the name, what are you going to search in the App Store?

00:46:32.260 --> 00:46:36.500

You're probably going to put, "Well, it's an AI thing and I want a profile picture." And then

00:46:36.500 --> 00:46:44.340

boom, that's your app, right? Yes. To be honest, though, that is a side effect. Obviously, the ASO

00:46:45.620 --> 00:46:52.500

gains here are good, but it was, to be honest, a lack of imagination on my end, in the sense that

00:46:52.500 --> 00:46:59.380

it's more like, because this was quite rushed for V1, I didn't have the time to really

00:46:59.380 --> 00:47:08.980

give the app that much character, right? So, actually give it a cute name with identity,

00:47:08.980 --> 00:47:15.140

with a nice app icon that really fits

00:47:15.140 --> 00:47:19.820

to what kind of identity you want to give to the app.

00:47:19.820 --> 00:47:21.780

The app had to be generic on day one,

00:47:21.780 --> 00:47:23.660

because what does the app do?

00:47:23.660 --> 00:47:27.100

It creates AI avatars or profile pictures.

00:47:27.100 --> 00:47:29.700

This is AI profile pic.

00:47:29.700 --> 00:47:34.220

The actual app icon is someone who has actually used our app

00:47:34.220 --> 00:47:37.940

and just created a support ticket because they

00:47:37.940 --> 00:47:39.860

needed help.

00:47:39.860 --> 00:47:42.580

And they sent us the avatars they had created.

00:47:42.580 --> 00:47:44.580

And we're like, these actually look pretty good.

00:47:44.580 --> 00:47:45.740

Can we use them?

00:47:45.740 --> 00:47:48.660

I basically bought the rights to use it as an app icon.

00:47:48.660 --> 00:47:50.180

Ha, that's pretty cool.

00:47:50.180 --> 00:47:52.700

In exchange for free credits.

00:47:52.700 --> 00:47:54.140

That's awesome.

00:47:54.140 --> 00:47:56.580

It reminds me of Clubhouse.

00:47:56.580 --> 00:47:57.500

Well, I say remember.

00:47:57.500 --> 00:47:58.740

I guess they're still around.

00:47:58.740 --> 00:48:02.740

But how they would have a different picture every time

00:48:02.740 --> 00:48:03.260

of--

00:48:03.260 --> 00:48:04.340

I guess that was a real user.

00:48:04.340 --> 00:48:06.100

I don't remember what the story was there.

00:48:06.100 --> 00:48:07.380

But that's pretty cool.

00:48:07.380 --> 00:48:11.980

Yeah, exactly. So it kind of showcases what kind of results you could get from the app.

00:48:11.980 --> 00:48:15.480

But at the same time, well, V1 had my face on it,

00:48:15.480 --> 00:48:18.680

and I couldn't wait until, you know, removing my face from it.

00:48:18.680 --> 00:48:20.680

So that's kind of how it worked out.

00:48:20.680 --> 00:48:24.180

That makes sense. Yeah, I would not like that myself.

00:48:24.180 --> 00:48:25.180

[laughs]

00:48:25.180 --> 00:48:25.680

Nope.

00:48:25.680 --> 00:48:31.280

So we already talked about Lenza coming in, and in your case,

00:48:31.280 --> 00:48:37.280

they did a lot of the legwork, essentially, of helping build this as like a

00:48:37.280 --> 00:48:44.320

legitimate, trustworthy thing, I guess. So in your case, that was mostly helpful. What about

00:48:44.320 --> 00:48:53.120

like, because I have looked in the App Store recently, even since we talked like a month ago,

00:48:53.120 --> 00:48:58.880

and it seems like there are just 1000 of these apps. Now, how has that been? Like, have you

00:48:58.880 --> 00:49:03.280

have you felt the sort of land rush of a bunch of other people coming in? And how have you

00:49:03.280 --> 00:49:08.320

differentiated. I guess having a back end is probably a piece of that. I feel like the top

00:49:08.320 --> 00:49:16.800

five apps are getting 90 plus percent of the market and then it's just a long tail of these

00:49:16.800 --> 00:49:25.680

smaller apps making peanuts and getting almost no users etc. And are you in the top five apps?

00:49:26.480 --> 00:49:27.600

I'm not even sure.

00:49:27.600 --> 00:49:30.080

Like, the app is doing quite well,

00:49:30.080 --> 00:49:33.920

but it's far from making the millions.

00:49:33.920 --> 00:49:36.360

So the trend went down as well.

00:49:36.360 --> 00:49:38.080

Linza was making about--

00:49:38.080 --> 00:49:39.720

I imagine, like I saw that on Twitter,

00:49:39.720 --> 00:49:43.120

but they were making about $1 million a day

00:49:43.120 --> 00:49:46.200

for about 20 days straight, right?

00:49:46.200 --> 00:49:51.560

AI Profile Big did not make $1 million a day any day.

00:49:51.560 --> 00:49:55.800

So it is a smaller app, but it's making decent income.

00:49:55.800 --> 00:49:57.960

But yeah, the big players came in.

00:49:57.960 --> 00:50:04.720

It was such like this Dreamboost and stable diffusion technology was such a good way to

00:50:04.720 --> 00:50:08.000

revamp apps that have existed on the App Store.

00:50:08.000 --> 00:50:12.800

They're like face tuning apps or removing backgrounds or blemishes.

00:50:12.800 --> 00:50:15.240

Which is a whole category on its own.

00:50:15.240 --> 00:50:20.680

Yes. And it was such a good way for them to revamp their product and add these magical

00:50:20.680 --> 00:50:23.040

new features, right? These magic avatars.

00:50:23.240 --> 00:50:30.680

And then it is true that the App Store now is crowded with this kind of app, with avatar apps.

00:50:30.680 --> 00:50:36.520

That's why it was quite important to be there very early on, because having had a one-month

00:50:36.520 --> 00:50:44.440

head start has also meant that the app has had, AI Profile Pick has had a lot of App Store inertia.

00:50:44.440 --> 00:50:49.720

So it's got a lot of reviews, ratings, downloads,

00:50:49.720 --> 00:50:53.560

so it just boosts the app on the App Store.

00:50:53.560 --> 00:50:58.680

And right now, the name of the game is ASO,

00:50:58.680 --> 00:51:00.360

and adding interesting features.

00:51:00.360 --> 00:51:03.160

So I'm working on a few new features.

00:51:03.160 --> 00:51:05.240

There's a few in the pipeline.

00:51:05.240 --> 00:51:07.560

Once your model has been fine-tuned,

00:51:07.560 --> 00:51:11.400

since it doesn't cost that much to do extra generations,

00:51:11.400 --> 00:51:13.160

it's to do interesting things with that,

00:51:13.160 --> 00:51:14.720

be it being able to do.

00:51:14.720 --> 00:51:19.960

So with this app store in inertia, it's also about how to keep the ball

00:51:19.960 --> 00:51:23.240

rolling because there's still an influx of users.

00:51:23.240 --> 00:51:27.120

There's still an influx of shares of generations on the app.

00:51:27.120 --> 00:51:34.240

And the idea is to how how to keep the ball rolling for for the long

00:51:34.240 --> 00:51:35.240

while to come.

00:51:35.240 --> 00:51:42.720

One of the things we talked about a while ago was like how fast the whole

00:51:42.720 --> 00:51:48.640

space is moving. And I think when we were talking last month, you were like, some new paper came out

00:51:48.640 --> 00:51:53.760

this morning that I'm trying to figure out somebody's in the process of trying to implement

00:51:53.760 --> 00:52:02.480

that paper now. Does that pace of change in that space? Is that coming into play with this area?

00:52:02.480 --> 00:52:07.280

Or is it kind of like those are kind of moved in a different direction than what would affect

00:52:07.280 --> 00:52:16.960

like your app and business now? I feel like stable diffusion v1.4 or 1.5 was really the mother load.

00:52:16.960 --> 00:52:25.280

That was kind of the big thing that shook up the industry and just gave so much

00:52:25.280 --> 00:52:33.680

to product makers. I do feel that it's slowed down a bit since then. So interestingly,

00:52:33.680 --> 00:52:39.840

it comes in waves. It doesn't feel like it comes on a daily or weekly basis. This kind of unlock

00:52:39.840 --> 00:52:49.760

that this technology or AI in general creates, it does come in waves. I don't think that the wave

00:52:49.760 --> 00:52:56.320

has been as... The few waves of recent times have been as big as the stable diffusion release,

00:52:56.320 --> 00:53:03.280

but I do think that there will be other medium-sized or bigger waves to come.

00:53:03.280 --> 00:53:09.440

Do you feel like you have to kind of be on top of that so that when one of those comes through

00:53:09.440 --> 00:53:14.240

that could have an impact, you know, you're one of the people that's jumping on that quickly?

00:53:14.240 --> 00:53:20.880

Thinking forward, it definitely is stressful to be on the lookout for something and then

00:53:20.880 --> 00:53:26.960

to ride that wave. So I don't think I would want to ride every wave.

00:53:26.960 --> 00:53:30.160

Well, right. Because most of them, you know, once things get going,

00:53:30.160 --> 00:53:36.160

most of them probably won't produce, you know, actual results for your business specifically.

00:53:36.160 --> 00:53:39.360

So that would start getting really overwhelming.

00:53:39.360 --> 00:53:46.240

Yes. So there's the aspect. And there's also the aspect of lessons learned from shipping AI

00:53:46.240 --> 00:53:53.440

profile pic. It's understanding the value of a mode, the value of not wanting to go

00:53:54.000 --> 00:54:02.320

head-to-head with the big players with no mode that can outspend you when it comes to ad spend,

00:54:02.320 --> 00:54:08.560

for example, or have a bigger marketing team or have these big influencer collabs, right?

00:54:08.560 --> 00:54:09.120

Mm-hmm.

00:54:09.120 --> 00:54:09.280

Yeah.

00:54:09.280 --> 00:54:15.280

So going head-to-hand with Lenza is impossible. It just dictates a lot when it comes to doing

00:54:15.280 --> 00:54:18.640

business, which is quite an interesting challenge as well. So it's still fun.

00:54:18.640 --> 00:54:22.960

Do you feel like you were going head-to-head with the big players? It sounded like when

00:54:22.960 --> 00:54:27.360

when you're describing it earlier that them entering the market was helpful

00:54:27.360 --> 00:54:30.320

for you? Or was it, do you think that was helpful for you in the short term,

00:54:30.320 --> 00:54:33.360

but then long-term that becomes more of a challenge?

00:54:33.360 --> 00:54:38.840

It never felt like I was going head to head with Lenza or the other players

00:54:38.840 --> 00:54:42.720

because I just do not have the user base.

00:54:42.720 --> 00:54:45.200

They have user bases with millions of users,

00:54:45.200 --> 00:54:50.280

meaning they can shoot up to the top of the store within a

00:54:50.280 --> 00:54:51.640

sending a push notification.

00:54:51.640 --> 00:54:57.360

Like this is the actual value that they can propose that I do not have.

00:54:57.360 --> 00:55:03.520

So I was just glad that the interesting thing is like Lenza as a player, so big

00:55:03.520 --> 00:55:07.800

that they created a small way for me to surf on, so they're a whale.

00:55:07.800 --> 00:55:12.520

Lenza's tail just hit the water and it created this ripple for

00:55:12.520 --> 00:55:14.000

everyone to, to write off of.

00:55:14.000 --> 00:55:16.520

Well, it's, I mean, it's kind of like what, how a lot of people

00:55:16.520 --> 00:55:22.080

describe getting Sherlocked by a platform owner like Apple or Google or

00:55:22.080 --> 00:55:27.240

whatever, where it's like, when they enter that space that you have had an

00:55:27.240 --> 00:55:32.440

app for a while and they implement it as a native feature in the platform, it

00:55:32.440 --> 00:55:36.680

can sometimes be beneficial for you because it like highlights this as a

00:55:36.680 --> 00:55:40.160

thing that people might want and legitimizes a business.

00:55:40.160 --> 00:55:43.160

And then more people end up seeking you out because you're the more, you know,

00:55:43.160 --> 00:55:44.680

advanced version or whatever in that case.

00:55:45.560 --> 00:55:47.560

It works quite well when

00:55:47.560 --> 00:55:52.840

Lenza, for example, launched the Magic Avatars feature,

00:55:52.840 --> 00:55:54.720

but it was paid, right?

00:55:54.720 --> 00:55:59.240

And we were almost offering it at the same price or even cheaper.

00:55:59.240 --> 00:56:03.280

So it kind of made sense, like a lot of people wanted to try both, right?

00:56:03.280 --> 00:56:05.560

Because the avatars are not the same, right?

00:56:05.560 --> 00:56:10.760

We produce more photorealistic avatars,

00:56:10.760 --> 00:56:13.240

and we just have a different value proposition.

00:56:14.560 --> 00:56:20.480

Contrasting that with getting Sherlocked is I would imagine it to hurt when Apple

00:56:20.480 --> 00:56:26.680

Sherlock's a feature and offers it for free and installed by default on the OS.

00:56:26.680 --> 00:56:27.720

Yeah, that's true.

00:56:27.720 --> 00:56:28.200

Right.

00:56:28.200 --> 00:56:32.120

It's like I already have this feature and it doesn't cost anything.

00:56:32.120 --> 00:56:38.000

So, so now the value proposition of the existing product has to be so much higher

00:56:38.000 --> 00:56:42.840

for people to download it, install it, launch it and, and pay for it.

00:56:43.640 --> 00:56:48.840

The last kind of big thing I wanted to hit on was, and you sort of mentioned this, but like,

00:56:48.840 --> 00:56:58.360

clearly this was a big wave situation. That wasn't going to last forever in terms of the

00:56:58.360 --> 00:57:03.160

height of how popular it was. But hopefully it has a long tail. How has that played out?

00:57:03.160 --> 00:57:07.880

And did you realize it was like a big wave and you didn't know what it would end or when it

00:57:07.880 --> 00:57:11.640

would end while it was happening? How did that kind of go through in your mind?

00:57:13.480 --> 00:57:18.840

The whole thing so far, and even from the beginning, even though in monetary terms,

00:57:18.840 --> 00:57:26.200

it could be a cash grab or land grab, it was also just an experiment, right? It was

00:57:26.200 --> 00:57:31.480

even in part just to have fun. It's like, this seems pretty cool, like the technology is pretty

00:57:31.480 --> 00:57:36.600

magical, so let's just launch something in the space and learn as much as possible, right?

00:57:36.600 --> 00:57:41.560

So there was always that aspect involved. One other aspect of it is

00:57:41.560 --> 00:57:46.560

is learning how to launch viral products.

00:57:46.560 --> 00:57:50.680

And this is where it leaves me.

00:57:50.680 --> 00:57:52.700

It leaves me into thinking

00:57:52.700 --> 00:57:55.820

what my next product would be, right?

00:57:55.820 --> 00:57:59.240

And it's always quite fun and interesting

00:57:59.240 --> 00:58:02.560

to learn about marketing on one end,

00:58:02.560 --> 00:58:05.600

but also how social networks are born

00:58:05.600 --> 00:58:07.560

or how virality works.

00:58:07.560 --> 00:58:08.400

- Yeah.

00:58:08.400 --> 00:58:11.120

- So I just consider it as kind of a V1

00:58:11.120 --> 00:58:15.200

or experiment number one in a series of experiments

00:58:15.200 --> 00:58:16.920

in launching viral products.

00:58:16.920 --> 00:58:20.000

- Right, not necessarily a V1 just of this product itself,

00:58:20.000 --> 00:58:25.000

but in terms of your sort of career of launching products

00:58:25.000 --> 00:58:27.360

sort of in this area.

00:58:27.360 --> 00:58:28.840

- Yes, exactly.

00:58:28.840 --> 00:58:33.560

And that is also very much related to

00:58:33.560 --> 00:58:35.040

how to create products where

00:58:35.040 --> 00:58:38.960

users do the marketing for you, right?

00:58:38.960 --> 00:58:42.680

So product-led growth or product-led marketing,

00:58:42.680 --> 00:58:47.080

instead of having to force product onto users

00:58:47.080 --> 00:58:49.640

by spending hundreds of thousands of dollars

00:58:49.640 --> 00:58:56.840

in ad spend, and once you turn that spend off,

00:58:56.840 --> 00:58:58.880

the product dies off, right?

00:58:58.880 --> 00:59:02.120

So as an indie developer, it's always

00:59:02.120 --> 00:59:05.600

interesting to seek out these kind of alternative ways

00:59:05.600 --> 00:59:11.640

of marketing products, just to see if you can create something that can sustain itself

00:59:11.640 --> 00:59:16.400

without having to push it onto users.

00:59:16.400 --> 00:59:26.000

So I treat it as experiment number one in this overarching, in this arch of potential

00:59:26.000 --> 00:59:27.440

products in the future.

00:59:27.440 --> 00:59:33.240

I'm excited to see, you know, where this app itself goes, but also, yeah, what you have

00:59:33.240 --> 00:59:41.320

in store for the future. With the AI advances, just the landscape is so bright. It's just

00:59:41.320 --> 00:59:48.800

such an exciting time to be alive, to be honest. I wonder if we're getting close to some kind

00:59:48.800 --> 00:59:54.200

of inflection point. A lot of people are, and a lot of smart people are talking about

00:59:54.200 --> 01:00:02.440

AGI and that it could happen within our lifespan. Actual artificial intelligence, but the singularity

01:00:02.440 --> 01:00:03.720

basically. Is that what you're talking about?

01:00:03.720 --> 01:00:04.720

The singularity.

01:00:04.720 --> 01:00:05.720

Yes. So.

01:00:05.720 --> 01:00:06.720

OK. Yes.

01:00:06.720 --> 01:00:09.480

So general intelligence, right?

01:00:09.480 --> 01:00:10.720

Yeah. OK. That's the word.

01:00:10.720 --> 01:00:15.320

To have a kind of a sentient machine that is

01:00:15.320 --> 01:00:17.640

as intelligent or more intelligent than humans.

01:00:17.640 --> 01:00:20.760

And a lot of smart people are betting on it.

01:00:20.760 --> 01:00:24.720

So that is it's just interesting.

01:00:24.720 --> 01:00:28.880

Right. And on the way to get there, if we do ever get

01:00:28.880 --> 01:00:31.840

there. But it doesn't matter because on the way, there's just

01:00:31.840 --> 01:00:34.640

so many interesting advances happening. And the pace is so

01:00:34.640 --> 01:00:37.720

fast that as a technologist, it's just so exciting.

01:00:37.720 --> 01:00:42.880

Yeah, there's it's constantly spawning new opportunities, new

01:00:42.880 --> 01:00:47.960

ideas for, you know, interesting products, right? It's like the

01:00:47.960 --> 01:00:51.640

space race. We're not an interplanetary species, like we

01:00:51.640 --> 01:00:54.840

thought we would be in a decade in like the 70s. But we sure

01:00:54.840 --> 01:00:58.200

got a lot of cool stuff out of, you know, lots of big money

01:00:58.200 --> 01:01:02.040

spenders coming up with new technology at a rapid pace?

01:01:02.040 --> 01:01:03.720

Yes.

01:01:03.720 --> 01:01:09.840

Well, that's, that's a good note to sort of leave off on. So I'll,

01:01:09.840 --> 01:01:12.840

I'll ask you the question that I always try to wrap the show up with,

01:01:12.840 --> 01:01:17.760

which is what's like a person or people out there that have inspired you that

01:01:17.760 --> 01:01:19.360

you'd recommend other people check out.

01:01:19.360 --> 01:01:24.560

Someone that appeared on your podcast previously is Christian Salig,

01:01:24.560 --> 01:01:25.520

the maker of Apollo.

01:01:25.520 --> 01:01:27.000

Yes.

01:01:27.040 --> 01:01:32.040

I just love his spirit, his wholesomeness, uh, and the product he creates.

01:01:32.040 --> 01:01:35.780

So overall, yeah, he's definitely an inspiration on my indie journey.

01:01:35.780 --> 01:01:38.280

Cannot agree more on that.

01:01:38.280 --> 01:01:43.160

Uh, got a chance to hang out with him in person, uh, over the summer at WWDC.

01:01:43.160 --> 01:01:44.680

And he is.

01:01:44.680 --> 01:01:50.220

As he is online in exceptionally kind human being, and also makes something

01:01:50.220 --> 01:01:53.800

that's a massive source of inspiration just as a dev and designer.

01:01:53.800 --> 01:01:56.040

So, uh, yeah, hard agree on that one.

01:01:56.040 --> 01:01:56.960

Awesome.

01:01:56.960 --> 01:02:02.760

Cool. Well, I guess we can go ahead and wrap this up. So where can people find you and

01:02:02.760 --> 01:02:03.760

your work?

01:02:03.760 --> 01:02:13.620

People can find me mostly on Twitter. So it's twitter.com/ronyfadel. And they can find my

01:02:13.620 --> 01:02:15.980

apps on fadel.io.

01:02:15.980 --> 01:02:21.860

Thank you so much for coming on this first episode of the 2023. Yeah, this is a fun one.

01:02:21.860 --> 01:02:24.100

Charlie, this was a pleasure. Thank you for having me.

01:02:24.100 --> 01:02:25.540

Yeah, no, the pleasure's all mine.

01:02:25.540 --> 01:02:32.340

Thanks for listening. This episode was edited by Jonathan Ruiz. If you'd like to discuss

01:02:32.340 --> 01:02:36.880

the show, you can find me on Twitter at underscore Chuckie C, or you can reach the show directly

01:02:36.880 --> 01:02:43.100

at Launched FM. You can also find me at Mastodon.social as Charlie M Chapman or the show at Launched,

01:02:43.100 --> 01:02:47.060

and you can find show notes and more at LaunchedFM.com.

01:02:47.060 --> 01:02:49.640

(upbeat music)

01:02:49.640 --> 01:02:52.220

(upbeat music)

01:02:52.220 --> 01:02:54.800

(upbeat music)

01:02:54.800 --> 01:02:57.380

(upbeat music)

01:02:57.380 --> 01:02:59.960

(upbeat music)

01:02:59.960 --> 01:03:02.540

(upbeat music)

01:03:02.540 --> 01:03:05.120

(upbeat music)

01:03:05.120 --> 01:03:07.360

(gunshots)