WEBVTT 00:00:00.000 --> 00:00:12.560 The lesson, if there is one lesson here, is regular expressions. We'll just keep it at that for now. I think ChatGPT is a secondary lesson here too. 00:00:12.560 --> 00:00:18.352 If you want some really bad regular expressions, you can ask Chargy PT and it will hallucinate you some. 00:00:20.687 --> 00:00:33.328 Hello, everyone, and welcome to episode 65 of the Metacast Behind the Scenes podcast. The Holiday Edition. The Holiday Edition, yes. I am Ornab Deca. I am Ilya Bezdelov. 00:00:33.328 --> 00:00:47.664 So today, we'll try to keep it to 30 minutes. We are going to talk about three things. A quick recap of the year so far, because this might be the last episode of the year, as far as you'll hear. We'll talk about a... 00:00:47.664 --> 00:00:57.968 very big outage we had over the weekend and a very serendipitous solution to that. And a little bit about Blue Sky. 00:00:57.968 --> 00:01:12.400 I wonder why you're so noncommittal about this being the last episode of the year, because it is going to be the last episode of the year. Unless something really disrupts our plans. Exactly. Unless we have an emergency or something like that. Yeah. So we're recording this. 00:01:12.400 --> 00:01:26.463 at the end of November. So there's still a month to go. So we'll see. But I think it's highly likely that this will be the last you'll hear from this podcast in this year. Yeah. So do you want to start with the holiday stuff first? Sure. Yeah. So... 00:01:26.463 --> 00:01:38.304 At the beginning of the year, where were we, Elia? When did we launch our closed beta? In February or March or something? Yeah, closed beta, I believe it was February 12th or something. 00:01:38.304 --> 00:01:51.280 Because I think it was end of January or very early Feb when Apple announced transcripts in their app and we just scrambled and launched. Yeah, and we were like, we have to do this immediately. 00:01:51.280 --> 00:02:05.328 Till then, you, me, and Jenny, we were maybe using the app like ourselves. In February, we launched Open Beta. We were in closed beta since August last year. We had a handful of users at that time, yeah. 00:02:05.328 --> 00:02:16.912 Right. Yeah, yeah. And then we launched the open beta in February 2024. It was a miracle how your standards change when timelines change. 00:02:16.912 --> 00:02:31.247 When you really want something and there is an external factor and you just throw your perfectionism out of the window and things just start to happen. For me, it was a very interesting lesson when we actually did that. Well, I think even with that urgency, it still took us... 00:02:31.247 --> 00:02:41.536 A few more months, like we launched in Android Play Store in September, if I remember, or maybe end of August. And then iOS App Store in October. 00:02:41.536 --> 00:02:53.743 And here we are in end of November, end of the year. It's by no means a hockey stick kind of growth. Otherwise, you probably would have heard otherwise from us already. But... 00:02:53.743 --> 00:03:08.015 We're seeing healthy signals, I think. We're seeing an active number of people continue to use the app, continue to use the powerful features. The number of subscriptions, premium users are trickling, but increasing over... 00:03:08.015 --> 00:03:16.192 time. Healthy percentage, I would say. And most interestingly, I think we have... 00:03:16.192 --> 00:03:24.991 Somehow, without us putting any marketing dollars in, it's been all organic marketing so far, people are finding the app. 00:03:24.991 --> 00:03:39.312 Every month we are seeing on and on new people come in, they try the app, some of them try the free trial, some of them stick on, some of them drop off. We have reasons to know like why this is happening. So it's not looking bad, I would say at the end. 00:03:39.312 --> 00:03:51.376 the year. What do you think, Ilya? I definitely agree with you. We crossed $500 in revenue about a month ago. Actually, I have not checked the all-time revenue recently. 00:03:51.376 --> 00:04:05.295 I just know what we were making in the last couple of weeks from the RevenueCat app because it doesn't show you the all-time values. It only shows you the last 28 days, I think. Well, last two weeks, I don't remember what it was. Yeah, it only shows that on your website. 00:04:05.295 --> 00:04:06.879 The all-time revenue. 00:04:06.879 --> 00:04:21.247 What I really like about this stage is we get the users who email us, people who comment on the Reddit posts, DM us, and most of those are pretty thoughtful messages and emails. And even when people drop off, 00:04:21.247 --> 00:04:34.240 Some of them actually drop off after they ask for something and explain their reasoning, and we tell them it's going to take a while. I don't think we really outright denied anybody's request. I think it was mostly like, it will just take some time. 00:04:34.240 --> 00:04:44.560 And they're like, yeah, it's a blocker. And then they hopefully will come back later. And I think one of the biggest actually things that we've learned so far is the private podcasts. 00:04:44.560 --> 00:04:51.216 people who pay for strategy in the new york times or whatever all these podcasts or patreon podcasts 00:04:51.216 --> 00:05:04.495 Those people actually are likely to pay for a premium podcast app because they are already paying tens of dollars for a set of subscriptions, right? And we don't support that yet. That's going to be one of the things that will be coming pretty soon. 00:05:04.495 --> 00:05:14.495 But yeah, I'm really enjoying the amount of positive response and thoughtful response. You know, even if it ends up in the user dropping out. 00:05:14.495 --> 00:05:22.160 The interaction is always positive and we always learn from our users. I think so far we've only gotten hate from one person. 00:05:22.160 --> 00:05:36.848 that unhappy redditor from the overcast subreddit who comes to our subreddit and just shits in our posts. And we called him out or her, like, why don't we just jump on the call and we talk and you tell us more because you have really good thoughts. 00:05:36.848 --> 00:05:44.560 And I actually don't think they've commented on any of our posts afterwards. Yeah, it's okay. They'll always be, especially on Reddit. 00:05:44.560 --> 00:05:58.831 They'll be like that. Well, it's Thanksgiving week in the US. So I'm going to say, yeah, we're very grateful to have the people who are continuing to use the app, the new people who have come on. And actually, I just checked since you said the five. 00:05:58.831 --> 00:06:09.055 $200 all-time revenue. We've grown more than 20 plus a handful of percentages since then already. We are over 600 now? Yeah. Cool. 00:06:09.055 --> 00:06:16.959 600 plus some change. I think at this point you can just tell the number. 619, so yeah. 00:06:16.959 --> 00:06:31.279 So it's not bad. And it's been about a month, I think, since we crossed 500. So that's pretty good without any marketing dollars. And I think the features that you talked about, we have some powerful playlist features coming in, private podcast. 00:06:31.279 --> 00:06:45.487 And today, actually, we're going to talk about the outage, which turned into a really nice feature for us. That's the second segment of this we're going to transition into shortly. So I think this year we have put together the foundation to work on more. 00:06:45.487 --> 00:06:57.807 of those really cool and useful features that we can ship. I want to say pretty quickly, but as far as software development goes, they are never quick, but I think we've done most of the really difficult work already. 00:06:57.807 --> 00:07:12.079 at least for the foundation. And now we can really iterate on value-add features, and I'm really looking forward to that. I think by the time this episode comes out, we will have one of those newer things shipped, and we want to keep shipping more and more of these, and we will talk about that. 00:07:12.079 --> 00:07:25.759 that in a second. At the same time, though, I do feel now that we're in production and we have active users that are using the app, the importance of things going wrong is much more now. 00:07:25.759 --> 00:07:40.175 With just two and a half, three people, it can get taxing sometimes. So that will have a weight on us delivering value-add features faster and faster, these kind of outages. To this one, we locked into. So let's get into... 00:07:40.175 --> 00:07:52.543 like this outage over the weekend yeah so this weekend i noticed that the transcript was taking a long time to generate for one of my podcasts sometimes it happens i dismissed it i'm like whatever 00:07:52.543 --> 00:08:04.160 But then we got an email from a user who told us that they've not been able to see transcripts for some of the newer episodes that they've been listening to, which means that something was off. 00:08:04.160 --> 00:08:13.216 A premium paying user, right? A paying user who had been with us for a long time. And if you're listening to this, you know who you are. 00:08:13.216 --> 00:08:25.216 So we're really thankful for all the feedback you've given us. And this is also special because we know the user. They've given us their time from very early days and they've also given us their money and they've also given us feedback. 00:08:25.216 --> 00:08:37.423 It's the kind of user that you want to always keep happy. The ideal user, yes. The ideal customer, yeah. Exactly, yeah. We use a third-party provider to run AI models for transcriptions. 00:08:37.423 --> 00:08:51.696 And I looked at their dashboard from my phone, and I'm seeing that we have a bunch of jobs that have been started, but never completed. They were still in the starting phase. So going a one-minute technical rabbit hole into this, what's happened? 00:08:51.696 --> 00:09:05.903 That's like four and a half hours of audio. Audio is already like massive amounts of data. 00:09:05.903 --> 00:09:20.111 four and a half hours of that. Running AI models on that is pretty expensive. So what we don't want to do is have a fleet of servers that we own and we keep running all the time waiting for these kind of jobs. 00:09:20.111 --> 00:09:34.320 What we instead do is it's almost like a serverless mechanism. So we don't own our own servers. Whenever there is a transcript job that's coming in, we have set up our ML models to be run in a... 00:09:34.320 --> 00:09:48.528 serverless fashion on servers that somebody else owns and they have made a SaaS out of it. We call that service and they do the orchestration and running on the ML jobs and then we extract the output from it. So that service, there is no error. 00:09:48.528 --> 00:09:54.416 anything. It just says the instance is booting and that's it, right? It's stuck. And we have not changed anything. 00:09:54.416 --> 00:10:06.207 We haven't changed anything and there's no controls on our site to like reboot the instance or fix it or something like that. So basically it's stuck. We created a ticket with them, but there's no response for like over a day. 00:10:06.207 --> 00:10:20.559 I think it's been two days at this point. Two days, yeah. And we had like 10 plus premium paying customers. Transcripts fail like this, getting stuck like this at that point. So we had to do something about it. Yeah, but then to the rescue came who? 00:10:20.559 --> 00:10:34.720 Our friend, do you want to name him? Well, we didn't get his permission yet. But I guess we'll just say that his name is the same as mine. I actually tagged him on a LinkedIn post already. Oh, you did? Okay. 00:10:34.720 --> 00:10:49.488 So I'm pretty sure that you got confused by that email exchange where you had two Ilias, especially when Gmail only shows the first name. Oh my God, that was crazy. Even on Apple Mail App, because it's just Ilias, Ilias and you. 00:10:49.488 --> 00:10:57.152 And it was so bizarre. But based on the context of who's saying what, I could totally figure it out. But it was pretty mad, yeah. 00:10:57.152 --> 00:11:11.440 Yeah, so I met Ilya, I want to say a year ago, maybe a year and a half. Our kids used to go to the same Russian preschool in Florida. Yeah, we just met at one of the playdates and we started talking about who does what and I told him about the app. 00:11:11.440 --> 00:11:25.648 actually had similar ideas in the past, and he got really, really excited. He also had been a beta user from the very early days. He gave us a lot of feedback, also lots of technical inputs, because he's also an engineer. I think about many months ago, he sent 00:11:25.648 --> 00:11:37.392 us an email. He's like, would you use my service if I built you a transcription service? And we said, yes, but at this price point. And out of the blue, I think a week ago or so, 00:11:37.392 --> 00:11:51.663 Yeah, like I think last Tuesday or something. This is like four days before the outage that we had. Yeah, he sent us an email. Actually, it landed in spam. We also didn't see it for two days. Maybe because the email had links and he sent this to one of our Google Groups. 00:11:51.663 --> 00:11:52.528 places. 00:11:52.528 --> 00:12:06.799 It got filtered. I just was scrolling through those spam reports and I saw an email from him and I approved it. And the rest is history, right? Yeah, basically he said, like, I have this service that I bootstrapped for you and I was able to get the cost to... 00:12:06.799 --> 00:12:12.943 what you expect. Are you interested? And then you start experimenting with it. And I guess, yeah, why don't you tell more? 00:12:12.943 --> 00:12:27.279 Yeah, so we started experimenting just by kicking off some jobs to it, looking at the quality, comparing it with the transcripts that we were generating already. And then we liked it. It was also way faster than our current mechanism with the server. 00:12:27.279 --> 00:12:31.567 provider that we were talking about like five six times faster yeah 00:12:31.567 --> 00:12:45.840 Because our serverless provider is also doing some optimization on their end. They are not keeping the servers running either. If it is cold, if that ML model has not been run in a while, it's cold. It takes some time for everything to boot up. 00:12:45.840 --> 00:12:57.567 And this was way faster. The price point was comparable to what we had. And so we were like, OK, let's get this into the app. But just for like our internal you, me and Jenny. 00:12:57.567 --> 00:13:03.519 as a feature so that we can toggle between the two types of transcripts that we have and compare. 00:13:03.519 --> 00:13:17.792 And so I think we did that for about a couple of days and we were already thinking, let's switch over to this. Maybe next week we can quickly switch over to this. That was probably like Thursday or Friday. We decided, okay, next week, let's just switch over to it. It won't be too much work. 00:13:17.792 --> 00:13:27.936 And then bam, the outage happens on Saturday. And you and I quickly talk on Slack and we figure out, okay, we have to do this like right now. 00:13:27.936 --> 00:13:40.495 Meanwhile, we had some terms and actually one of the other things is we wanted to use Ilya's service, but we were also like we need to at least have some basic terms or conditions kind of agreed. 00:13:40.495 --> 00:13:54.799 Thankfully, he also shipped it out within the same day, I think, the terms and all that. I think almost everything we asked for, because we also asked, I think, for some changes to the API, he shipped pretty much the same day on the next day, which was incredible compared to... 00:13:54.799 --> 00:14:09.375 that other company, right? We opened a ticket and there was no response. There is no way to track it. I remember I sent you a message. I'm like, so we just have this outage and I forgot exactly what I said, but the point is we're not doing anything. Yeah, that's when we decided to just make the switch. 00:14:09.375 --> 00:14:22.960 Yeah, I did some hacky work on Saturday so that all transcripts for all users are getting generated via this new service now. Yeah, it's basically fixed. 00:14:22.960 --> 00:14:36.975 I don't think we actually told Ili about this. We just switched our production app to it. Well, we did tell him that we're starting to use this for all our internal users, which is three. But yeah, we didn't tell him that we switched over completely. We can tell him now. 00:14:36.975 --> 00:14:40.720 Before this episode comes out, yeah. Yeah, but I think it's a good problem to have. 00:14:40.720 --> 00:14:54.991 If I was in his shoes, I would love to have that problem. And he was at the right time, with the right price point, with the right product. When he first sent that email and he started doing the prototyping, I was like, this is not the right time. I mean, I didn't tell you this, right? But I was thinking, this is not the right time to do this. 00:14:54.991 --> 00:15:08.384 But I want to keep the momentum with Ilya going. But that turns out to be the absolutely right thing to do. Because at first I thought it offers us redundancy, but it offered more than that. Turns out we actually flipped the standby. 00:15:08.384 --> 00:15:15.648 Right. We decided that we will flip the switch even before the outage. And then we wanted to keep our current mechanism as a fallback. 00:15:15.648 --> 00:15:27.679 But if everything doesn't work, it just broke. I'm sure that operational problem will get figured out and it'll start working again. But we'll flip over. But now that will become our secondary, the standby mechanism. 00:15:27.679 --> 00:15:42.399 One of the things that I'm always concerned about is small developers, especially like single developer operations. What if something goes wrong? What's their operational process and all that? Actually, I think when they just have one customer, which is us, us not being... 00:15:42.399 --> 00:15:56.687 a huge scale operation, I think it's okay. Maybe when you have at least tens of thousands of transactions per day, I would say, and you have maybe like 100 customers, that's when if something goes terribly wrong, you will just be drowned by ops. But I think when you have 00:15:56.687 --> 00:16:00.864 a really small number of customers. The only problem is, what if this happens in the middle of his night? 00:16:00.864 --> 00:16:12.096 So I was actually thinking, like on Saturday, right, when I did this hot fix to switch over to the new service, is what if we didn't have the service at that time? What we would do? 00:16:12.096 --> 00:16:26.416 And we do have a solution, actually. I don't know if you'll like the solution, but we do have the solution. At the end of the day, we are running ML models. We're running it either in the serverless server we were using or now via this other. 00:16:26.416 --> 00:16:39.279 service. But these are open source models that we are using with some tweaks in either way. And for those people who are listening to this, you might hear some noise coming from our website. It will soon go away. 00:16:39.279 --> 00:16:48.192 It's fall, so there's a leaf blower going on outside. Hopefully the timing won't be interrupted too much, but we'll see. But yeah, worst case, Ilya. 00:16:48.192 --> 00:16:55.759 You, me, Jenny, we just roll up our sleeves, see what's failing and run those models locally on our machines. 00:16:55.759 --> 00:17:08.832 Our scale is thankfully not at getting a thousand transcript requests a minute. So we can manage this by hand still. It's going to be painful, but that's what we'll have to end that maybe. 00:17:08.832 --> 00:17:15.663 If that's the thing, we live like that for a couple of days while we boot up a server of our own. I mean, if that's the only way, right? 00:17:15.663 --> 00:17:29.935 We boot up a server of our own, do the orchestration part there and move it over there. It felt like a very Amazon Sev2-ish situation for me over the weekend. But I was very happy, very proud to... 00:17:29.935 --> 00:17:41.983 get it fixed like almost immediately. And also figure out that if we didn't have this, we would still have a way out of this problem. Right. We could also download files, put them into Descript. 00:17:41.983 --> 00:17:56.271 It's going to be more and more painful, but yeah, there are lots of ways, yeah. Kudos to Ilya for predicting the outage. That's a pretty darn good prediction that he made, that our service will fall apart. And he was just there with his solution. 00:17:56.271 --> 00:18:03.263 which is amazing. If we had conspiracy theories in this podcast. Hmm. 00:18:03.263 --> 00:18:17.903 Anyway, let's not get there. What did he do to that other provider? Did he hack it? Yeah, I know. Okay, on the theme of Thanksgiving, that's the, I think, third thanks goes out to Ilya. 00:18:17.903 --> 00:18:32.208 for totally saving the day for us. So far, counting our users who have been with us for over a year, the premium paying customers and the other people who are using and giving us feedback. Second goes out to the other person who told us about that. 00:18:32.208 --> 00:18:46.415 outage who's also been with us for like over a year at this point uses our app for podcasts all the time and all that so that's great now let's quickly switch on to the app updates bit and one of the most beautiful things 00:18:46.415 --> 00:18:59.887 to come out of this outage and switching over to Ilya's service is how fast it is. We already said it's five or six times faster, right? Life gave us lemons and we made some lemonade. Yeah. 00:18:59.887 --> 00:19:05.791 Maybe we put in a shot of like tequila in there too and make it a margarita even better now. 00:19:05.791 --> 00:19:20.079 You know, I don't remember when it started, maybe about eight years ago. I started having a really adverse reaction to tequila. I used to really like tequila, but then I was in Mexico and I was drinking those high-end tequilas in high-end... 00:19:20.079 --> 00:19:28.832 all-inclusive hotel. And I was having hallucinations. And I'm like, maybe there's tequilies. Maybe something's wrong. I know it's an off topic. But then I tried some tequila in the U.S. 00:19:28.832 --> 00:19:41.056 Same exact thing. I drink a small margarita and I'm having wild hallucinations at night. So I have not had a tequila for a very long time. I think when you came to our place, did you try one? 00:19:41.056 --> 00:19:50.223 Because I love making margaritas. I think you did. That was the last time, probably. Yeah, maybe I had so much of it that it overpowered my hallucinations. 00:19:50.223 --> 00:19:57.008 But yeah, tequila is one drink that I really feel weird about. Maybe it's also a specific kind of tequila. Maybe yours was different. 00:19:57.008 --> 00:20:11.311 And for the other side, this is amazing how timing works, right? Till about a year and a half back, I hated tequila. And then we went to Mexico, had some really good tequilas there. And then I was like, okay, this is actually tequila. What I was drinking was shit. 00:20:11.311 --> 00:20:17.392 So because of this, our transcripts are now incredibly fast. 00:20:17.392 --> 00:20:30.175 If you use our app, you won't even know that you're requesting a transcript because we've made it all like behind the scenes. You just open up the podcast. Right now, it's estimating how many minutes and all that. And then it shows you. Usually it's... 00:20:30.175 --> 00:20:35.968 I want to say between 5 and 15-ish minutes to generate the transcript for an episode. 00:20:35.968 --> 00:20:49.824 Depending on how long and how many speakers are there and all that, it takes different time. But now it's going to be like less than 30 seconds. And in almost all cases, even for a less long episode, it's going to be less than a minute. 00:20:49.824 --> 00:21:04.112 It's going to be so fast that we're going to remove all that calculation of how long it's going to take, show you all that details. We're just going to rip out all that and make it super simple. Yeah. And I've heard from a few people, like I'm listening to 30 minutes. 00:21:04.112 --> 00:21:16.127 and I'm having to wait for like 10 minutes for the thing to be available. And by the time it is available, you already passed the premium paywall threshold. So it's all blurred. That also helps solve that problem. 00:21:16.127 --> 00:21:24.256 So that update is coming very soon to the app? Actually, that update is already in the app because it was a backend update. The UI is coming, but the speed is already there. 00:21:24.256 --> 00:21:37.279 The UI still shows you like, oh, eight minutes, but then it's ready like within 30 seconds right now. Even the sandbagging in the UI. Yeah. We're going to remove the sandbagging, yes, and pour some water on it. 00:21:37.279 --> 00:21:47.951 All right, so the other feature. There's a bit of a tale to this one too, but I'll let you start, Ilya. Yeah, the other feature, I just want to preface this. I was working on something else. 00:21:47.951 --> 00:22:00.511 And I'm like, while I'm in this file, I can add a few more lines of code and... Just make this other awesome thing. Yeah, create a super cool feature that wasn't even a high priority, but I'm like, why not? 00:22:00.511 --> 00:22:10.000 Long story short, ended up being 500 lines of code changed. 173 comments on pull request and 37 commits. 00:22:10.000 --> 00:22:21.887 and a week of calendar time and two hour long phone call and countless messages in Slack and a bit of frustration from both me and Arnav. 00:22:21.887 --> 00:22:36.528 With the code, with the project, with each other. But the end result is that now, if you go to the description of the episode, actually, as you listen to this episode, if you go to the description by tapping on that podcast cover icon in the middle, 00:22:36.528 --> 00:22:47.359 which actually I think is not very obvious. It's a button, we need to do something about it. In there, we always have the timestamps for different topics that we discuss. Now you can tap on those and the player will jump right there. 00:22:47.359 --> 00:22:59.647 So if you're already playing the episode, it will just jump to the other timestamp. If the episode is not playing, then it will open the episode and it will jump you to that part. So it works pretty much like YouTube. 00:22:59.647 --> 00:23:13.248 Not just our podcast, like any podcast you're listening to on Metacast, as long as they have timestamps in the description, this will work. Right. Yeah. Like Lex Friedman, he has dynamic ads, I think, or he has dynamic ads at the very end. 00:23:13.248 --> 00:23:24.048 which is why timestamps actually work, even if the length of the audio is slightly different, because it's not in the middle of the episode. Yeah, so that feature is really cool, especially for podcasts that 00:23:24.048 --> 00:23:38.384 I very rarely actually jump between topics in podcasts because for podcasts that I listen to, I usually listen to the whole thing. But there are some things like you would send me a podcast and you would be like, oh, listen to this thing. And like, I'm not interested in spending the full hour listening to it. 00:23:38.384 --> 00:23:53.056 I can just jump to a specific topic. So now we made this possible with just 500 lines of code. The lesson, if there is one lesson here, is regular expressions. We'll just keep it at that for now. 00:23:53.056 --> 00:24:07.551 But we had a lot of debate over that for multiple days. I think ChatGPT is a secondary lesson here too. If you want some really bad regular expressions, you can ask ChatGPT and it will hallucinate you some. Right. 00:24:07.551 --> 00:24:17.519 Okay. That feature is coming to, it's going to be what? 1.14? 1.14. Yeah. So hopefully by the time this episode comes out, it's already out there. 00:24:17.519 --> 00:24:31.344 Okay, so let's slowly start going towards the end of the episode, towards clearer blue skies all around us, hopefully. Isn't that how time works? Yes. You mean time is always going slowly? 00:24:31.344 --> 00:24:37.968 That depends, but it always drifts towards the end. Because maybe we're going towards a black hole or something like that. 00:24:37.968 --> 00:24:52.240 Okay, it's getting very distracting also here because my dog is now starting to react to the leaf blower, which is right outside now. So before everything goes crazy, let's talk about some bluer. 00:24:52.240 --> 00:25:06.640 clear blue skies. Yeah, so I'll let you start because you were a Twitter user. I never really was. Well, blue sky is a social network, but it's also a open social network protocol. 00:25:06.640 --> 00:25:18.864 called ATProto. Twitter's former CEO, who is Jack Dorsey, he started Blue Sky a while back, 2018, as an open alternative to Twitter. 00:25:18.864 --> 00:25:30.991 When Twitter started having all sorts of debates about censorship and left and right and all that, he started this. I have been on it for, I want to say, a year and a half at this point, probably. 00:25:30.991 --> 00:25:45.248 In terms of features, it was pretty good. Like early on, there were a lot of missing features itself, but recently there were features, but all these social networks, they need the social network for it to work. And that was missing. 00:25:45.248 --> 00:25:52.336 And then you probably have heard already, listener, is that towards the middle and end of November. 00:25:52.336 --> 00:26:06.672 A massive migration of people into Blue Sky has happened, right? As we speak right now, they're growing like a million more people every day or some crazy number like that. They're at about 22 or 23 million right now. That's when... 00:26:06.672 --> 00:26:18.112 I think the social network starts making sense is when all the people that you want to talk to are already there. Right. And it's essentially Twitter users migrating mostly. Yeah. 00:26:18.112 --> 00:26:29.423 And a lot of Twitter tech users held on to X and Twitter for a long time, especially indie hackers and people like that, because they have a big following on Twitter. 00:26:29.423 --> 00:26:42.544 It's very hard to like say, okay, I'm going to leave all of this and go over. But a few influential people, I think, started this trend. I want to say of the two that comes to the top of my head. 00:26:42.544 --> 00:26:54.864 Kelsey Hightower, a month back, I think he made this high profile. Okay, I'm done. I'm moving. Here's Blue Sky. And he has been very active and promoting Blue Sky quite a lot, helping people out quite a lot. 00:26:54.864 --> 00:27:06.127 And then most recently, Justin Jackson, who was also in our episode like last year, he moved over. And I think that made a lot of people move over as well. 00:27:06.127 --> 00:27:20.432 Yeah, that made me actually go out and test it out. A lot of indie hackers, I think, saw and took notice of that. But at the same time, I mean, Justin Jackson is in the tech indie hacker space. A lot of celebrities, lots of bigger influential people. 00:27:20.432 --> 00:27:31.632 same move. So there's been a mass migration. You came on what on Saturday or something? Sunday morning, I think. Okay. Yeah. So we are both on blue sky. 00:27:31.632 --> 00:27:45.968 So my first feel for, okay, this is actually working was Saturday night. I was watching a soccer match. I was having actual fun conversations about the match on Blue Sky, like replying to me. 00:27:45.968 --> 00:28:00.048 memes and posting memes and replying to comments and all that. I used to love doing this on Twitter way back when. But maybe because I don't pay for Twitter or for whatever reason, nobody actually looking at my tweets at all. 00:28:00.048 --> 00:28:13.503 So that was completely gone. I think I had moved over to Reddit for this sort of stuff. And this Saturday was the first time I was on both Reddit and Blue Sky at the same time having total fun with this. So yeah, it was great. 00:28:13.503 --> 00:28:21.776 Yeah, I can share my impressions of the social network. So I've used a couple of Russian social networks before going on Facebook. 00:28:21.776 --> 00:28:34.400 Then there was Facebook, then there was Instagram. But I came to those networks when they were already mature. Well, I guess sort of mature. Not like right now. Now it's like mainstream. There were people to talk to on those networks already. 00:28:34.400 --> 00:28:48.016 Google Plus was the only network where I started using as an early adopter. And then I think it died out before. It became, well, it never became mainstream, I guess. Did you ever use Google Wave? No, I missed that. 00:28:48.016 --> 00:28:52.463 Twitter, I tried multiple times. I could never grasp. 00:28:52.463 --> 00:29:06.768 Like what makes Twitter interesting? The short blurbs of text or like memes. For me, actually, it became interesting only after Elon Musk bought Twitter. I'm basically just like following his account and I'm looking at all those memes. 00:29:06.768 --> 00:29:18.064 It's gotten a lot darker recently and also a lot more politically motivated. But overall, for me, it was just basically like the place where I see what Elon is doing. The feed was always overwhelming to me. 00:29:18.064 --> 00:29:29.887 Then the threads came out when they wanted to kill Twitter. And threads was just like Instagram, but without the pictures. Well, sometimes with the pictures too. Because they were reposted from Instagram. Or like cross post. Yeah. 00:29:29.887 --> 00:29:34.271 I still use threads. I want to say a few times a week I'll open it up. 00:29:34.271 --> 00:29:48.624 It feels like the National Geographic or REI. For me, that's the kind of stuff I get is hiking or fall pictures or winter pictures and all that. Animals. And it feels very much like a nice... 00:29:48.624 --> 00:30:02.943 magazine, but it doesn't feel like people or small community of people, like it doesn't feel like I'm talking to somebody. It feels like I'm browsing a nice magazine with National Geographic photos in it. It feels like you're being broadcasted. Yes. 00:30:02.943 --> 00:30:11.503 It doesn't feel like I'm actively talking to somebody in front of me. Whereas that was the early Twitter and now Blue Sky is more like that, yeah. 00:30:11.503 --> 00:30:25.152 I'm personally a big fan of the channels approach. Like Telegram has channels. You subscribe to a channel. A podcast is essentially a channel. You go to the podcast and you see what you want to consume, right? Whereas those other networks, they have the feeds. 00:30:25.152 --> 00:30:33.215 I mean, Blue Sky also has a feed, but here's how I feel it. So when I came to Blue Sky, I think it was yesterday, right? 00:30:33.215 --> 00:30:47.663 And it feels like a bunch of people who are assigned to the same classroom and they are just trying to meet each other. So they are sort of in this together because they're in the same room. But it's like a very large classroom, maybe like a university class, like a thousand people. 00:30:47.663 --> 00:31:01.119 very large. But it's before clicks started to form. And then you start to meet people, just comment on random people's posts, just go and talk to random people at an event. You don't know them, they don't know you, but everybody is there to meet. 00:31:01.119 --> 00:31:06.160 Yeah. You should post this on Blue Sky. I think this is a great analogy. Yeah. Love it. 00:31:06.160 --> 00:31:20.463 And this feels so good compared to other social networks like LinkedIn or even Twitter, I think. It's overrun by people who want to get something out of their audiences. The influencers, right? I don't have anything against influencers per se. 00:31:20.463 --> 00:31:34.127 But like on LinkedIn, I always feel like I'm being pitched something. You don't even realize that you're being pitched something sometimes. Yeah. And sometimes also you look at some post, it has whatever, hundreds of thousands of likes. And it's like... 00:31:34.127 --> 00:31:37.776 dumb as hell it's like the going gets tough the tough gets going 00:31:37.776 --> 00:31:52.079 You know, it's like this kind of banality, is it the right word? It's like cliche, right? So things like this, and it gets like hundreds of thousands of likes and reposts, you know, like I'm starting to lose hope in humanity. I'm like, this is just so obvious. If this is the kind of content that goes viral. 00:31:52.079 --> 00:32:02.816 I don't know if I want to be in that network. Whereas on Blue Sky, what I really liked, Blue Sky has a really nice growth hack called the Starter Packs. It's just so funny, like in my feed. 00:32:02.816 --> 00:32:17.008 Everybody in their dog is creating a starter pack of indie hackers. I've seen probably at least 10 of those already. So a starter pack is basically a list of people who do certain things. You can follow them all by opening the starter pack. 00:32:17.008 --> 00:32:26.240 And like every couple of hours, somebody else would be like, oh, I'm starting a starter pack of indie hackers or bootstrappers. The reason why people do this is they put their name on it. 00:32:26.240 --> 00:32:32.240 In that way, I'm probably a bit lucky because I was on Blue Sky before this wave started. 00:32:32.240 --> 00:32:46.511 Because I had astronomy and my soccer and my indie hacker starter packs like way before. Like I didn't create them, but I found them before the mass migration happened. Now the thing is like, you have to be on them. 00:32:46.511 --> 00:32:56.688 So by the time Justin Jackson joined, for example, or he became more active, he was in one of those packs and he and a bunch of other people, right? That's how they got to a thousand followers really, really quick. 00:32:56.688 --> 00:33:08.016 Oh, no, no. I'm not saying I have many followers. I'm saying I found the right people to follow before this happened. And through them, I'm getting the right sort of content. But right now... 00:33:08.016 --> 00:33:22.352 I'm following anybody and everybody whose account I see and I find even mildly interesting. I'm like, yeah, like you said, right? It's like a university campus. You're meeting people for the first time. I'm giving everybody a chance that like, okay, you post good things and I'll... 00:33:22.352 --> 00:33:35.567 keep following you yeah yeah i think i followed somebody through one of those starter packs because i followed a bunch of those and i have like 250 people that i follow through those starter packs mostly and somebody posted something like what are they working on today 00:33:35.567 --> 00:33:49.903 And I'm like, I'll just respond. And I was the first one to respond. I'm like, I'm writing unit tests for regular expressions, right? And he actually had a very good point there. He's like, you can set up monitoring for the RSS feeds for different formats and run those episodes. 00:33:49.903 --> 00:34:04.112 descriptions through ai through an llm and ask it does the regular expression that we have still do what we needed to do for this particular episode i'm like hmm this is interesting so like we don't use ai to parse it but 00:34:04.112 --> 00:34:18.831 can use it to sort of monitor whether it's still doing the right thing because we can't detect it otherwise because it needs some smartness in there anyway so but this was the discussion i had today and i'm like where else would you have this kind of discussion in our slack maybe 00:34:18.831 --> 00:34:27.423 Yeah. Four people DM'd me in Blue Sky just saying, hello, and what are you building? There was no selling. 00:34:27.423 --> 00:34:41.744 One person I found out is from like Pakistan, very similar background to me, who's in Vancouver building something of their own ex-Google. It was like amazing. This is the kind of connections I used to find on Twitter, like maybe eight, 10 years. 00:34:41.744 --> 00:34:48.512 years ago. Yeah. Actually, I only got one DM and that person invited me to subscribe to his newsletter. 00:34:48.512 --> 00:34:57.327 Well, there's always going to be those. So, yeah. Yeah. And I also saw a post where somebody was asking the person tagged blue sky. 00:34:57.327 --> 00:35:10.175 And they're like, can you add the count of views to the posts? And actually I respond to that. I'm like, I actually don't want the count of views because when you have impressions, it starts to drive the influencer sort of behavior where people start to optimize for reach. 00:35:10.175 --> 00:35:23.295 Like, how can I game this? That's what they start to think about. Maybe this was the underlying motivation for that person's post too. I don't know. I kind of actually enjoy that it does not have the impressions. It removes some of that anxiety when you post stuff. 00:35:23.295 --> 00:35:38.000 Yeah. In related news, it looks like X is removing that showing the number of likes and all that views from tweets. I haven't been on X for a week or so, but apparently you can swipe your post and hide those things now. 00:35:38.000 --> 00:35:46.751 Cool. So yeah, we are both on BlueSky. We are trying it out. And you can follow us by tapping the link in the episode description. 00:35:46.751 --> 00:35:58.784 Yeah. And I think our handles, we're going to switch our handles to use Metacast domain custom domain, which is actually another really cool thing because it's the open protocol. And the way it's built is... 00:35:58.784 --> 00:36:09.568 If you own the domain metacast.app, you can actually own the handles in blue sky with metacast.app and somebody else can't like impersonate you. 00:36:09.568 --> 00:36:24.239 Right, yeah. Mine is ilia.metacast.app. I think you haven't set it up. Well, you will have set it up by the time this episode comes out. It's going to be rnap.metacast.app. Yeah, but it shows the affiliation of you with the thing that you're claiming to be yours. 00:36:24.239 --> 00:36:33.103 which is really cool. Yeah. And it's open protocol. So lots of people are building interesting things on top of it, I think, like the early Twitter again. 00:36:33.103 --> 00:36:40.592 Twitter had a lot of cool things around it, yeah, before they decided to shut down third-party API access and all that, yeah. 00:36:40.592 --> 00:36:54.960 A lot of indie hackers are doing follower growth charts and stuff like that. So it's exactly the thing that you're observing on LinkedIn when LinkedIn starts to get more mainstream. All those vanity metrics and gaming the algorithm, all that stuff is coming to blue sky. 00:36:54.960 --> 00:37:00.335 faster than to other networks because there is already a playbook and the pattern established elsewhere. Yeah. 00:37:00.335 --> 00:37:13.295 But I'm glad I get to see like cat pictures and I don't know, underwater whale pictures and all of that in my feed right now. And it's kind of vibrant. So that's cool. Yeah, I've seen some space pictures. It's really cool. 00:37:13.295 --> 00:37:23.615 So let's conclude this with wishing people Happy New Year and Merry Christmas, I guess. We wish you a Merry Christmas. I won't sing the rest of it. 00:37:23.615 --> 00:37:31.007 Yeah, because it's going to be the last episode of the year. Well, if it's not the last episode of the year, I don't know if we'll be actually in the mood to record another one, so... 00:37:31.007 --> 00:37:45.791 I'll just declare this to be the last episode. If this is not the last episode of the year and you are going to hear from us again in 2024 through this podcast, it's either very good news or very bad news. Yes. 00:37:45.791 --> 00:37:54.639 And if it's very bad news, we'll probably be too busy sorting out whatever happens. To record a podcast, yeah. So what are you reading and listening? 00:37:54.639 --> 00:38:08.976 I've been listening to a lot of Rogan recently. I've gotten some flack for that on LinkedIn the other day. Somebody I know, he's like, do you listen to this moron Joe Rogan? I decided to not even respond. Basically, whoever is listening to this, 00:38:08.976 --> 00:38:23.791 you might have opinions about Joe Rogan. But I did listen to a bunch of his more recent interviews. And it's kind of really nice to just like go out for a walk and just listen. Because his podcasts are less of an interview, more of a conversation. 00:38:23.791 --> 00:38:38.128 And I know people have different views about Rogan's views. Frankly, I don't care anymore because the cool thing about podcasts, you can listen to whatever you want to listen to, right? Your listening time is between your ears and AirPods. 00:38:38.128 --> 00:38:52.335 you and don't let anybody tell you what to listen to or not to listen to something. Exactly. And besides, it's the most popular podcast in the world. That's the thing, right? So I think there is always a vocal minority. Actually, I think Niklas Taleb talks about this. 00:38:52.335 --> 00:38:56.927 in the skin in the game book. There is a thing called tyranny of minority. 00:38:56.927 --> 00:39:08.751 where there is a vocal minority that goes against the grain and then they make everybody conform to their views because they were the vocal ones, even though actually the majority just might not care enough. 00:39:08.751 --> 00:39:23.023 So there is a certain group of vocal people who don't like Rogan, and they'll create this impression, especially on social networks like LinkedIn, that Rogan is persona non grata. But if you look at the stats, it's the most followed and most listened to podcast in the world. The thing that I found the most... 00:39:23.023 --> 00:39:34.367 interesting about the Rogan podcast recently. He published an interview with Donald Trump on October 26th. I think just after the election, so like two weeks in, maybe three weeks in after the publication. 00:39:34.367 --> 00:39:39.391 It became the most downloaded podcast episode of 2024 on Apple Podcasts. 00:39:39.391 --> 00:39:53.711 Yeah, I know you don't care. And I told you already, like you should not care about other people's opinions. But since we are recording the podcast, I'll also give you my opinion on this. I used to be a big Rogan. 00:39:53.711 --> 00:40:06.096 fan i actually still like the way he runs the conversations but i think increasingly over the last couple of years the number of crazy conspiracy theories 00:40:06.096 --> 00:40:20.527 It's not necessarily he doesn't bring it about, but his guests bring it about. And he doesn't push them back far enough. He doesn't ask the hard questions. So I don't like that side of it anymore. I don't listen to all of it, but I did. 00:40:20.527 --> 00:40:34.735 listen to the Donald Trump one. I actually think that was a nice conversation. And I think Kamala Harris, I know he reached out to Kamala Harris to do a podcast episode. I feel like she should have done it too. It would have allowed people 00:40:34.735 --> 00:40:39.615 to get that nice conversational style of your personality out. 00:40:39.615 --> 00:40:53.311 Anyway, again, ending it with that, you should not care. What you're listening to is between your AirPods and your head, and there's nobody out there looking at it, or even if they are, don't care about them. Just listen to whatever you like. Just keep an open mind. 00:40:53.311 --> 00:41:07.632 That's it. And I'm not talking about you, Ilya, just anybody who's listening to this podcast. Actually, one thing I wanted to bring up, and maybe I say this and we move on to what you're listening to. Some of the conspiracy theories that he gives the spotlight. 00:41:07.632 --> 00:41:18.320 aren't conspiracy theories, but they're labeled as conspiracy theories. And I don't think we should go into the details because it just became so politicized. I don't agree with everything he says, right? So like, don't get me wrong. 00:41:18.320 --> 00:41:32.592 But I also do agree with a lot of things that he does say or he does say, right? Not all of them by any means. I probably only listen to like 5% of what he publishes anyway. In many ways, I'm grateful that there are people like him and also like Lex who don't... 00:41:32.592 --> 00:41:46.800 Don't hesitate to actually talk about very difficult subjects that they will be labeled as conspiracy theorists for or like cancel, but they are too big to cancel. I guess Elon is one of the people who are in that category. He also can say all sorts of crazy stuff. 00:41:46.800 --> 00:41:53.552 right? Not all of the stuff that he said is crazy. Yeah, I think it's up to every individual to decide what they want to take in. 00:41:53.552 --> 00:42:04.063 Yeah, we as a society have a very binary, like you're either left or right, you're either blue or red kind of lens increasingly. And that does bother me. 00:42:04.063 --> 00:42:17.039 I do love Lex's episodes. And some of the actually most interesting ones are, I forget the person's name, but this is the guy who has a Netflix series on... He's a Graham Hancock. 00:42:17.039 --> 00:42:26.255 Yes, yes. He's widely debunked by the anthropology society as like, this doesn't make any sense, but... 00:42:26.255 --> 00:42:33.871 That episode was actually very thought-provoking and interesting on Lex's episode about a month back, I think. So... 00:42:33.871 --> 00:42:48.239 What am I listening to right now? Coming towards the end of the year, it's all sorts of sports stuff going on. Formula One season is winding up. There's two more races left. Tennis has Davis Cup. 00:42:48.239 --> 00:42:58.047 all that. As we speak, the finals are going on. And one of my all-time favorite players, Djokovic, actually announced Andy Murray. 00:42:58.047 --> 00:43:10.096 one of the other all-time greats as his coach. And the whole tennis world is going like, what? So I'm all into sports. And very soon, in another week or so, all of this will be over. 00:43:10.096 --> 00:43:24.463 I'll be deprived of sports until about March or so. Because I don't really follow the football, like NFL and all that. I watch it once in a while or hockey that much. I would watch it if there's like a playoff game or something. 00:43:24.463 --> 00:43:33.168 that I follow, soccer, tennis, Formula One, they all end around November and then they'll start around March or something again. Cool. 00:43:33.168 --> 00:43:45.599 All right. And on this note, I guess we wish everybody Merry Christmas. And a Happy New Year. And we'll see you in early January. All right. Thanks for listening. 00:43:45.599 --> 00:44:00.192 Thank you. Where can people get our app? Oh, on our website, of course, metacast.app. Well, if somebody is listening till now to this podcast and they don't know where to get our app. 00:44:00.192 --> 00:44:13.679 They're fired. Well, they should send us an email at hello at metacast.app. No, hello at metacastpodcast.com. Right, hello at metacastpodcast.com. Do we also have hello at metacast.app? We don't, right? No, we don't. 00:44:13.679 --> 00:44:22.735 Okay, we should maybe, just so we don't confuse ourselves once in a while. Maybe, yeah. All right. Thanks for listening. See you next year.