AI Zeitgeist: Birth of a Podcast

0:18 Hey, welcome to the A.I. Zeitgeist. I'm Austin and I'm Luke and this is us following through

0:26 on a promise we made to each other about a week ago at a friend's kid's birthday party.

0:35 Exactly. And this is a friend that we both know for different reasons and they've got

0:43 three kids so we see each other about three times a year and that's the only time we talk

0:49 and every single time we just get into this A.I. conversation and talk to the whole party

0:55 about A.I. Yeah, we monopolize the party. Yeah, and my wife just looks at me like, stop

1:02 talking and pay attention to what's going on. So we're like, okay, maybe we should pay

1:07 more attention at their birthday party and we can take these conversations and make a

1:11 podcast. And that was the idea of the party. We were like, yeah, Luke was about to leave

1:15 and I was like, we should like jokingly, we should start a podcast. And he's like, yeah.

1:19 And then we're like, wait, are we going to do this? And it was like, the thing is that,

1:25 you know, three, four years ago, if you had said, let's start a podcast, man, like think

1:32 about all the things you had to do to like get that set up and the amount of lift and

1:36 the weekly commitment. But, you know, the reason that we were able to jump on here just

1:42 one week later to record our first podcast, Austin was able to build out an app using

1:47 A.I. that basically is letting us create this podcast in an automated way. So that way we're

1:55 actually able to just focus on recording the content and then all the downstream stuff

2:00 gets handled. But I'll let you talk about the specifics of what it's doing.

2:03 Yeah, so we're both we're not in the same place. We're both at our houses and our kids are

2:10 asleep and we're both using good microphones and it it records both of our mics locally

2:17 at each of our computers and uploads like every 10 seconds chunks of the audio to a

2:22 server and then it mixes our voices together and then sends it to open A.I.'s whisper to

2:29 transcribe what we're saying and then it will edit the whole podcast and cut out things

2:36 that we don't want in there and any silences and stuff like that. So it's our idea was to

2:43 have it as as little effort as possible.

2:46 Yeah. Like you even have it set up so like it's drafting descriptions and stuff, too.

2:51 Yeah. Yeah. It generates everything for social media posts and stuff. And I'll probably like

2:57 the cool thing with A.I. is I have experience doing some of the stuff and now I'm trying to

3:05 have A.I. do as much of it as possible. So I just use cursor and kind of just one shot at this

3:11 and then it was pretty much ready to use right off the bat. But then I iterated on it, added a

3:17 few features. But but I'm finding with A.I. like even with my technical background, that sort of

3:25 gets in the way. And so I found that it's easier for me to like I'll type out a prompt and I'll have

3:31 too much technical detail on it. And so then I'm like, wait, I should just delete most of this and

3:36 have A.I. do it and see what happens.

3:38 OK, so cursor that sounds complicated and everything. What is like a way that somebody could kind of get

3:44 their feet wet? How would they jump in on this?

3:47 So probably the easiest thing is, I mean, a lot of people are using A.I. in some form or another with it,

3:54 whether it's like ChatGBT or Gemini or whatever. And so you can just start ideating with whichever A.I.

4:03 you use and just say like, hey, I'm thinking about making this thing. How would I even start? I have no

4:07 clue. And the cursor is really cool. Like my brother in law, he doesn't really have any background in

4:13 programming or anything, but I was talking to him about it and he's like, I want to make this like

4:18 Dungeon and Dragons thing so I can play with my kids and I don't have to be the Dungeon Master.

4:24 I don't I don't do that. So I think it's called Dungeon Master. And so he's he's been building this A.I.

4:29 thing to do all that part of it for him. And he's been going crazy and he's been using cursor and he's

4:37 been using GPT to tell him how to use cursor. And so he like asks different questions to get it going

4:44 in cursor. And he's built this pretty crazy thing. He's burned through a lot of tokens, though, and

4:48 he's already he's already hit his limit for the month, I think. Yeah, but it's the thing is that

4:54 that is the key right there. You don't have to know how to do it. You just have to be able to

4:59 ask the A.I., hey, how am I doing this? What's what's the next thing? What's what's how do I?

5:04 It's like you explain your goal and then say, like, how do I do what I'm trying to do? And then

5:12 it walks you through how to do what you're trying to do. And like you have to open up your mind to.

5:18 Like, I don't I don't know what I'm doing.

5:21 And so you just have to open your mind up to like.

5:23 Yeah, no, it's like exactly there's like what it is is you have to just have the the like this is

5:32 possible mindset, right? I don't know exactly what the way is to get there. Not like not like why

5:37 can't I do this? Right. Or I can't do this. But think of it like, why can't I do this? And ask

5:44 you or whatever? Yeah, it's like, how do I do this? Exactly. It's like you're sitting there

5:50 and you're thinking about what you wanted to do. And it's like, I have no idea how it's

5:56 going to do this. That's right. You don't need. Yeah, that's what a lot of people stop themselves

6:01 with. They're like, well, I don't have any programming experience or like I don't even

6:05 know that I need to be programming for this thing to work. And you you don't need to know. You just

6:11 you just need to be able to ask an A.I. how to do it. And it will tell you how to do it and tell

6:16 you the tools. Yeah. And then they'll actually do it for you. Exactly. One of my favorite thought

6:22 leaders, Andrej Karpathy. And one of the things that he says that's like really profound,

6:28 it's it sounds sounds silly, but it's like English is the latest programming language

6:35 like we are. Right. Yeah. Yeah, exactly. Like you just say what you're trying to do. So like

6:40 if you know how to write and you know how to read English, which I mean, pretty good bet that you're

6:46 literate if you went through school, et cetera. Yeah. But even if you're not like the A.I. understands

6:51 you, even if like you're not spelling things correctly or formatting words, you could use

6:56 voice text, you know? Yes. Yes. That's a lot of thing. A lot of people do that now. Yeah.

7:02 But so one of the things that is interesting about this whole phenomenon is that because

7:08 you can build stuff, right, it means that you no longer have to like get pigeonholed into the

7:18 standard application and you can make something custom for yourself. Like the first example that

7:23 we're talking about is the way that we're recording this podcast instead of going to

7:27 off the shelf application. Austin was able to build something that was custom, but we actually

7:32 needed for it to do exactly what it needed to do now. And there are there are tools like that

7:37 already in existence, but they cost money. And this was I mean, I'm paying for hosting for the

7:43 server, but that was something I was already paying for. Exactly. And it's it's not even

7:47 necessarily the money. It's more of the fact, you know, because that's part of it. But the other

7:52 part of it, too, is that you're getting a tool that is dialed into exactly what your workflow

7:57 is, which and a tool that I can keep adding to it. Like as we realize, like, oh, this doesn't

8:02 work out for us really great. I can add tools and tweak it for sure. So like that is the model.

8:09 And one of the one of the very small example. But the other day, one of my friends, he sent me a

8:16 application like a Google Play app. And it's like, this is a this is a way you can track your

8:22 your putting stats. I play disc golf, and so I'm trying to practice my putting.

8:25 And it was doing groups of 10. And like, I don't have 10 putters, I have three putters.

8:33 So I need to be in multiples of three. And it was able to create an application for me

8:37 that basically is able to let me do like that specific thing without having to like reconfigure

8:46 the way that I'm throwing right to like, match the way that everyone else practices.

8:51 Right, right. And then as you're using it more and more, you can decide like,

8:55 oh, maybe this would be helpful if I added a timer for whatever reason to see like,

9:01 maybe I buy a few more discs. And now it's like I get one more. Right, right. You can add Yeah,

9:06 you buy one more at a time. And yeah, you can step into it. Right, for sure. So but yeah,

9:11 I mean, just like it's, it's literally, you know, that that simple. And one other story,

9:19 like the other day, I was just watching somebody on their phone, and they were pulling up

9:22 information. And they pulled up the information that they were, it was a random question that

9:27 we're like, Oh, I wonder what, you know, when you, you have a question, you pull out your phone,

9:31 you Google it, right? They jumped into Google, and then they scroll down, and they click the

9:36 Wikipedia link, and they pulled up the Wikipedia page, and they're reading through the article.

9:39 Oh, yeah, this is the fact. And it was like, it felt like somebody had a landline rotary phone.

9:48 And they were making a phone call. Right, right. You know, it's like, to me, the way that I would

9:53 do that now is I would jump into a chat, and I would ask the question. And it would give me the

10:00 specific answer to the exact question I was asking, not like a tangentially related Wikipedia.

10:05 And you can ask follow up questions. And yeah, I drive a lot for my work. And so I'll ask,

10:12 I use Gemini a lot. And I'll ask it a certain thing. And then I'll just like spend the whole

10:17 drive, just talking to AI and learning while I'm driving, and asking follow up question after

10:23 follow up question. I mean, you if you have a curious mind, and you just want to learn,

10:28 like this is probably the best time to be born in. Oh, yeah, for sure. For sure. It's crazy.

10:33 I've been using, I've been using AI since probably like a week after

10:37 ChatGPT came out. And I use it an insane amount. Like, every minute of the day,

10:45 my wife is like, she'll like look, and I'm on, I'm on, I'm on like a chat with Gemini.

10:50 And she's like, you talk to that more than me. Well, it knows more than you.

10:56 So I want to I want to come back to that. But first, I think you had something you want to

11:00 talk about an application you built that you so you were doing like paging at your company?

11:05 Oh, yeah. So my company, yeah, my company, our phones stopped working, like Comcast or something

11:12 cut it off. And this is a big thing. So our phones route for like a week, and we got them back on.

11:17 But I realized maybe like a few weeks later that the paging for our, for our office, like,

11:24 so and so you have a call online to wasn't working. And that's when someone picks up

11:28 the phone and they they can talk over the loudspeaker. And so I was trying to get a

11:33 paging adapter. And it kept getting delayed and stuff. So I had this thought that I could build

11:40 a way for people to type in what they want to say. And then it would do a voiceover

11:45 over loudspeakers automatically. So I had it. I knew I wanted to use OpenAI's API to generate a

11:53 voice. And then I was also trying to think of like how what's the easiest way for them to input the

11:58 text. And I have thought to do it over Teams. And so I had Cursor help me, Cursor and Gemini

12:06 helped me create an application that is running on a laptop up by our amplifier. And

12:13 it integrates into Teams. So now people just type into a channel, they tag this pager,

12:18 and they type whatever they want. And it says it a loudspeaker. And I have all these settings for

12:22 repeating it, how many times you repeat it. And then I have a transitional word.

12:27 So it'll say like, Austin, call on line two. Again, Austin, call on line two. But it just

12:35 repeats the message and I can have it do as many times as I want. And it was just something like

12:39 I had a thought and I had it working like within a day, which I can't even imagine doing that two

12:46 years ago. But now you've given the AI brain a voice. A voice. Yeah. But everyone loves it at

12:56 work because no one wants to hear their own voice over the loudspeaker. And it's far clearer than

13:02 like people actually talking on a phone. So it's just, it's nice. It's pretty cool. It's crazy that

13:08 I could do it. That's true. Like you don't necessarily want to hear your voice over a

13:10 loudspeaker. That's a good point. Yeah. It's like the receptionist who does it the most.

13:15 She was like, I love this. And she's also, she's, she's got an accent because English isn't her

13:24 native language. And so it's probably even better for her because she can get out what she wants

13:29 very clearly to everybody else. That's awesome. Yeah, that's great. Okay. So I want to go back

13:34 a little bit. You were talking about using Gemini a ton and like you've been basically using it like

13:41 like power user, like since day one, basically I'm in the same boat. Like I jumped in kind of

13:47 around when GPT 3.5 kicked off. Like I saw the change from three to 3.5 and I was like, okay,

13:53 it's actually a thing. So I've been using it a lot. I have this like theory and we were talking

13:58 about at the birthday party. And this is kind of, I think what is like related to our name,

14:02 AI Zeitgeist. Yeah. Like everybody is in these chats, not everybody, but like the people that

14:08 are using AI a lot. Right. And there's like four, three or four big models. Like there's Claude,

14:15 there's ChatGPT, there's Gemini, there's Grok. Maybe there's a couple of open source models

14:19 that we're using, but those are the big ones. And they, they kind of have their own personalities.

14:23 Yeah. But they're all, but they're all kind of the same. Like they're, they have their own

14:27 personalities, but they all, they use the same technology to build them. So, and they're all,

14:32 they're all large language models. So they all kind of have a similar shape, even if they're

14:37 slightly different with different personalities, like they're all, they're all AI. And my theory

14:42 is like, we're all like thinking alongside of AI, like a co-thinking almost honestly. And so like,

14:50 I just, there's, there's like almost like a Zeitgeist, like a moment in time where everything

14:54 is like tuned to the same frequency where you can kind of like, I don't know, it's kind of

14:59 hard to describe. Well, you were, you were saying before, like how you'll have a thought about

15:05 something and then like the next day you'll hear your thought in a podcast. And so it's kind of

15:09 like, we're all getting the same, we're all talking with the same person and, and they're like crafting

15:15 our minds. And so there's like this, this mind hive or Zeitgeist. Exactly. That's the creepy way

15:20 to think about it. But I think one of the other explanations though, is that like, we're all

15:26 seeing the same information and we're all like processing it and through AI and that processing

15:34 is going to lead to similar like ideas because we're co-thinking with, with the, you know,

15:42 Oh, like context is a big thing. Like this is that the other, like maybe the idea of,

15:48 okay, we need to start thinking about coding agents, like whatever the topic du jour is,

15:54 there is like this point where there's like an inflection of something that's going on that,

16:02 that multiple people are hitting at the same time. And then they're all interacting with the AI

16:08 and it's helping them find very similar solutions. And then you start to see those

16:13 solutions happen simultaneously, like popcorn, like within, you know, days of each other.

16:20 Well, it's crazy. It's crazy. Like how, how fast are things moving? Like the, a new thing every

16:26 week, something new comes out and it's like the latest and greatest technology. And, Oh,

16:31 I was even hearing, like, we listened to a lot of different podcasts. And so like a lot of the

16:36 podcasts, they talk about different eras in AI and like, they're talking about like two months ago

16:42 was this era. And so it's like, it's just like weird how fast it's going. And like the things

16:49 they're talking about were in the past just happened like a few weeks ago. Yeah. And now

16:53 it's like old, old news history. A hundred percent. One of the, we were talking about

16:57 this earlier. And one of the things that I've been finding is that like tools and like things

17:03 that I've made that I relied on, like heavily for my workflow, like six months, three months ago,

17:09 I'm like dropping them not because they broke, but because like the next version

17:15 of chat GPT does it natively. It's so right, right, right. It kind of makes it

17:22 useless. And that's like, I think probably the most important thing when it comes to like the

17:27 speed of this technology is iterating is to stay flexible and like stay ready for it to take over

17:33 more of what you are doing. Like from a technical perspective, it may negatively be able to like

17:39 handle a lot more things like as it gets better. But the other thing that I think is important to

17:44 is to like understand what your core beliefs and values are, like have those set up pre ahead of

17:52 time because you don't want to be like completely unmoored and right. Have the AI influence you.

18:00 Right, right. So that's my advice anyway, because I think, I think we definitely are in a zeitgeist

18:07 and you want to still retain some individuality. Your humanity. Exactly, exactly. Yeah.

18:16 So I, in that, that mindset of things updating so quickly and like, oh, now Anthropic's the

18:22 best one. Oh, now Gemini is the best one. Oh, now it's open AI. I try to keep, like I'm using cursor,

18:28 which actually now is becoming kind of having its own model with, with,

18:34 with Grok. Right. And Compose. Yep. Yeah, Compose. And so like, but cursor, you can use

18:42 ChadGBT, you can use Anthropic's models and stuff. And so it's,

18:48 it's nice being in a program that I'm kind of invested in where I can do different models.

18:53 Whereas like, if you're using open AI, they have Codex. And if you're using Anthropic,

18:59 they have Cloud Code and stuff. And so what I'm trying to use in Buildout is,

19:04 isn't dependent on a specific AI company. Totally. So I can quickly like transition things.

19:09 One of the things that I've been working on in that same vein is I've been working on an

19:13 application that basically records a bunch of my notes using an open source,

19:19 the open knowledge format that Google released about a month and a half ago.

19:23 And it's basically like if you, the technical side of it, it's kind of like a skill MD,

19:27 it's got a front matter. And then it's marked down. So anyways, it's just a bunch of like text

19:32 files that the AI records. And what I'm able to do is I can use that with ChadGBT. And then I can

19:39 have Jim and I look at that same context. I can have Claude use that same context. So like exactly

19:45 what you were saying with Cursor. It's like whatever system you build, whatever system you

19:49 set up, whatever system you use, say, I'm going to use this one. Because they're all kind of

19:55 leapfrogging each other and things are changing every day, it makes the most sense to be as much

20:02 independent of picking one model to really run with as you possibly can be. Because you don't

20:08 know where it's going to end. Right. And the newest, greatest thing could be on a different,

20:15 from a different lab next week. You're like, well, I want to use that thing now,

20:19 but I'm locked into this thing. Right. Yeah. And I think that's the most important thing is to keep

20:24 all of your information, your information, because that lets you go anywhere you need to go.

20:31 So another project I've been working on using AI is I work on commercial sound systems at

20:37 grocery stores. And one of the hard things is like going around and trying to listen and see

20:44 if a speaker is working or not. And it's really hard to do with noise around. So I came up with

20:49 this idea to play a high frequency that humans can't hear over the loudspeaker. And I've built

20:56 this device that I've used AI heavily to design and to create software for that. It's got a

21:03 screen, touch screen and a microphone on it. And I can walk around the store and I can tap on a

21:09 map of the store and like create a speaker and I can record its volume, like the ambient noise

21:15 and that high frequency. And so I can go around to each speaker and do all this. And I can see,

21:20 I have like a visualization of what's going on with the speakers. And it's,

21:25 I know most people don't do this specific job, but it's something like really niche to me that

21:32 is like extremely helpful. And it's just something that I had an idea for. And so I made it.

21:38 So, and yeah, like typical software development, like there's something called TAM, like Total

21:44 Attainable Market, and they will go out and they'll say, okay, what is, you know, how many

21:49 people out there do commercial sound in grocery stores or retail stores? You know, what's the

21:55 number of companies, how then they'll figure out, okay, which ones are willing to buy specific

21:59 software. They're just going to do it manually. And then that now I have to think about marketing

22:03 and acquisition and customer churn. And then you put all that number into a formula and then it

22:09 determines whether or not it's profitable to make that software. Right. Even if it's even worth it.

22:13 Yeah. So like there's software that does not get made because it's not something that is going to

22:18 be, there's not a large enough market for it to be profitable. Or you could just make it yourself

22:22 with AI. Right. Right. So like, I've been, I've been doing this job for like probably like five

22:28 years and I had early ideas of this type of thing, but now like, now that AI is so, that I'm using it

22:38 so much and it's so much easier to use now, like this is way, I couldn't have imagined with my

22:47 And I think there's something there that's like broader than just this specific example.

22:54 Like there's going to be a long tail of software where there's a bunch of different niche cases

22:59 that are going to get specific applications built out by the users themselves. And that's going to

23:08 allow for people to be able to have like very specific tools to do the work that they're doing.

23:14 So things are going to get easier because you're going to have like very specific software that's

23:20 built specifically for the thing you're trying to do. Now, the one thing that I see a lot of people

23:25 try to do is they try to figure out, okay, what's the way to like cash in on this? Because obviously

23:30 it feels like the nineties with the internet, like, okay, there's, I know this is a thing,

23:34 like how do I get my.com like going? Like, but I think the, that's the trap there is to try to

23:42 sell something to other people when I think the best use of AI, at least right now, like as we're

23:49 kind of still figuring it out is to make something that's useful for yourself. And I think that's

23:56 probably the best place to start. Yeah, for sure. So I think that is a good place to land for our

24:05 episode. So we will wrap up here, but I'm super excited to jump into our next one. And I think

24:12 this has been a great start for sure. All right. See you next week. Bye. I left the wrong call.