Broke to the Frontier of AI

Transcript

Title: Broke to the Frontier of AI
Source: https://youtu.be/gvufVSE6RE8
Generated: 2026-04-27T08:01:45.381Z
Note: Speaker labels are heuristic and may require edits.
[00:00:00] A: The beginning of social media, all the, all the companies had no idea what they're doing, but then the kids did, and the kids were just good at social media, and then the company started hiring kids to, like, run social media. It's kind of the same thing. It's like, this is a huge wave of new technology. No one in the old world knows how to wield it. Uh, the people that do are kind of, like, concentrated at the labs, and then it's like, if you pick it up and play with it, you might become an expert at the next game. We should have
[00:00:25] A: ideas and experiments at the ready for helping people ease the transition, because it is likely that things move faster than people's ability to adapt. And I'm talking to Reed Hoffman, right, co-founder of LinkedIn, one of the earliest investors in Open AI. We talk for like five hours. Basically, Reed's just picking my brain about, like, this journey. The same conversation we're having right now is like, what have you learned in the last exploration of language models?
[00:00:51] B: Being broke is bad,
[00:00:52] B: but being stupid is what's really bad.
[00:00:55] B: And what's really,
[00:00:55] B: really bad is being broke and stupid.
[00:01:00] C: Welcome to episode eight of Young, Smart, and Battling Broke. Today's guest is Parth Patel,
[00:01:06] C: a builder and thinker at the frontier of AI,
[00:01:09] C: repeatedly reinventing his career and perhaps the way we all work.
[00:01:14] C: He took an econ degree from UCLA and decided against the traditional path.
[00:01:18] C: He entered startup land and started stacking skill after skill.
[00:01:22] C: He joined Clubhouse in 2021 as a founding data scientist,
[00:01:26] C: contributing to its wild rise to a $4 billion valuation.
[00:01:30] D: Facebook has taken bold moves into two new product categories.
[00:01:33] D: It's taking on Clubhouse with Facebook live audio rooms.
[00:01:36] C: When Clubhouse faced post-pandemic headwinds and had mass layoffs,
[00:01:40] C: he saw it as an opportunity to go all in on his hobby projects.
[00:01:44] C: He spent every cent he had made,
[00:01:46] C: including his 401k.
[00:01:48] C: And every waking hour to better understand the power of LLMs.
[00:01:53] C: A year later, he was appointed by Reid Hoffman,
[00:01:55] C: LinkedIn's co-founder and early OpenAI investor,
[00:01:59] C: to lead frontier AI projects.
[00:02:01] E: This new world of AI avatars will certainly make a lot of people uncomfortable and by the way will certainly have some foot faults and errors but again what might we be able to make?
[00:02:10] C: And he's as knowledgeable about current model capabilities as anyone.
[00:02:14] C: you one.
[00:02:14] C: We get into his unbelievable bet on himself,
[00:02:16] C: the AI future he sees,
[00:02:18] C: and what this means for young people and careers.
[00:02:21] C: I hope you enjoy our conversation as much as I did.
[00:02:23] C: Thank you.
[00:02:25] C: Parth, I'm so excited to have you on, man. It's genuinely been...
[00:02:29] C: In the last two weeks occupying my brain,
[00:02:32] C: how many questions I could ask you to just like fully quench my AI curiosity thirst, man.
[00:02:38] C: Put very simply,
[00:02:40] C: you are one of the most experienced and esteemed vibe coders. But that's like the most simple, dumb explanation of what you do.
[00:02:48] C: So if you could just start with telling me what you're currently doing, Parth,
[00:02:50] A: Yeah
[00:02:50] C: that'd be wonderful.
[00:02:52] A: Yeah. Yeah. One of the first vibe coders vibe coding from before vibe coding was a thing.
[00:02:56] C: Mm-hmm.
[00:02:58] A: Yeah, that's definitely.
[00:02:59] A: A good amount of my day is generating code with language models.
[00:03:03] A: I don't say vibe coding anymore.
[00:03:06] A: I think because there's this like stigma around vibe coding
[00:03:08] B: Yeah.
[00:03:09] A: that it's not performant, that it's a little bit like it's more like.
[00:03:12] A: like it's a it's a more casual form of programming Mm um
[00:03:15] C: -hmm.
[00:03:15] A: And I'm, and I have friends that, like, we were kind of like, okay, what other work, like, like, it's like, if vibe coding is talking to AI, smoke and weed, and, like, sitting back while it works, then, like, what is the version of programming where you fire off seven or eight more of these agents on other problems, other projects in parallel? And, like, because that doesn't feel like vibe coding. It's like, if this thing is going to work for 45 minutes on my behalf,
[00:03:39] A: I could either sit back, I could go to the beach,
[00:03:41] A: or I could fire off another one.
[00:03:43] A: And now it's like, why have one when you could have 20,
[00:03:45] A: 30, and then the bottlenecks, the bottlenecks are different.
[00:03:48] A: I think when you get good at one, when you get good at making one thing with AI,
[00:03:51] A: then it's like, it's a question of like,
[00:03:53] A: do you turn that into like your break time or you turn it into like,
[00:03:56] A: okay, well, let's hit the gas pedal and see how far this goes.
[00:03:59] A: I have a very obsessive personality and I'm a gamer. So I kind of see this as like.
[00:04:05] A: why talk to one AI when you could talk to, like, a fleet, and
[00:04:07] D: Hmm.
[00:04:08] A: like, have that kind of working, um, in the background on your behalf. That's, like, my kind of, like, longer term kind of objective with working with AI. And, but it's been a, it's been, like, a, at this point now, like, four year journey for, like, I guess, like, since, like, three, we're in 2025, so 2022 November, so it's like three, three year journey since, like, Chat GPT, and, uh,
[00:04:33] A: all as the scaling paradigm i mean that's where i'm gonna ramble a lot so you're gonna have to try to keep me I'm on excited
[00:04:38] E: on for it.
[00:04:38] A: Yeah, you're gonna have to bring me back to the points you want, you hope that I make. But basically, like, um, when Chat GPT came out, I was working at a startup. I was working on Clubhouse, and I could tell that, like, oh, wow, like, this is, like, way better than C3PO,
[00:04:54] A: and C3PO is, like, fairly advanced technology. Like, you can talk to this thing. It, it can translate every language. It can also, like, use a computer if you allow it to. Um, at the time you had to copy paste what you were doing into Chat GPT and then be like, oh, make this, and then you write a program, then you copy paste that back into your IDE. And I didn't even know how to set up my environment, so I was like, hey, teach me how to run this program. It's like, okay, here's how you run it. I was like, well, I don't have Python. It's like, here's how you install Python. So you can just self
[00:05:20] A: teach yourself every single step up until the process of vibe coding and then vibe coding was coined at the beginning of this year but a lot of us have been coding with llms since the beginning i think the first commercial application of llms was github copilot Hmm and that came out before chat gpt and for a lot of people at least engineers first they were like oh my god like this yes they can finish the next block of code
[00:05:43] A: But if you extrapolate that out,
[00:05:45] A: it's like, can it build the entire program?
[00:05:46] A: Can it write a multi file program?
[00:05:48] A: Can it build Facebook one day?
[00:05:50] A: Like those questions of like,
[00:05:52] A: if you keep scaling this,
[00:05:54] A: where does that stop?
[00:05:55] A: And then for me, it's like, it's like,
[00:05:57] A: okay,
[00:05:57] A: well,
[00:05:58] A: let's get good at it while it barely works.
[00:05:59] A: And then see what happens. And then inevitably every eight months there's some, like, huge leap in the model capabilities, and then everything that you were doing that barely worked then either comes online or is, like, massively amplified. Um,
[00:06:11] B: Hmm.
[00:06:11] A: And so that, and that pattern just keeps playing out, playing out. So, like, what you were doing in Chat GPT is now baked into the coding environment. What you're doing in the coding environment, with models that had a context window of, like, 1.5 pages just three years ago, now they have a context window of like
[00:06:27] A: like 450 page books. So
[00:06:29] C: Mm-hmm.
[00:06:29] A: Like, they go from, like, intern to, like, competent co-worker
[00:06:33] C: To
[00:06:33] A: Yeah.
[00:06:34] C: a one prompt home run.
[00:06:35] A: to
[00:06:35] C: Yeah.
[00:06:36] A: zero shotting very complex problems that would have otherwise taken you three weeks, and watching that, like, watching the model progression and working with it. So it's like, you could call it vibe coding, but, like, getting to see front row, like, interacting with a model as it gets better more quickly than any person you've ever worked with
[00:06:55] A: has been just, like, a very trippy couple years. Also just asking it dumb questions and be like, hey, yo, like, strapping yourself to this exponential curve and seeing what you can learn. Um, me being not a, like, a full, not a software engineer three years ago, and I don't say I'm a software engineer, but other people call me a software engineer now, and I do spend most of my day software engineering. So it's like, it's not like the title is, I think, I think of titles as, like, kind of BS, but, like,
[00:07:22] A: when other people give you a title, and then that kind of means something, and then, like, you could, because you've earned it through the way you act and things that you're able to make. And I think there's, like, so people say vibe coder, and I think that, like, I like that word. I think a lot of people should be vibe coding. I think everyone is going to be vibe coding in some capacity, even if they don't know it. But the, like, there's this other version of this, which is, like, what might the top one percent vibe coder look like? How might they interact with the computer?
[00:07:50] D: Mm
[00:07:50] A: how
[00:07:50] D: hmm.
[00:07:50] A: Because that's, if it's anything like playing video games, uh, the top one percent, like, strategy gamer is never going to drop a game to someone that's in the 98th percentile, because the gap between the top one percent, like, just the top 200 players in a particular video game, the way they play the game, the level of depth that they have in thinking about that game is so different from the guy that's at the 95th percentile,
[00:08:11] A: or, like, you know, near the top but not at the absolute top, and the way they work, the way they think about the game is different. So I think about, like, yeah, vibe coding might be, like, the everyday person's experience with coding agents. And even now it's like, we're still so early that most people are not interacting with coding agents, or they don't realize they are. And then I think about, like, some of my friends, um, Adam Silverman
[00:08:35] A: and, like, the Agency AI team in SF. They were like, we should call it, we should call it hyper engineering. Like, this other version, like, what might that, like, power user version of programming with language models be called? We haven't named it yet. I mean, hyper engineering is one name. I like that name because, like, what's the version of this where you lean in and, like, you're like, I need more? Um, oh
[00:08:53] E: I really hope hyperengineering does not become a nice LinkedIn buzzword.
[00:08:56] E: That'd
[00:08:56] A: My God, dude, you
[00:08:57] E: be so gnarly.
[00:08:57] A: know the moment it goes a term
[00:08:59] A: both mainstream, I'm just like, okay, well, I can't use this anymore, because there's no longer, you know, you say agents, and that three years later everyone and their mom is talking about
[00:09:06] B: agents Yeah.
[00:09:07] A: So it's like, okay, why don't we just not say the word agent, because that's the only way you can stand out, right? So, like, I'm, like, working with a startup, and they're building an agent, but one of the main things is, like, the founder and I were talking about, it was like,
[00:09:18] A: what if we just don't say the word agent,
[00:09:19] C: Yeah
[00:09:19] A: and then, like, that ends up being part of the, like, way we stand out, because everyone's going to hear agent, and it's like, sounds like the same, the same thing. And I think there's a lot of fatigue there, like, trying to not sound like. I think that's another thing, that's maybe one of my character traits, is
[00:09:34] A: like, I'm, like, anti-establishment or anti-mainstream to a fault. Like, I, like, I'm an Android user. Like, if, the moment everyone is doing something, I'm like, all right, like, time for me to head out, like, time for me to find the next interesting thing. Um, so that, that's, that's definitely part of my personality, and it shows up in, like, how I make decisions. Like, every one of my friends was becoming a management consultant, and I was like, well, you guys all want to do that because none of you know what you want to actually do.
[00:09:59] A: I actually know what I want to do, which is I want to work at a startup and no one was like
[00:10:03] A: go become a management consultant, then you can go into startup. Like, that didn't, but me coming from the Bay Area, I, I could see that. That was clear to me. Also, even in high school, all my, all my friends wanted to become engineers. I was like, well, someone should do something other than engineering if we're all going to potentially work together. Um, so then I was like, oh, I'll go become a business
[00:10:21] D: I'll
[00:10:21] A: guy. Yeah.
[00:10:21] D: go learn how to talk to people.
[00:10:22] A: Yeah, I'll go, like, yeah, exactly.
[00:10:23] D: I'll
[00:10:23] A: Yeah.
[00:10:23] D: be that guy.
[00:10:24] A: I'll be the business guy. I'll go figure out how finance works. That was me in high school. Um, and then I would bring that comp sci skill set to the finance
[00:10:32] A: finance space, and
[00:10:32] E: Mm-hmm.
[00:10:33] A: so, like, I was the most automated of all the finance bros that you could ever imagine. And so, but then that lens of, like, of, like, like, being good at a computer and applying that computer skill set to a problem space that doesn't have very many people that are good at computers ends up being how I stand out in that space. Um, but
[00:10:52] A: Yeah, I mean, we're kind of rambling already.
[00:10:54] A: So let's,
[00:10:54] F: Yeah,
[00:10:55] A: Yeah,
[00:10:55] F: I'll try. I'll try my best
[00:10:56] A: Yeah,
[00:10:56] F: to reel us.
[00:10:56] A: Yeah,
[00:10:56] F: So about three years ago,
[00:10:58] F: when you did start this kind of, yeah,
[00:11:01] F: it doesn't sound like vibe coding what you did start this journey as. It definitely sounds much more like hyper engineering in past stories. And when we've talked about it,
[00:11:08] F: you were pulling 14 hour days sprinting on trying to understand everything you could about this new technology.
[00:11:14] F: Could you walk me through those early days and what kind of sparked that journey?
[00:11:18] A: Yeah.
[00:11:19] A: So ChatGPT comes out in November of 2022.
[00:11:22] F: Mm-hmm.
[00:11:22] A: And I was working at Clubhouse and there were people in the app talking about this became instantly the most popular topic in the app and almost every single country in every single language in the app, like ChatGPT was in the title of the room.
[00:11:35] A: And you would just join the room and people would be talking about how they're using ChatGPT.
[00:11:39] A: And I talked to a lot of users. That was, like, my thing on Clubhouse, was like, I just, I gotta make friends with people. It's like, you can talk to people all over the planet, so you now, you get to, like, meet people that you would never meet in the real world otherwise, thanks to this app. They, like, reduce the distance between any two people, two people on the planet, down to a conversation, and anyone can discover that conversation and join that conversation.
[00:11:59] A: that's magical it's a magic that will be rediscovered by another platform in the future but we had it and I was able to just go into rooms and ask people how they were using ChatGPT and there were so many moments everyone is using a different way like photographers artists etc like and then I had one moment where I entered a room with it was a it was a group of farmers from my home state of Maharashtra in India
[00:12:25] A: And they were talking about using ChatGPT to help plan their crop cycles.
[00:12:31] A: And I was just like, are you serious?
[00:12:34] A: Like, that means this is a computer.
[00:12:36] A: Like, for all intents and purposes,
[00:12:38] A: we finally have something that's like.
[00:12:41] A: Jarvis like but like anyone can have it's not just for the billionaire main guy and it's not just for Tony Stark yeah
[00:12:47] B: Protagonist, yeah.
[00:12:47] A: it's not just for like one guy using Jarvis and on the planet but like now we have this the beginning and it's like sure I was like oh man they're using it for that it's not very good yet
[00:12:55] A: but that means it must get really good. We have to make it much better. It's not, it's general purpose technology, which is why everyone can use it, and then everyone's discovering the use cases, right? Like, they're using it in a way that I could never have imagined. I'm, like, over here, people are like, oh, it doesn't write my emails well. I'm like, this feels like a skill issue,
[00:13:12] A: like, you are applying this in a way that is, like, low leverage. Meanwhile, other people may apply it in a way that, like, changes their entire, like, the way they, the way they live. And that's why it's important to talk to people, because, like, you're like,
[00:13:24] A: like, you're like, okay, I'm not even seeing one percent. Like, you, tip of the iceberg on, like, possible use cases of the technology, especially because the technology is a general technology. It's not like this is for business, this is for life, this is for art only. It's like, actually the intersection of all, it's, it's, it's like these models that have been trained on all of human knowledge are now a representation of the collective intelligence of the planet, of, like, of all of humanity, including people that died.
[00:13:54] A: that's amazing now you can talk to these like ghosts you can you can construct these ghosts and talk to them and draw from the model insights that you would not be able to draw from the people that live near you and I think about that a lot right so like I go to clubhouse to answer questions about things that other industries and like other parts of the planet I try to understand like you meet people to understand something that you wouldn't otherwise have in your own neighborhood
[00:14:20] A: and now the model is a kind of, like, artifact that we can also call on to answer questions. That it's like, oh, I don't have a tutor for this, but now I don't need one. Like, I'm not, I'm not at zero for not having someone in my neighborhood that doesn't know this.
[00:14:33] C: Yeah.
[00:14:33] A: I can at least start by talking to the model, and that is a huge unlock. So this is, like, a realization over November to March.
[00:14:42] A: And I knew people were using it for programming and some of my teammates left Clubhouse to go to OpenAI in the months leading up to ChatGPT. So I was like, oh my God, when ChatGPT came out, I was like, that makes sense. Like Clubhouse is a natural language data set. It's like people meeting all over the internet and talking to each other.
[00:14:57] A: And now you have language models,
[00:14:58] A: which is.
[00:14:59] A: kind of like in the same space. It's like the conversations we have, language models, it's a conversational experience. It speaks every single language, and, um, it just kind of brings us together in a different, um, way. So it made sense that people who worked on social audio products ended up working in frontier labs and making these products. Like, that makes sense in retrospect, but
[00:15:19] B: Mm-hmm.
[00:15:20] A: also, um, the model just starts getting really good really quickly. So March 14th of 2023
[00:15:28] A: they dropped GPT-4 and GPT-4 is like it's a multimodal model chat GPT had a context window of like 4,000 tokens at the time then there was a 3.5 turbo which had 16,000 context window and GPT-4 comes out and it's like 32,000 context window and I'm like well we are going very quickly like the the the if you think of the context window as like
[00:15:51] A: the conversational bandwidth of the model, or, like, the, the, like, the model's RAM. It's like, how much can it work with in one prompt? Like, that is quickly, it's just doubling multiple times. And I thought, okay, my dad worked on semiconductors his entire life, and I was like, yo, this thing went from, like, barely holding together a page of information to, like, working with multiple pages at the same time, plus getting vision, plus getting, like, you
[00:16:16] A: really good at programming, and in the span of months. And my dad was like, oh, this reminds me of computers, right? Like, the
[00:16:22] C: The
[00:16:22] A: transistors. Yeah.
[00:16:23] C: Moore's Law,
[00:16:23] A: 
[00:16:23] C: yeah.
[00:16:23] A: Moore's law exactly like this is gonna get better twice as good and you're not gonna think of how to use it every single time like you're not thinking about how to make use of all the increased capabilities as quickly as the capabilities come online and um so GPT-4 32k context and that was like a a version that wasn't even available I was like I was using the
[00:16:44] A: the smaller GPT4, and I was like, man, if they release 32k context in December, I'll name my firstborn child GPT4. Like, now, because I had, I had co-pilot systems that I was building. At this point I was building chatbots, connecting them to tools, and, um, they were, it's like, I could tell it to do one thing at a time and it would go do that thing, like, analyze this,
[00:17:03] A: search the web, write a file, but it would be, like, one turn versus, like, one action of, like, competent work. And then you're like, okay, well, can you get it to work by itself? If you tell it to write a plan and then to just keep iterating on that plan, that was working. And then it was like, well, actually, like, think step by step. The, all the prompting techniques were like, oh, you should also give these things to the co-pilots, they get better at doing the job. Like, think, if you allow it to think about the problem, and if you allow it to write code, it gets way better at solving the problem. And, like, okay, you're, you're
[00:17:32] A: these techniques are coming online, but then the model's capability also just increases, and then all those techniques start, like, flooding, or, like, you realize, you're like, every day, 14 hours a day, you're working with this, and you're still not discovering, like, the full range of its power. And then, um, we had layoffs at Clubhouse, and I, we had, the nice thing was, like, we got to keep our laptop, and we got four months of severance. Yeah.
[00:17:57] D: Wonderful.
[00:17:57] A: So I'm sitting there, and I'm like,
[00:17:59] A: well, this is like, this is perfect. Now I'm not moonlighting this after work. I can actually just spend all my time, all my waking energy for the next 120 days. I'm just gonna sit here and talk to this model and figure out what it can't do. What, what is it not useful for? Where does it fall short? Like, I'm just gonna do that, and then at the end of the 120 days, then I can decide whether to go get a regular job again
[00:18:22] A: or, um, you know, like, yeah.
[00:18:25] B: You want to marry the AI?
[00:18:26] A: It's like, okay, no, it's like, it's like either I figure out how to use this and it's like a useful skill, or I go back to my old, like, job and try to get a regular job. Yeah. And, uh,
[00:18:36] A: and, and it was interesting, because I met a guy on Clubhouse, and he was like, he was the guy when GPT4 came out that was like, you should be programming with this. So then I was like, okay, no more excuses. Like, I've been dodging programming my whole life, and then I was like, well, now this thing is just going to write the code. Like, all of a sudden everything became easier. So then I just sit there, I was like, we're just going to code with this. And of course I'm starting from zero. Like, not completely zero, I learned Java in high school,
[00:18:58] C: Mm
[00:18:58] A: but
[00:18:58] C: -hmm.
[00:18:59] A: it was so much, it was so much easier than anything I'd ever done in programming. Like, everything that used to take six months could be done in, like, three prompts, and I was like, okay, that, that's a crazy, like, you could tell it to, you know, teach me how to clone my voice. It writes the program, then you get the program running. You're like, two prompts, and you did something that if you were to try to build that from scratch would have taken a team. But because you're just leaning on the model to surface all the solved versions
[00:19:24] A: of the different parts of this problem, you're, like, accelerated through a lot of everything that's been done. You can, like, quickly go through, because the model is really good at stuff that's already been done. Of course, there's a frontier now of, like, how good is the model at things that no one's ever done?
[00:19:37] D: Hmm.
[00:19:37] A: And that's a frontier, but that doesn't mean that you can't just use it for everything we know. That's, like, not an impossible problem. That's like, doesn't, like, how many people are working on problems that actually require novel physics versus, like, solving a problem that
[00:19:50] A: someone else has probably solved before? The model being trained on everything that we've already done makes it so useful for just, like, copy pasting solutions from, like, existing other spaces. So
[00:19:58] E: Mm-hmm.
[00:19:59] A: they're already useful. And I was like, you know, you could, if you froze improvement at 32k GP4, I could still use this to, like, explore Mars. Like, I could still use this, and it still changes the world. But then if you tell me you're gonna, like, 100x compute, you're gonna 100x everything, and I'm like, okay, well, then
[00:20:18] A: a lot of these rough edges are just going to be, like, solved by scale, and then a lot of other techniques are going to come online. We're just going to bootstrap past a lot of these existing limitations. And sure enough, by December they released, I was expecting 32k GPT4 in 2023,
[00:20:34] A: uh, November, and they released 128k context GPD4, and I was like, okay, that's 50. So this is, like, so much bigger than I thought I was gonna get.
[00:20:43] F: And where is this in terms of your severance?
[00:20:45] A: Oh, that was, I am already burned
[00:20:47] F: through Okay,
[00:20:47] A: my severance.
[00:20:48] F: so you hit your four months, you hit your almost 1,000 hours of AI coding,
[00:20:51] A: Yeah, yeah, yeah.
[00:20:52] F: and
[00:20:52] A: I, because
[00:20:52] F: what was the mindset when you hit that four-month mark?
[00:20:55] A: Okay, so the good thing about, like, starting severance, I was like, I'm just gonna
[00:20:59] A: do this every day for 14 hours a day. Like, sit here, talk to the model, figure out what, like, what this is for, and then, you know, I'm gonna do that for 120 days. The nice thing is then, um, and I'm just gonna do it on the weekends too, because, like, when you are jobless, like, every day is the same, right? The weekend is only just a day that some other people are down to hang out and drink. It's like, it's like, but, like, your social construct start falling apart,
[00:21:25] A: especially when all you do is sit there and talk to AI yeah
[00:21:28] B: Yeah, there's a lot of way that those four months could have ended.
[00:21:31] A: It could be a lot
[00:21:32] B: I
[00:21:32] A: of you
[00:21:32] B: just imagine you hoodie in the basement just working
[00:21:34] A: That is true. Yeah, literally true. It's just like me sitting there
[00:21:37] B: EDM in the background.
[00:21:38] A: yeah blasting blasting music putting on mixes and then like the cats are just like what is he doing and then every once in a while just so then I run out of severance and I'm like
[00:21:50] A: I feel like I have more questions than answers.
[00:21:52] B: Mm-hmm.
[00:21:53] A: And I had friends that were seeing me because I just show my I had my friends from like previous jobs and I was like, guys,
[00:22:00] A: here's what I'm learning working with these models.
[00:22:03] A: This is how it changes definitely like data analytics,
[00:22:06] A: which is my bread and butter my like superpower I was like this thing totally changes data science and data analytics because instead of us writing SQL queries by hand building dashboards by hand you prompt the model to do both of those things and then that means that like I feel more like
[00:22:20] A: it, it used to be the CEO asked me and then I go build the analysis, but now I sit there and tell the agent to do the analysis. So it's like, I'm moving up an abstraction layer. And then also it's like, instead of relying on an engineered instrument, the thing, and, like, put the analytics into the code, I could just be like, add, instrument this, so that we can, you know, add logging, instrument this.
[00:22:40] A: So, like, I am both the engineer that implements the analytics code and the data engineer that structures the code. That was previously other people in adjacent roles to me,
[00:22:47] B: Mm
[00:22:47] A: but
[00:22:48] B: -hmm.
[00:22:48] A: now it's like I get to expand into all of them. So I become the CEO and the data engineer and the data analyst and the, the, like, engineer to some degree, and so it's an expansion of the role I'm seeing. So, like, even if I go back to data analytics, this is how I'm going to do it. But then you go back to, like, a bunch of companies, like, they used to work for, and you're like,
[00:23:07] A: you guys playing with language models? They're like, no, we banned it. It's like, all right, you guys are gonna get, you guys are gonna get rolled then. Like, like, you, you banned it, and I'm over here seeing how, like, I can't imagine playing the game without this now. Without the, in, you play this game without these technologies, and, like, you will be, you will lose to someone that plays the game with these technologies. Um, at least, like, just the old game is done.
[00:23:32] A: But then a lot of people just weren't, um, they just weren't. At that moment it made sense, because they had the day job, they had their obligations. They hadn't, you don't have three months to spend talking to AI, being like, oh shit, this changes everything. But then the people that were my friends were like, yeah, you have definitely discovered, like, the future of our roles. Um, I feel like you should just keep doing this. So then I was like, hey, this isn't insane. Like, if the smartest people I know are just like, you'll keep turning over these leaves, turn over these cards, like, because they
[00:23:59] A: benefit from knowing how their own role changes, then I was like, okay, cool, like, I think there's more to learn here. Um, and also I thought, like, can't go to school. They don't teach this to school. Like, the students would find, students from UCLA would find me at a hackathon, and they'd be like,
[00:24:15] A: you want to just give like an underground lecture at UCLA we'll just take a room we'll give you you could plug in your laptop and just tell us what you think is going to happen and I was like absolutely dude like there's no one who's going to update the curriculum as quickly as now there are my friends are like working with UCLA to like use agents to update the curriculum but over then it's like I was like I'm just going to show up vibe code the deck in the ride over there and then like explain what we're doing with these technologies by using the technology
[00:24:42] A: And then it's like, you know, the kids are kind of like, oh, but is it cheating? I was like, cheating? The real world is going to pay you half a million dollars to be good, good at using the stack. So
[00:24:50] B: Mm-hmm.
[00:24:50] A: Like, you're, you're cheating in the context of the old
[00:24:53] B: The
[00:24:53] A: world
[00:24:53] B: old rules.
[00:24:54] A: Yeah.
[00:24:54] B: Yeah.
[00:24:54] A: The old rules, the old game, but that game is no longer even the game that's worth playing. So it's like, you, if you were willing to reject that and, like, discover this new next game, then, like, you could become good at the next thing, and then
[00:25:08] A: Yeah, sure, most people don't know how to put a price on it, what it's, what it is. They don't know. But, like, if it is the future, then eventually they'll catch, they'll come around, and they'll realize, like, leverage, this is leverage, and then the rules will be defined, and then you'll be able to be, it's like, you know, in the beginning of social media, all
[00:25:26] C: Mm-hmm.
[00:25:26] A: the, all the companies had no idea what they're doing.
[00:25:29] A: But then the kids did and the kids were just good at social media and then the company started hiring kids to
[00:25:33] D: Mm
[00:25:33] A: -hmm. like run social media.
[00:25:35] A: It's kind of the same thing. It's like this is a huge wave of new technology.
[00:25:40] A: No one in the old world knows how to wield it.
[00:25:43] A: Uh, the people that do are kind of, like, concentrated at the labs, and then it's like, if you pick it up and play with it, you might become an expert at the next game.
[00:25:50] E: What an interesting analog.
[00:25:52] E: That's super fascinating for youngins entering these funky labor markets.
[00:25:56] A: Yeah. Oh, that's the way. You have to, you cannot go for yesterday's game.
[00:26:02] A: It's not even clear to me that yesterday's game is worth playing,
[00:26:04] E: Mm hmm.
[00:26:06] A: and it's, like, or even more valuable than the next game, but
[00:26:10] E: it's it's a funny one because I don't think the establishment or the traditional companies have these like new defined roles of the hyper engineering or something of the sort.
[00:26:18] A: No.
[00:26:19] E: So.
[00:26:19] A: I think every single company is going to should be hiring vibe coders and the number one request I get when I'm talking to people is how do I hire it's like one they try to hire me
[00:26:28] A: and I have a job but then then they'll be like okay how do I hire people like you and that's it's a very and then I was like oh shit like how would I find someone who's kind of in this mindset where is that person and it's hard I mean it's it's it's like then they try to hire my friends which is that's a cool thing is like I have friends that I've been bringing along on this journey people that I was bouncing these ideas off of a lot of them eventually like they might have experienced layoffs or they were like well I think I should just take time off
[00:26:54] A: To kind of go deep,
[00:26:55] A: figure out the next thing and then come back into the job market with this.
[00:26:59] A: new, new skill set. And, and because Parth kind of took that risk and mapped it out, like, I can, like, because I did that, I can just give you the, like, I can get you to level two, level three faster, because, like, you don't need to, you still need to put a lot of hours in to become good at anything, but you wouldn't be guessing as much, because you have a peer group that's, like, also exploring.
[00:27:19] A: So I knew I would I was like in the best case scenario,
[00:27:21] A: I do this work so that most of my network doesn't have to start from scratch and then I can kind of see the technology within the network and then the whole network becomes more resilient and then the longer term value is that if we're still right and that it is like an upgrade to the way we work and think,
[00:27:37] A: then six months from now you would be teaching me something that I didn't know about how to use the tech to create these feedback loops of community.
[00:27:44] B: So fascinating.
[00:27:46] C: Okay, but you still haven't hit on how you started generating revenue because business-brained Parth
[00:27:50] A: Yeah.
[00:27:51] C: couldn't
[00:27:51] A: Oh
[00:27:51] C: have been in this post-severance period like
[00:27:54] A: man
[00:27:54] C: I can just have fun with my AI in the basement for the next year and just keep on vibing. You were probably starting to get a little stir-crazy.
[00:28:03] A: Yeah. Three months, three months into not, uh, like, after severance, and three months into just burning my money. And I mean burning my money, like, the more interesting and creative my ideas were, the more compute I would burn,
[00:28:14] A: burn trying to achieve that. Or, like, you create a chat bot, you put it, put it in a group chat, and you leave it on all night, and then you realize you wasted, like, two thousand dollars on generating the word, it's just saying thank you to each other. It's like, oh shit, like, what did we learn here? It's like, okay, maybe you need a manager in the room to, like, end the conversation so we don't spend all this money. And I was like, did I just spend rent money? Like, oh man. Like, so then they look, the lessons became expensive, and then my friends, one of my friends, like, Part, just, like, pick up coins along the way,
[00:28:42] A: like, turn this into money. My mom was saying the same thing. She's like, everyone's gonna catch up to you. And then a couple months later I was like, no one's gonna catch up to you, because you don't have a life, you don't have a wife, you don't have kids. Like, as long as you don't have a life when you gas pedal, like, it's gonna be, you're gonna have this lead, which you can't do forever. It's not sustainable. And for me it was like, I started, then I sold all my crypto, I
[00:29:04] A: And I don't recommend this to literally anyone.
[00:29:07] A: And in retrospect,
[00:29:08] A: even if I asked AI,
[00:29:10] A: it probably would have told me to like do something other than.
[00:29:13] A: early withdraw my 401k. Yeah.
[00:29:14] C: Yeah,
[00:29:15] A: 
[00:29:15] C: that's a little bit crazier than selling all your crypto when you just
[00:29:17] A: with
[00:29:17] C: say out loud.
[00:29:17] A: How many finance people we know, uh, like, this is the problem. Like, when I say, like, I'm anti-establishment slash mainstream to a fault, like, asking anyone in finance what the right thing to do in this scenario, they wouldn't have recommended this, and they might have suggested a lower risk kind of way to, like, just buy more time. And I guess I was in a state of, like, isolation of, like,
[00:29:41] A: I just need to move to the next like the next day I'm going to learn the next thing the next day I'm going to learn the next thing
[00:29:45] C: Yeah.
[00:29:45] A: and I wasn't thinking rationally about like uh I was like okay if I just sell all that I still have x number more months of like play time and and I was also like you know you go to the
[00:29:59] A: and you're like, okay, well, are you insane? It's like, yeah, probably, but also, like, you have figured something out, and you can't go to the regular, the real world is just not even paying attention, really, outside of San Francisco, to what this means. And, uh, you can just get better at this tech. Like, then I was like, okay, would I rather a year from now have no money to my name, like, literal net worth zero, and be good at using this,
[00:30:24] A: or would I rather have like a regular job and not know any of this stuff and I was like I would much rather have no money but be good at using this because then someone's going to probably pay for me to think like this and that was like
[00:30:40] A: like, once you say that and once you believe it, and then it's like, okay, cool, like, we're gonna go until we run out.
[00:30:45] B: your one decision to make a thousand, basically, like, that had decided what the next 100 days would look like, make doing that
[00:30:52] A: Yeah.
[00:30:52] B: calculus that's
[00:30:52] A: Yeah, yeah.
[00:30:53] B: so interesting
[00:30:53] A: And then there are other mantras, like, you know, you're talking to the cats and you're just like, don't worry, guys, like, the boys ain't gonna starve. The boys will have wet food once again. And it's like, with that mindset, is more like, look, we're not gonna start, like, okay, let's say even if I run out of all my money,
[00:31:09] A: Okay,
[00:31:09] A: worst case scenario,
[00:31:10] A: I could go ask my friends, you know, if they need an analyst,
[00:31:14] A: like what I was good at before AI,
[00:31:17] A: I would be 10 times better at now.
[00:31:20] A: So like, I feel confident in at least getting an amplified version of my old role.
[00:31:25] B: Yeah.
[00:31:25] A: Right?
[00:31:26] A: So like, we will make money if we need to make money,
[00:31:28] A: we will make money.
[00:31:29] A: It's just that I would rather
[00:31:32] A: be in a headspace every single day of using the new capability than playing yesterday's game. Not, not everyone can do that, like, and not everyone has enough time or learns quickly enough to skip to the next game, even through experimentation. Um, and I noticed this a lot, because, like, um, like, especially if you have, like, a wife and kids, like, you don't have nearly as much time to learn and explore. And I thought of it as, like, and I was like,
[00:31:59] A: had another thing you tell yourself to justify lighting all this on fire, is like, oh, they don't teach this. Like, my mom was like, maybe you should just get an MBA. I was like, then I definitely don't learn this. Like, what, like, get an MBA, that's down to lose money.
[00:32:12] B: Learn to
[00:32:12] A: Yeah.
[00:32:12] B: smile more.
[00:32:13] A: Yeah, it was like, then I lose money, and it's like, I have not, I definitely, they don't teach this in MBA, and I would lose, lose my money. And, and, uh, I say lose my money, it's not that I don't value an MBA. It's just I've never in my life been in a room looking around, like, gosh,
[00:32:26] A: oh shit, if only we had someone with an MBA. Like, that's never happened. Um, but then I was like, like, okay, they're not teaching this in school. So then I was like, okay, another way to think about this is, like, I'm investing in my own education. Yeah.
[00:32:43] B: And when did we start seeing the ROI on that investment?
[00:32:47] A: Yeah, pretty much. I was like down to like two months of
[00:32:49] B: runway Okay.
[00:32:49] A: And, uh, my mom, my parents, I visited my parents, and they could tell, because I was, like, very honest, because, like, they were like, oh, why don't we, let's, we want to travel here, uh, in three months. And I'm kind of like,
[00:33:00] A: yeah, I can't really think about three months out, because I need to think about how to make money three months from now. And, like, I know you want to plan that trip, but I'm not going to be able to come, because I need to, like, figure out my life, right? And then they were just like, oh shit. They were like, what? Like, do you need help? Like, honestly, like, I know that my parents would have allowed me to move back home and just live in the garage if that's what I needed to, like, find my place. And maybe part of it is, like, my own, uh, just, I was like, you know, I will figure this out.
[00:33:26] B: Mm And-hmm.
[00:33:26] A: also I was optimistic because I had just gotten like the first like contracting gig started coming online where I was working I was basically like I wasn't completely alone I mean I was alone in my work but being in LA I would I would go to events all the time like every week I would go to like community events AILA largest community around AI in the city.
[00:33:50] A: And I made friends,
[00:33:52] A: I was introduced to the founder of AILA, Todd Tarazas, and he's amazing.
[00:33:55] A: He's a great connector.
[00:33:56] A: He calls himself the nerd herder because
[00:33:58] C: Hmm.
[00:33:58] A: he gets nerds together and we, and he's focused a lot on like the evolution of Hollywood.
[00:34:03] A: But at the same time, I'm focused on the like the automation and the coding agents and the like automation of cognition.
[00:34:09] A: And so we're, we're like hanging out.
[00:34:12] A: And like we're throwing events,
[00:34:14] A: we're trying to get people together to like learn what is the builder community,
[00:34:17] A: the creator community doing with this because we're not the only ones seeing this like.
[00:34:23] A: transformative movement. It's like, it's not in isolation. You get a bunch of people together, you realize you're not crazy. You're not the only crazy person. There's a couple other crazy people. It's a subculture. It's not mainstream yet, but at least we can create our corner of the city where we're experimenting. We have hackathons where people are using image and video models to tell short stories and make music videos. And, uh, so in LA the hackathons are, like, creative hackathons. Then you also have technical people here. And so, um, in, working on the community meant that people would see what my
[00:34:52] A: my abilities were. So you go to a hackathon, you get to know people, and they're like, oh, this guy has a skill set which isn't traditionally really defined yet. And then people would be like, oh, we'd love to learn how to implement this in-house. So, like, I knew that language models are useful, and there's, like, a, there's, like, a progression curve of an organization adopting language models,
[00:35:11] A: and I think, like, there's the level zero stuff of, like, let's just make sense of the information we have, create structure from all the unstructured data. Then, you know, level four or five is agents, but everything up until that point is, like, on the path towards becoming a more efficient company. So, like, the skill set is useful. Can you implement it? And so I was taking contracting gigs, and I was like, okay, great, like,
[00:35:31] A: my buddy John Milinovich, I worked with him at Clubhouse, and he was like, he gave me, he was like, Parts, you have finally, he's like, good, like, you have found a market for this new skill. I was like, great. He's like, now you just need to figure out what the price to charge is. I was like, how do I do that? So, just raise your rate with every new person you work with, and then when they say no, you may have discovered the neighborhood of the price. And so as I would get these contracting projects, I would just charge a slightly little higher amount, and then no one would ever say no. I was like,
[00:35:59] A: oh shit, like, this is actually, like, potentially, like, even today it's like,
[00:36:04] A: oh, dude, I could be charging, like, a stupid amount for this. But then I, for my friends, I just, like, let's just get you alpha, right?
[00:36:10] B: Yeah.
[00:36:10] A: Like, I would rather invest in you and then see the, the, the, the, like, like, the payoff for me is, like, where you are two years from now, having, now knowing this new game
[00:36:19] B: Mm
[00:36:20] A: exists
[00:36:20] B: -hmm.
[00:36:20] A: instead of me trying to, like, um, extract value from every single interaction that I have.
[00:36:25] B: I love that.
[00:36:26] A: Um, at least for my friends, right? Like, randos, maybe I'm down to charge people. Yeah, no, to two, maybe 20, maybe that's,
[00:36:32] A: maybe that's, that's the question,
[00:36:32] C: mm
[00:36:33] A: is like, maybe we should be, like, at BC, when I, we used to work, I used to work at BC. One of the first things I did was like, we should just slap an extra zero on what we're charging clients.
[00:36:44] A: -hmm, and it turns out they wouldn't say no. This is the whole
[00:36:46] C: it actually adds
[00:36:47] A: Yes.
[00:36:47] C: to your ethos, the price. Yeah.
[00:36:49] A: Yeah, they're like, oh, we got to take them seriously, because, like, they're actually charging us real money, right? But it's not like a donation to a bunch of students. It's like, if you think that's what you're worth, and you charge that, and then people actually pay it, and you deliver the value,
[00:37:00] A: you, like, it's win-win actually. But so that was, like, an exploration in pricing. And also I got a job, um, creating courses for Coursera, and they were like, we wanted you to teach. They basically brought me on to, to create a bunch of courses related to language models for data analysis. That's my core skill set. But then also language models for software engineering, which was like, whoa.
[00:37:22] A: I'm like not a software engineer but I'm teaching a course on language models for code generation and software engineering my buddy was like part one of the highest leveraged things you could do is be someone that teaches software engineers
[00:37:31] A: like, that's
[00:37:32] C: Hmm.
[00:37:32] A: a valuable, that's a very high leverage thing. And I was kind of just like, imposter syndrome. But then it's like, I have a course on LLMs for software, and at the time it was like, I was using coding copilots, but, like, the course was like, let's just do Chat GPT, or, like, let's just, you, how do we use an API? And I was like, so, very intro level stuff, like Chat GPT for code gen, and then, like,
[00:37:54] A: embedding models and retrieval systems, which is, like, very easy for me at this point because I've been playing with it for, like, so long. But I was also like, oh, this shows, like, the, the, like, we need to, I need to come back to people's, like, level and give them the, like, first thing, the first piece of leverage. What's the most important first thing you learn, such that you might discover for yourself all of these other things? Because I can't tell you what the most important thing is, but, like, if I give you the first couple wins,
[00:38:20] A: the model will reveal to you a lot of these other things that even I haven't seen and so but the doing the Coursera was like when I say it's like is it better than the influencer course selling thing
[00:38:31] A: I didn't own those courses.
[00:38:33] A: Coursera owned them.
[00:38:34] C: Basically, it's just less profitable.
[00:38:37] A: Yeah. So I get the stamp of Coursera on my bio. And actually, like, I didn't put it in my bio because I was kind of like I was kind of just jaded by I was like, man,
[00:38:44] A: I did something very useful, but I don't own a stick.
[00:38:47] A: So it's like I don't even promote the course because I don't get paid more for more people taking the course.
[00:38:51] A: So like I didn't even put it in my bio because I was kind of just like jaded by this.
[00:38:57] A: Coursera, if you're listening,
[00:38:58] A: you might want to change that.
[00:38:59] A: Creator revenue model. Um, course creator revenue model. I mean, honestly, like, if you want to get a good teacher and you want them to stick around and really contribute, you got to give them some kind of vested interest in contributing to the course. But then, uh, then you go to a language model and you ask it, okay, who is Parts Patel? Two years ago it would just be like, there's, like, nine Parts Patels out there, and it would just blend them all into one guy. Then it was like, there is one Parts Patel
[00:39:25] A: notable for working with language models and he has a Coursera course I was like damn I didn't put that in my bio like the language model could go find those details and then I was like okay I'll put in my bio so that the language models know because the language like because then when you have a conversation with language models like who is Parth and then it's like it's nice that the models know that I have Coursera courses even if I am not personally promoting them it's just like
[00:39:48] A: it does lend credibility to work with a notable brand. Um, I do think that if I did the same course as an actual create, if I build a creator brand and build my own personal brand and I sold those courses, I would make way more money. Yeah. Um, that was the lesson there. The lesson there was like, okay, like,
[00:40:06] A: you built a course, and it's on someone else's platform, and you don't even own a percentage stake. Now your motivation isn't there. You're kind of jaded. But actually, like, it's like, okay, then I was like, okay, I gotta build my own brand, actually. Like, it's more important to build. So that's actually why I came to work here with a bunch of creative people, is, like, creators in LA is like, they are, like, their business is the brand. The, their business is the creator journey. And I am at the very beginning of that, of, like, realizing that, like, okay,
[00:40:32] A: okay, you don't have to build everything on other people's platforms, and actually you should have a corner of the internet that's your own, especially because the language models will be surfacing you in conversation. And it's like, how do you become the answer to the question of who someone is looking for? And so that, that has been an interesting thing of, like,
[00:40:49] A: unless you own your own presence on the internet, you let it be defined by other people. And then the nice thing is that language models make it easier to find some of that stuff, so that you end up, you end up showing up in conversations as the answer
[00:41:01] B: to You surpass your name twins.
[00:41:05] A: Yeah, yeah. Now, now if you go to a language model and you say, like, who's Parth Patel, like, it will actually be able to delineate between the different Parts Patels, and there's, like, nine of us at least on the planet, nine of us that at least work in tech.
[00:41:17] A: Heck, there's probably, like, a hundred thousand of us, but, like, most of them aren't on the internet. And it's like, okay, so then I was like, started taking speaking gigs, because I was like, oh, turns out that also helps, right? Like, coalescing who you are. And then I went back into Corsair, I logged in, and I updated my bio so it's more consistent with my presence on the rest of the internet, so that the language model seeing
[00:41:37] A: 500 links would be like okay this is Parth and this is a bunch of other people also named Parth Mm but
[00:41:42] B: hmm.
[00:41:42] A: that this is the Parth that's known for working with language models at the bleeding edge of coding agents that like I want to be known for that so then I put that on the internet internet is like evidence of how that how I should be known for that yeah
[00:41:55] B: Yeah.
[00:41:55] A: 
[00:41:55] B: Yeah.
[00:41:56] B: So you do the course,
[00:41:58] B: you're starting to get consulting gigs.
[00:41:59] A: Yeah.
[00:42:00] B: And at this point, you're probably feeling a little less imposter syndrome surrounding this skill set.
[00:42:05] A: Yeah.
[00:42:06] A: It was cool because that was two months after my parents were really concerned about me.
[00:42:09] A: And I was like, I think I can, I think I'm on the verge of this.
[00:42:12] A: And then, then my fate kind of, well, basically I went cashflow positive and the cashflow
[00:42:19] B: That's
[00:42:20] A: started
[00:42:20] B: big,
[00:42:20] A: to increase.
[00:42:20] B: baby.
[00:42:20] A: Yeah.
[00:42:21] A: Yeah. Wet food.
[00:42:21] A: Wet food.
[00:42:22] A: You know, the boys ain't going to starve.
[00:42:26] A: And then, um, a mutual friend out of the, I was working at this, I was just co-working at a hacker house with a bunch of AI LA people, and one of the guys at the hacker house, Ben Relis, um, he, he, he was like, oh, like, do you know how to build these, these, I was really good at building chat bots, and he was like, you think you could build a chat bot on top of, like, a corpus of an established creator? And I was like, yeah. So I built the first version of that that afternoon for, uh, Reed Hoffman,
[00:42:53] A: and then they were playing with it and then they were like oh this is really like good and then he was like you should a couple weeks later he was like you should come to the bay area and then I come to the bay area he introduced me to Reid Hoffman
[00:43:03] B: Hmm.
[00:43:03] A: and we we talked for like five hours basically Reid's just picking my brain about like this journey the same conversation we're having right now I was like I was like what did you like what have you learned in the last like exploration of language models and it's like well here's how it applies to every domain I've ever worked in I spent one month talking to language models about music
[00:43:22] A: It turns out they're really good at operating a digital audio workstation. But then if you combine that with the fact that, you know, they're good at code, then you start wondering, well, then they're going to be really good at using the computer. Then actually I should be able to talk to my software. Like, then it's like, what is the role of voice? Like, all of these, these, like, trend lines that I've observed in exploring the models. And I'm talking to Reed Hoffman, right, co-founder of LinkedIn, one of the earliest investors in Open AI,
[00:43:47] A: and he's probably the first person I talked to that's like, like, we can have a real conversation about the second, third order effects, and
[00:43:55] B: Wow
[00:43:56] A: that, like, I am not insane, but that this is actually, like, the game has changed. And, um, that ends up being a four and a half hour conversation, and, um, next day, um, he reached out, and he's like, you should, you know, like, would you be down to work for me?
[00:44:13] A: And, uh, I said yes. And so I've been working with him for almost two years now, uh, kind of doing the same stuff, but now I have a team of people, and we, like, people that I build tools for, so I get more feedback, uh, than just working in isolation. And also, like, uh, Reed is one of the most creative people I've ever worked with, and so he's just got, like, a thousand ideas, and
[00:44:34] B: Mm
[00:44:34] A: that's
[00:44:34] B: -hmm.
[00:44:34] A: perfect when you have hundreds of agents that are coming online, and you're like, hey, like, we should attempt and start figuring out all these things. We're no longer bottlenecked
[00:44:42] A: on dev productivity if we can, like, delegate a lot to this, to these models that are coming online. And, like, uh, so, like, the projects are ranging. I think Read AI has now become its own, um, more sophisticated thing. Like, it's gone on speaking tours. It's probably going to be on more podcasts.
[00:44:57] B: Read AI is the
[00:44:58] A: Yeah.
[00:44:58] B: avatar.
[00:44:59] A: It started off as a chatbot, but now we have a visual avatar.
[00:45:03] A: We have like an interactive avatar you can talk to. It's not widely available yet,
[00:45:06] A: but it has been on like Colin and Samir's podcast.
[00:45:08] A: And I could totally imagine this thing being on, it's been on the news in different forms.
[00:45:13] A: And it's been an interesting project because it blends language models and video models and avatar technology,
[00:45:22] A: retrieval systems into this like avatar experience. So I'm known now.
[00:45:26] A: Now for digital twin projects and like AI clones.
[00:45:31] A: As a side effect of being good at working with language models is
[00:45:33] B: Mm
[00:45:34] A: that
[00:45:34] B: -hmm.
[00:45:34] A: this problem space became, like, and it was just because, like, we're like, oh, what if we did this? Like, and I was talking to readers, like, what if we connect this to, like, an avatar? We could create that scene from Tron where Clue, like, where Jeff Bridges talks to himself. And I was like, that would be amazing. And then we, like, get access to hologram technology, and, and Van Nuys, and then they're like, oh, we can put this avatar in a life-size box, and you can show up and talk to it. And so they, it just keeps
[00:45:58] A: getting, that project has been a magical surprise, because it's just like
[00:46:02] A: it's like, oh, we would love for Read AI to keynote speak at the Computer History Museum. I'm like, are you kidding me? Like, that, I went there as a kid. It's one of my favorite field trips. And then I get to come back, and, like, a thing that I made is now giving a keynote presentation. And so it's, you could never, it's, that's one of those weird things where, like, you don't know where it's going to go, but, like, you're rewarded for playing with technology. And also, because what I learned from Read was, like, I was, because I was mind blown at how, like,
[00:46:29] A: like, it opens doors, but it's, what I learned from it was, like, well, if you're early to something and you experiment publicly with the technology, then what happens is, like, everyone else knows you as that guy, and then eventually it'll go mainstream, and it'll be, they'll take it for granted one day, but they will remember you as the person that introduced it to them. And so we did that with avatar technology, but it also was like, I did that with GPT for my network. I was the guy, I was like, and I used to, one of the mantras when I was unemployed was like,
[00:46:56] A: I could be, like, the Johnny Apple Seed of GPT4. I just go around to just, like, sprinkle GPT4 across everyone I meet. It's like, you know, you, you know, you meet the cashier, it's like, you should download Chat GPT, and he's just like, what? Trust. But it's like, you know, you never know. It's like, you get your Uber driver, it's like, have you, have you heard of the Lord and Savior, Chat GPT? And then next thing you know, they're like, oh my God, like, this changes. So when you actually, so then, but then, like, my, my friend's like, oh, you should be using Cursor
[00:47:23] A: before anyone else. I'm using Cursor. Now everyone's using Cursor, right? So before anyone else, like, I recognize the technology is really good, then I planted across my network, partly because I'm like, it starts off small. It's like, okay,
[00:47:35] A: did everyone hate that, or did it change the way they work every day? Because if it, if, if they hated it, then I'm like, okay, why do they hate about it, and then, like, move on to the next thing. But then it's like, you, some things you show people and everyone agrees this is a game changer, and they're using it in a different way. Like, voice transcription technology, same thing. Coding agent, same thing. Cloud Code, same thing. It's like, this technology sells itself. And so I'm like the Johnny
[00:47:59] A: Apple Seed of that. I was like, I'm just going to go and plant orchards by dropping apples all over the country, and then a couple years later it's just like, ah, yeah, now everyone's on it, thanks to, like, a couple people just, like, being out there and creating content and then just, like, putting it out there. Yeah.
[00:48:13] B: How was, at the end of that four and a half hour conversation, your brain? What were, like, the chemicals swirling in your brain, like, after doing so many months of
[00:48:25] B: sprinting on this problem space, probably?
[00:48:29] B: You had to live by these mantras to get you through a
[00:48:32] A: Yeah, yeah.
[00:48:33] B: lot of this period and doubt started to creep in.
[00:48:36] B: And then there's this extreme moment seemingly of validation and maybe there was several moments of payoff in the months leading up to this. But this really does feel like a pretty damn cool moment that just kind of validated all the months before. I'm curious.
[00:48:50] A: Yeah, it was definitely the final, like, like, it was like, okay, cool.
[00:48:55] B: Yeah, it's
[00:48:56] A: Maybe there aren't that the many
[00:48:57] B: final scene of Pursuit of Happiness.
[00:48:58] A: Yeah, it's like, yeah, yeah, yeah. Well, it's like, I mean, look, all my smart friends understood. They're like, he's not insane, he's just, like, early. And then I, and then you meet Reed. It's like, okay, no, this, this is, like, actually, like, most people have not even recognized that the game has changed,
[00:49:15] A: and they're still playing yesterday's game.
[00:49:17] B: Hmm
[00:49:17] A: And, um, but hearing that from him and, um, getting to bounce these ideas off of him for over four hours, I was like, we all have now more questions than answers, and that, like, uh, like, we need to find more people that are also just, like,
[00:49:35] A: um, in this wild west, right?
[00:49:37] B: Yeah.
[00:49:37] A: There aren't many of us. It's a subculture, and there will be, it will eventually go mainstream, but, like, right now we kind of need to just assemble, like, assemble the, the Avengers and just kind of, like, bounce the ideas off of the people that are, like, really pushing on them, and then learn from each other and have these conversations that otherwise you just, I mean, for me, now I've created enough of a network that I can,
[00:50:02] A: I can, like, call on my friends, even in LA, and have a conversation about this tech. But back then it was, like, isolating, right? You're the only person that knows or wants to talk about it, or you go to SF and every coffee shop's talking about it. But
[00:50:14] B: Mm-hmm.
[00:50:14] A: if you're not going to be in SF, then you better have, like, the internet. You better have communities that you create where you can explore it. Otherwise it's going to be very isolating. But meeting Reed, and the end of that was like, cool, like,
[00:50:27] A: if I could go more all in, I would, like, you know, I mean, of course now I still need to figure out money. But then the next day he offered me a job, and I was, okay. And then when, when we first had our first meeting, it was very much like, you can do the same thing, except now it's, like, not, it's
[00:50:41] B: not Have
[00:50:41] A: Yeah.
[00:50:41] B: you on the payroll.
[00:50:42] A: We'll pay you. Yeah, you got health insurance. Um, but, like, we're gonna do it even bigger, right? Like, uh, we're gonna go further. We're gonna learn more. It's no longer limit, it's not like, it's not the experiments that I, I, and also he started introducing me to people
[00:50:56] A: like within Microsoft that we're also thinking about because I was like okay who's working
[00:50:59] A: on this that's not constrained? Some people at Microsoft, you
[00:51:03] B: Yeah.
[00:51:03] A: know, you meet, you meet the guy who builds Google Docs, um, Sam Scolace. We have a conversation. He's like, yeah, I got, I got a couple DnD agents talking to each other, and it's, and he's like, that's, like, a failure, because, like, um, last I checked, it's just playing chess with some NPC. I'm like, isn't that victory? A lot of times you're playing DnD, and, like, people go on this side quest that you didn't expect, and then that is actually the magical thing, is that's the unscripted, like, side quest
[00:51:28] A: that they spend up. They're like, oh, we're just going to go to the tavern and buy drinks for everyone. That's going to be what we do today. And then, like, you may have planned something, but it's the thing that you didn't expect that ends up being the thing that everyone loves. Yeah. So I, I think of that a lot. I was like, okay, finally started meeting people that, because I was doing DnD bots, I was like,
[00:51:47] A: can you just have a DM, and can you have players, and then can you have, can you create an adventure in this, like, improv manner, when you create a couple different chatbots and you tell them to play DnD with each other and group chats and stuff? Like, this is the beginning of multi-agent stuff. And then, um, it's all relevant, because now that all of the models are better, they can use tools. So I'm reassessing those same ideas, but now giving them, it's like, what happens, you do a group chat of coding agents?
[00:52:11] A: Okay, well, now how do you make it so that when they work together today, it's like, it's not solved, but, like, because I've been working with these things even when they were imperfect, it's just like,
[00:52:22] A: cool, one day it'll work. It gets better. And also, like, maybe this is the wrong way anyways, but it's a lot of fun, and it's not useless. It's
[00:52:29] C: Mm-hmm.
[00:52:30] A: like, it just reveals something about what's possible, new form factors. And, um, I don't think it's, I don't think you're really wasting time. I think you're just, like, banking experience of, like, what worked, what didn't work, what might work. And maybe it ends up all being a waste of time when GPT7 comes out.
[00:52:48] A: I doubt it, though.
[00:52:49] A: And also until GPT-7 comes out, like these things are inklings of like useful paradigms to lean into.
[00:52:57] D: When we parse what knowledge means in this space,
[00:53:00] D: I think that that is actually like exactly what you're talking about right
[00:53:02] E: Yeah.
[00:53:02] D: now with like once GPT-7 comes out,
[00:53:04] E: Yeah.
[00:53:05] D: are the skills that I've built useful?
[00:53:07] D: And I think that that is for me a real curiosity surrounding hyper engineering or vibe coding or whatever the future of software engineering is, is is it like something that the intelligence threshold and knowledge threshold and skill threshold continues to basically?
[00:53:23] D: diminish and me who has not been playing in the same playground as you and just tried to basically vibe code with GPT-7 can start building similar applications and similar cool things to very skilled parth. I'm curious what your perspective on that is.
[00:53:39] A: Yeah, I have suspicion.
[00:53:43] A: Okay,
[00:53:44] A: so when I think about like, okay, a lot of times people say prompt engineering.
[00:53:49] A: doesn't matter, or, like, the models will just prompt themselves, which is true, the models will prompt themselves. But I don't think that prompt engineering doesn't matter. I think that, like, people have not really
[00:53:59] A: find prompt engineering as a, um, what I did a little over a year and a half ago was, I read, there was a paper that tried to create, use LLMs to figure out all the different techniques of prompt engineering. Just have them read a bunch of papers on prompting, and then it's like, do a taxonomy analysis. Like, how many techniques are there? And at the time there were, like, at least, I think there were, like, 80 something techniques back then,
[00:54:20] A: and I was like, wow, if there are 80, like, some of them are, like, few shot prompting, provide examples of how to solve the problem, or, like, allow it to use tools, or, like, image prompting. These are all different approaches to prompt engineering. So then I was like, okay, there's at least 80. You can also mix and match them, which means that, like, prompt engineering is a very, uh, multifaceted, uh, space of, like, bringing context into the model so that it has a better chance of, like, zero shotting your problem,
[00:54:48] B: Hmm.
[00:54:50] A: or, like, even just helping you with your problem.
[00:54:51] C: not And
[00:54:51] A: Even I
[00:54:52] C: zero shotting is so important because of the cost in both time and dollars,
[00:54:56] A: right? don't think it's that, it's more like, if I can get this thing to solve the problem in one word prompt, and it takes you, like, five turns of conversation, then maybe I'm, maybe I'm just better at this than you. And it's like, uh, I think that's, this is more like, okay, prompt engineering, I think, is sort of like the 10 000 hours.
[00:55:16] A: You gotta, you gotta run 10 000 prompts before you, like, judge a model. And I'm not saying that, like, like, you could, like, you get better by looking at other people's prompts. You can get better at using, you have to use the model to get better at using the model, knowing what you can ask for. But then there's this whole other vector, which I've been thinking about, which is, like, expertise, the role of expertise, and what levers that gives you. So if you're really good at something,
[00:55:39] A: You have a very rich vocabulary that comes from being good at that thing. If you're really good at photography,
[00:55:44] A: you know how to use your camera,
[00:55:46] A: you know all these levers of the different settings on your camera,
[00:55:49] A: or you know cinematography or you know direction,
[00:55:52] A: you know like anything.
[00:55:54] A: The better you get at it, usually there's an associated vocabulary that comes with that expertise. Like if you're a five-year-old working on...
[00:56:04] A: playing an instrument it's different from like Hans Zimmer working on like a movie soundtrack movie score and his exposure to so many like his experience his exposure to so many experts allows him to articulate his vision to these other experts which work on his behalf to create a more comprehensive ambitious thing than a five-year-old could ask for five-year-old can ask for something and I'm sure it will probably sound like a masterpiece but they're less in control because they don't have the control over the language
[00:56:31] C: Yep.
[00:56:31] A: And I think about the language is like your, um, I'm gonna get his name wrong, Fitkinstein, I'm
[00:56:39] C: Mm-hmm.
[00:56:39] A: gonna name drop. Did
[00:56:40] C: You nailed it.
[00:56:41] A: I, okay, so what did, what did Fitkinstein say? Reed's always name dropping him, and I'm always like, but then it
[00:56:48] C: You
[00:56:48] A: I
[00:56:48] C: do something regarding like the game.
[00:56:50] A: think he says, okay, he says, like, uh, yeah, look it up, look it up, pull that shit up, pull that shit up, pull that shit up. Um, I think he says, like, language is the limit of my, uh, cognition, or
[00:56:59] A: language is the limit of my ability to think pull it up find the actual quote though because i don't want to forget this or butcher it but i feel like i discovered that and then i read Wittgenstein because my buddy was a huge philosopher and mentioned it and then i was like oh this makes sense like i'm just there's nothing original it's just what's the quote what's the
[00:57:16] B: The
[00:57:16] A: quote. Yeah.
[00:57:17] B: limits of my language mean the limits of my world.
[00:57:19] A: Yeah.
[00:57:19] B: That's gorgeous.
[00:57:20] A: The limits of my language mean, and that's even better than what I was suggesting, because
[00:57:27] A: it applies to all AI, not just chatbots, but, like, every image model and video model and game model of the future. The limit of your ability to wield the technology, the limit, and their world engines, things that can literally generate worlds. So the limits of your world and what you can create are, it's, it's the language. It's
[00:57:44] B: Mm-hmm.
[00:57:44] A: like your ability to articulate that, and you could give that to a language model, but otherwise, in that scenario, you, you are still less, like, you have
[00:57:51] A: You are telling the language model to do some of the decisions that you wish you had more control over,
[00:57:56] A: but that would only come with experience and or expertise.
[00:57:59] B: Yeah, that makes so much sense. And I think once you have that output from the model.
[00:58:06] B: The expert will be able to have the taste to
[00:58:08] A: Yeah, yeah.
[00:58:09] B: say,
[00:58:09] A: 
[00:58:09] B: okay,
[00:58:10] B: where did it fuck up?
[00:58:11] A: Or, like, pull some more levers and get better iterations. The people are so obsessed with the zero shot, but I actually think that, like, the sixth or seventh version of something is, if it could do this, if it could go so far in one,
[00:58:23] A: then you see it and you play with it and you're like well actually what we need is slightly different Mm what we
[00:58:26] B: -hmm.
[00:58:26] A: want is different. Like, my vision is a little more ambitious or different, then the second and third rounds of iteration are also very powerful, of
[00:58:33] B: Mm-hmm.
[00:58:33] A: course you go to Replit it will Replit's like vibe coding app the first version will be impressive
[00:58:39] A: but it's like the eighth or ninth version, once it's, like, deployed with a couple people, that you actually are like, this is, like, the well-oiled machine. So
[00:58:45] B: Hmm.
[00:58:46] A: Iteration is key. Expertise is key. Even if the old job is gone, you having experience, it's a good, like, it's a good, like, bridge for you into the next game, is leaning on your domains of expertise. The other issue is that not everyone's an expert in anything. Like, many people are not an expert in anything.
[00:59:06] A: There's a positive there, which is, through, like, dumb luck and just, like, being a noob, you might discover something that experts would otherwise be too proud to discover. Like, I believe I'm one of those people where I'm like, I just asked the model, hey, is this possible? And it's like, yeah, it is. Okay, good thing I didn't, good thing I didn't think that this was my unique, like, like, that, like, I had to do this myself, that I might be
[00:59:30] A: uh, too proud to, like, ask for a possible answer from a thing that is trained on everything. Um,
[00:59:36] B: Mm-hmm.
[00:59:37] A: and I find that, like, that's been huge for me, like, humbling. It's just like, dude, you're not gonna learn everything. You're gonna get good at a couple things in life, but as also, like, you can just not be judged for asking dumb questions now, and then you might just stumble into, like, interesting solutions. Like, now when I do presentation, I used to hate doing presentations, because
[00:59:57] A: It's not that they're active,
[00:59:58] A: like doing.
[00:59:59] A: making a presentation is, like, annoying, but, but, like, that my mind would see a deck and I'd be like, this is off by a pixel. I'm like, why does it matter? Why do I care that it's off by a pixel? Like, like, isn't it the subject of the deck? Like, now I'm distracted. I'm like, why, three of the bullet points have a period, the fourth one doesn't, and I'm like, am I missing the whole point here, right? Like, by paying attention to the detail, and a lot of that being hand, manually done, it's like, man, we're spending so much time in the, the thing other than the idea,
[01:00:26] A: idea. Yeah. But then I went to a language model. I was like, you know, build a deck, and then it builds a presentation, and I'm looking at it, was like, wait, this is a web app that looks like a presentation. And it's like, yeah, what do you need? And I was like, give me a PDF. It just converts that into a PDF. I'm like, we're using web development to build presentations. And I was like, um, it's like, oh, it's, let's create a fake data set. Here's a fake data set for an ice cream company. You build a deck, it builds the deck,
[01:00:52] A: the deck is, like, pink, and it has, like, ice cream and stuff. I'm like, make it McKenzie, though. And then just rebuilds the deck in the style of McKenzie, and for all intents and purposes it's a presentation, but it's not built in PowerPoint, and it's not, it's built in HTML, CSS and JavaScript, and it's got, like, sleek animations and everything. But now I'm
[01:01:12] A: creating the entire thing by just prompting it. I can be giving the deck as a, I can be presenting the deck, and then someone asks a question, then I can switch over to the coding agent and be like, add a slide about X. By the time I get there, that slide is ready. And so now I'm like, well, I've discovered that you can vibe code decks. Why would I do, why would I make a deck manually ever again, if I could just talk to the model and it could generate the deck and it would look professional
[01:01:39] A: as I need it, and it's uniquely, and I can be like, oh, it's too verbose. Oh, yeah, it'll just rewrite every single slide. Make it less verbose. And it's like, oh, vary the slide layouts a little bit. Make it more, like, you know, you can't just have three bullet points on every slide. It adds more variation. So I'm like, wow, like, I feel like I'm the partner delegating
[01:01:56] B: Mhm.
[01:01:57] A: to the analyst, right?
[01:01:58] B: Yeah.
[01:01:58] A: except
[01:01:59] B: Yep.
[01:01:59] A: now the analyst is this bot that doesn't sleep, right? And the ramp time for that bot to understand the problem is, like, zero,
[01:02:08] A: which is interesting. But again, it's like, I've seen and gotten good at the old way, that was, that is now inefficient, and now I can ask for a higher quality output in, like, three prompts. So again, leaning on the expertise to get to a higher quality in your, in the thing that you know how to judge quality on.
[01:02:27] C: Yeah,
[01:02:28] C: it makes a ton of sense.
[01:02:30] C: Are you down to play a little game of
[01:02:33] C: But we'll call future headlines and we're going to try to we're going to try to test your predictive abilities for
[01:02:39] A: Sure.
[01:02:40] C: two to five and ten years out surrounding
[01:02:42] A: Okay.
[01:02:42] C: AI.
[01:02:43] C: Okay.
[01:02:43] A: I don't think anyone can see more than three years out, but let's play.
[01:02:46] C: You can though,
[01:02:47] C: you can.
[01:02:48] A: Maybe.
[01:02:48] C: Maybe. Okay.
[01:02:49] C: But I'm going to read you some headlines for possible things that we could be seeing in the AI world.
[01:02:53] C: Some are much more positive,
[01:02:55] C: some are a bit more negative.
[01:02:56] C: But yeah,
[01:02:57] C: yeah, of course you need those.
[01:02:59] A: And I just want to get your reactions. And if you have headlines that you think are much more likely than any of these,
[01:03:05] B: Mm-hmm.
[01:03:05] A: Please tell me.
[01:03:06] B: Sure.
[01:03:06] A: But for two years out,
[01:03:08] A: A,
[01:03:10] A: a former CTO licenses his AI clone to his former company,
[01:03:14] A: trained on years of notes and meetings to advise company executives and engineers for royalties.
[01:03:22] A: B.
[01:03:22] A: AI agents are seen as being overhyped and continue to require a human in the loop for even many low-level tasks.
[01:03:30] A: C.
[01:03:31] A: Goldman,
[01:03:32] A: McKenzie,
[01:03:33] A: and Meta cut 25% of their analysts in research teams after internal AI agents significantly outperform those hires.
[01:03:40] A: Middle managers are next.
[01:03:42] C: Yeah.
[01:03:43] D: Okay,
[01:03:44] D: let's start with the CTO cloning himself.
[01:03:45] A: Let's do it.
[01:03:46] D: I think that digital twins that people I think because right now a lot of the digital twins are being attempted to being deployed within companies,
[01:03:53] D: that seems like what the headline might be. But I think even if you did that, that would be more of a stunt than a practical thing.
[01:04:01] D: I think that, yes,
[01:04:02] D: like digital twins of us working with each other while we sleep makes sense.
[01:04:06] D: I think a lot of context that we have like doesn't enter the company in a way that other people can use it. Like it would be nice if I could loan my expertise.
[01:04:15] D: while I'm asleep, to people that I want using my, a version of me. And I'm working on digital twin of myself. Like, can you talk to this thing and use and reach into some of my experience through this thing? And, um, so I think there is a practical version of this. I just think that, like, the more interesting thing that I think is, like, you've seen Mulan? Oh
[01:04:35] A: I've not actually so
[01:04:36] D: My God, too many people. Okay.
[01:04:38] A: uncultured
[01:04:39] D: dude, okay, gotta watch Mulan, but
[01:04:41] A: For let's assume that the audience hasn't watched Mulan either.
[01:04:44] A: Give us give
[01:04:44] D: Are
[01:04:44] A: us a little
[01:04:44] D: you serious? Okay, okay, okay.
[01:04:46] D: Assume that the audience has not seen Mulan.
[01:04:49] D: Okay,
[01:04:49] D: in Mulan, Mulan is a story of a girl in China and then they're being invaded.
[01:04:56] D: And her dad gets drafted. And so this is very beginning, I'm not going to spoil anything. Her dad gets drafted, but in her, in his place, she takes her dad's place in the army, and then disguises herself as a man in order to enlist, so that her dad doesn't have to go to war, because she doesn't want him to die in battle. And so she goes, but then what happens is her, they have, like, a shrine of her ancestors, and the ancestor spirits come out of the shrine, and then they meet, and they're like, oh my God,
[01:05:24] D: we need to send someone to protect Mulan. So they send a spirit down in, uh, and he takes, his name is Mushu, he takes the form of a tiny dragon. And then it's like,
[01:05:34] D: Mushu isn't even, like, the spirit they wanted to send. They're like, you need to go wake up the spirit to save Mulan. But then he kind of messes up, and then he just, he's like, oh, I gotta go solve this problem myself. So then he goes himself to accompany Mulan on her journey and protects her. I think the more practical, beautiful version of
[01:05:50] D: Digital Twins is like, I would like to be able to lend my perspective on life to my great,
[01:05:58] D: great, great grandkids.
[01:05:59] A: Hmm.
[01:05:59] A: Right? Like, what if you could draw from the wisdom of your ancestors? So I think that the ancestor version of, we're going to get every version of this, good and bad. Digital twins will be every version of this. I think in many of these cases, like what you're talking about, that there, you're going to get every version of this, good and bad. I think one of the more interesting positive versions of this is, like, I wish I could talk to my ancestors, or a version of them, or an agent that, like, embodies that, like, their
[01:06:26] A: publish body of work that I can lean on. And it's not them. It's like, more like a painting in Harry Potter that's, like, animated, that you can talk to. And we're going to get that. I think that we're also going to get these corporate, like, digital twins. I just think that the CTO one is, like, whatever. Although the, like, if that person is someone that, like, really, the, like, like, the Steve Jobs, like, you're going to want to be able to talk to the clone of Steve Jobs, such that Apple, like, I think there's some people will want
[01:06:56] A: Apple to be able to draw on Steve Jobs'
[01:06:58] A: unique perspective or some version of that to keep the ethos maybe somewhat in that world.
[01:07:05] A: But the corporate version is kind of boring.
[01:07:08] A: The ancestor version is more interesting.
[01:07:10] B: It's much more beautiful corporate version though.
[01:07:12] B: I mean, I think
[01:07:13] B: I hear in some of my work about companies where somebody is going to retire and
[01:07:20] A: Yeah.
[01:07:20] B: they're like, we'll pay you, we'll keep you on the payroll.
[01:07:23] B: You don't have to work close to the hours that you used to do,
[01:07:27] B: just you have to be ready to hop on a phone call when we need you.
[01:07:29] A: Yeah.
[01:07:30] B: And
[01:07:30] A: Yeah.
[01:07:30] B: That seems like a really weird
[01:07:32] A: So
[01:07:32] B: real application.
[01:07:32] A: That's definitely expert networks and stuff like that is a good application of digital twins.
[01:07:37] A: I do think that this is very likely that
[01:07:40] A: because people keep trying to hire me, but I don't have a job, so I'm like, I say no. And then people, like, well, you cloned Reed, can you clone yourself? Because I would pay to talk to it. I'm like, that immediately, it's like the question of, like, is this a new market? What's the size of this market? Expert networks, me loaning my unique perspective on life to people and then charging for it. Can this clone pay the rent? I have not figured the answer out yet, but I think it's yes. Like, and you can scale your expertise through this kind of, like, experience. So
[01:08:09] A: Yes, I think it'll happen.
[01:08:10] B: Fascinating.
[01:08:11] A: Yeah.
[01:08:12] B: And let's double click on the leading companies across tech and finance.
[01:08:19] B: What do you feel like we're going to see in the labor force two years from now?
[01:08:22] B: Do you feel like there are going to be a lot of lower level analyst type roles that are being removed or are they just going to be higher agency roles because of the little management that they get to do of the AIs below them?
[01:08:34] B: How do you see this impacting those big ass companies?
[01:08:39] A: This is my opinion,
[01:08:40] A: right?
[01:08:42] A: So I think that we're in the short run,
[01:08:46] A: a lot of companies will opt to lay people off instead of retraining them.
[01:08:53] B: Yeah
[01:08:53] A: I have not yet seen any major company outside of like a Frontier Lab type company.
[01:08:59] A: actually embrace AI in a way that is, like, AI transformation projects. Like, I haven't seen any large corporation make that. They say it, they're not doing it. I mean, like, they're trying, but they're not trying hard enough, in my opinion, mostly because, like, you just can't moonlight. Like, the amount of time and energy I put into just changing my own perspective is not something you can expect of everyone. And, uh, you,
[01:09:26] A: it's easy to expect it out of all the frontier labs, because they, for them it's like religion. It's like, oh, we're going to do this. We're going to build all our software using our coding agent. We're going to build, we're going to have Gemini go and scan and, like, make everything more efficient. Like, that makes sense, because they're native. They're AI native. I think the, the thing is, like,
[01:09:43] A: making an entire large company AI native is not, I have not seen anyone do it, right? Um, but they will talk about it, because that will signal to their board that they're interested. They're like, oh, we, you have to have an AI strategy. Okay, well, we have an AI strategy. Is it working? In 90, in most cases, like, the pilot programs are failing. It doesn't surprise me, because they're giving it to someone as, like, a side project, um, rather than making it, like, the unique focus of, like, a team that's actually
[01:10:10] A: actually free to move with high agency within a large company. They need to, they need to do that. This is why the startups is reaping most of the benefits, because they're basically starting off AI native.
[01:10:19] A: They imagine themselves as, like, okay, we're going to launch internationally faster than ever before, because now we have models that can help us internationalize our content, our business. We're going to monetize earlier than we need, we, we would have had to before. We're going to raise less, or, like, our rate, fundraising strategy changes, because all of a sudden, like, the cost to create software has, has, has fallen, right? So the new play style is unlocked, but uncovered by startups rather than large companies.
[01:10:46] A: And then the large companies,
[01:10:47] A: I think the other thing is there is a huge middle management class that I think like you'll see deeper layoffs within to improve like the like cost structure
[01:10:59] B: the
[01:10:59] A: of larger
[01:10:59] B: bottom line
[01:11:00] A: companies at
[01:11:00] B: Yeah, yeah.
[01:11:01] A: the bottom line.
[01:11:01] B: I mean I think to this I really agree with the short term and I want to hear what you think in the longer term but I think the short term is so fascinating just because though agents and AI a few years ago you were very early on it now
[01:11:14] B: The stock market would disagree with you on being super early on it. Like right now, it is it is everywhere and everyone believes in it as like the future.
[01:11:21] A: Yeah, yeah.
[01:11:21] B: And I don't think that we're going to see the revenue.
[01:11:26] B: in any of these major companies reaped via AI in the next year or two in the way that would justify the valuations that these companies are now having, or the
[01:11:34] A: When you say these companies, you mean like the hyperscalers?
[01:11:37] B: Hyperscalers, yeah, yeah.
[01:11:37] A: Yeah, there's definitely like a huge bubble narrative right now.
[01:11:40] B: Yeah, the price to earnings ratio
[01:11:41] A: Yeah,
[01:11:41] B: is
[01:11:41] A: yeah.
[01:11:41] B: is not, because of mere revenue, is not going to change, I think, within the next year or two because of AI productivity gains. So I do think cost cuts seem like
[01:11:52] B: likely the way that AI is going to enable
[01:11:54] A: Yeah.
[01:11:55] B: these companies to hit some of their expectations in the immediate couple years.
[01:11:59] A: The thing is, like, we're in the very first innings of agents working. This
[01:12:02] B: Mm-hmm.
[01:12:03] A: is, like, the first year I've seen a coding agent actually, like, coding agents that can work for, like, eight hours in a row, um, that, that you feel like they're not, and then they're now starting to get more parallelized on the cloud, that you can spin up more and more of these, like, an ephemeral army of extra coding agents to work on something. And coding agent is a misnomer. It's just that coding agents are the most powerful versions of AI agents right now.
[01:12:25] A: And they're just less intuitive to the non-technical folk,
[01:12:29] A: but they are the ones that have the most abilities because code is, it's not just for building software, it's that we use software as a way to solve many other problems and that you can use coding agents to scale cognition in cyberspace.
[01:12:44] A: That being said,
[01:12:45] A: if you think about where the job cuts, it seems like the jobs that are at risk are a lot of the jobs that were
[01:12:52] A: uh wrote cognitive jobs like repeatable tasks that are cognitive in nature that you could drop an LLM in and potentially get better output than a human and
[01:13:00] C: Mm-hmm.
[01:13:01] A: then those are risky and also anything that can be done behind a computer.
[01:13:07] A: which is a lot. Like, like, there's a whole class, like, there's a whole, like, a lot of us work behind a computer, and if you use agents that can browse, use a web browser, that's just the first thing. But, like, you use coding agents that operate on, on the operating system level, like, they have the entire computer at their disposal to solve a problem, and that you can run many of them in parallel. It just is a very
[01:13:33] A: It's a new form of labor that is competing with the existing form of labor.
[01:13:38] A: And I do think that orchestrating that is a way out, right?
[01:13:44] A: Like if I think about data analysis in the old world doing stuff manually,
[01:13:50] A: writing scripts to automate,
[01:13:52] A: you know, business problems,
[01:13:54] A: but actually now you have language models writing those scripts.
[01:13:57] A: If I were to compete with the language model that's doing data analysis,
[01:14:00] A: it would be sort of like John Henry competing with the steam engine.
[01:14:04] A: I don't know if you know the folktale of John Henry.
[01:14:06] A: But it's like a classic American folk tale.
[01:14:08] D: It
[01:14:08] A: Man, bro.
[01:14:08] D: feels so illiterate.
[01:14:09] A: You gotta, like, we look for the inspiration, right? Like, John Henry, uh, I'll just go very quickly through it. John Henry, American folk hero of legend, considered, like, one of the great, like, the great steel men of our time. Yeah, it's not, I think he was a real person, but if not, he's a, he's a legend. So he was basically the greatest, the strongest steel man, and, and, like, and they were, they were trying to build a railroad through the Rocky Mountains, and at the same time, so he's like the most,
[01:14:34] A: he's like the strongest of the steel men. And then there's a steam engine, and so they do a race: who can break through the other side of the tunnel to the other side of the mountain? It's John Henry manually versus the, a crew of people manning this, this train steam engine drilling, you know, like. And then periodically the train would break down, and then the crew would have to patch it. And then eventually, like, John Henry breaks through the mountain first, wins the race, and then the train, the steam engine kind of, like, sputters at the end,
[01:14:59] E: Mm-hmm.
[01:14:59] A: but kind of, like, get second place. And then he dies of a heart attack as a result of, like, overexhaustion. And this is a classic tale of man versus machine, and it's like, would you rather be John Henry, this, like, the great data analysis of, like, the manual era, manual, but, like, or would you rather be designing the system that does the analysis? And it's, I, I saw, when I saw language models doing data analysis that I could talk to, I was like, man, we're the first inning of this. Like,
[01:15:26] A: I would not hire a data analyst to do this by hand. And I'm of the opinion that Excel itself is, like, like, I mean, I'm like, the Excel is not Excel. Like, I'm very good at Excel, but you won't catch me working in Excel ever again,
[01:15:41] B: Hmm.
[01:15:41] A: because I think it's just like we're in the weeds when actually we should be orchestrating the model that builds and works with the Excel model. And it may not be an Excel model. It may be actually code.
[01:15:50] B: It should be, it should be, yeah, just human language.
[01:15:53] A: Yeah, we, yeah, exactly. Like, language is our method of orchestrating
[01:15:58] A: these systems, and they can use code and all other tools in the computer at their disposal to help us solve our problem. Yeah.
[01:16:05] B: That's one of those that is like really tough for me to grapple with the like lack of parallels in progress within AI that I feel like we're still experiencing, where,
[01:16:12] B: for example, I spent like 65 hours this week doing my job basically all in Excel and creating a model that I just know is going to be able to be fully built by AI within the next two to five years.
[01:16:26] B: blown away
[01:16:26] A: Yeah.
[01:16:27] B: if not and yet
[01:16:28] A: All the best
[01:16:29] B: on
[01:16:29] A: analysts
[01:16:29] B: like are trained
[01:16:29] A: getting hired by the Frontier Labs to train the model to do that specific set of work better.
[01:16:34] B: bankers are getting paid like 200 hours an hour to train these models yeah and then today on my drive over to do this episode I interviewed essentially GP speaking GPT and had it play you and practice this little interview and that felt like such a more complex task
[01:16:52] B: than just understanding this cells and an Excel sheet and how they should filter and plug into one another.
[01:16:58] A: Yes.
[01:16:58] B: And, yeah, so it's one of those really fascinating lack of parallel in progress, and I think we're going to see probably a little bit of a leveling out in
[01:17:05] A: Yeah.
[01:17:05] B: the coming years there, but it's like we clearly have the innovation, and it just hasn't cascaded through.
[01:17:10] A: Yeah. I think I talked to my dad about this and it's like because I'm kind of frustrated with the lack of speed of maybe adoption.
[01:17:21] A: And then my dad was like, no, no,
[01:17:22] A: actually,
[01:17:22] A: like the first phase of this, if it's anything like the Moore's law and the computing revolution is the first phase is the infrastructure build out.
[01:17:31] A: And then that's why a lot like more people are making money on consulting on AI than they are even in like making apps and stuff like the startups haven't yet come online.
[01:17:40] A: But then the second once the infrastructure and the compute and all that is and the capacity is there and we have like the first set of like really working good agents,
[01:17:48] A: which I think we're in the very beginning of like cloud code and codex are amazing,
[01:17:51] A: but they're going to get so much better. And then they're going to evolve into better agentic products.
[01:17:58] A: And then the second wave is the.
[01:17:59] A: application layer, and we haven't yet had the second wave. The second wave, which is, like, you think about the Airbnb, the Ubers, the things built on top of the infrastructure. But no one looks at Uber and thinks, oh, what do they run on, AWS versus, like, whatever? Like, that stuff is abstracted away from the end user. And the application layer wave has, we're not even in the first, like, that hasn't even begun yet, I think. Um, the question is how quickly does all this happen. Um, I do think that
[01:18:27] A: like AI agents, I thought, I was like, when I got into, I was like, shit, I'm gonna be working on this for 10 years.
[01:18:32] A: Um, and now I'm like, I guess, like, year four of this, and there are people in AI that have been working on this for 10 years before me. And I'm like, yeah, 10 years. I mean, it's gonna be very useful up until that point, but even, like, like, it's, I could use this now. You could freeze it, and I'm still gonna have superpowers. The superpowers is just gonna get better. And also, like,
[01:18:53] A: I can think of more ways to apply them,
[01:18:54] A: but that only have like, I got, I got, I have a cloud code that just does my expense reporting once a month.
[01:18:59] B: Hmm.
[01:19:00] A: And I think that should be a product.
[01:19:03] A: I'm tired of anyone. No one should have to do it by hand.
[01:19:06] A: Cloud code will very likely do my taxes next year.
[01:19:09] A: Like I don't want to do that.
[01:19:11] A: And I think it's something that I can't, if I'm doing that on my computer,
[01:19:14] A: why can't this thing just do my taxes for me?
[01:19:16] A: Like.
[01:19:17] A: Um.
[01:19:17] C: Right.
[01:19:17] A: And think about all the scenarios that I'm not thinking about and find the optimization. So, but that also required, like, I spent all my time on this, and I know I'm still not applying it. So, like, you can imagine someone that doesn't spend nearly as much time on this, and the vast majority of the economy is not thinking about this. So, like, it's going to end up taking the capacity, and the capabilities will come online, but, like, the creative, creativity to orchestrate the agents to solve those problems is its own learning curve of, like, prompt engineering and then designing systems.
[01:19:45] A: So it's going to be a journey. Like, it's going to be a longer, and everyone wants it to be immediate, and maybe the prices reflect, like, like, I, I don't really think about the public markets. That, like, definitely feels like a bubble. I see a lot of, I see a lot of money, money going towards things that are LARPing as AI, and I'm like, well, that's, that's, like, red flag stuff. But at the same time, I think about dot-com era, and it's like,
[01:20:09] A: even after dot-com bust, it's like, well, a lot of that, those things were just, like, early, and
[01:20:13] D: Mm-hmm.
[01:20:14] A: then we ended up getting those same businesses, like, you know, five years later, ten years later. You
[01:20:18] D: Completely.
[01:20:18] A: know, so timing. Um, but I think, like, for me it's like working with startups, working with people that actually are building. Like, there's stuff that is obvious to me, and I like to stick to the things that are obvious. Like, it is obvious that structured outputs, taking a language model and creating structure from unstructured
[01:20:36] A: blobs of data, is one of the most important things you can be doing. Um, using them to, like, create tools and creating automated workflows, very obviously, like, impactful today. But you don't need to, like, I don't need to convince, unlike crypto, I don't need you to believe in this thing working for it to work. Like, it will work for me at your expense, right?
[01:20:58] E: Yeah,
[01:20:59] E: yeah.
[01:20:59] A: I think there's something very interesting here of just the
[01:21:03] A: like cascading through the system
[01:21:04] B: Yeah.
[01:21:04] A: of the current level of innovation, because it basically feels like we've already reached the magnitude of another internet in terms of just, like, how much this technology should impact us for the next five years, just as it cascades through the system. But then the innovation is going to continue, which
[01:21:17] B: is yeah
[01:21:17] A: just going
[01:21:17] B: to and
[01:21:17] A: be so fast
[01:21:18] B: the learning curve, any humans are the bottleneck. It's like the learning, our only learning curve is the bottleneck.
[01:21:22] A: Yeah, yeah.
[01:21:22] B: Every day I'm sitting there, like, man, I'm still the bottleneck. I'm like, why am I still doing
[01:21:26] A: why am
[01:21:27] B: this
[01:21:27] A: I still in the loop
[01:21:27] B: Yeah, I'm, like, more in the loop now. I was like, but it's like, I thought I was supposed to go to,
[01:21:31] B: go to, I was supposed to be on the beach by now. Like, you know, what am I doing? So there is, like, a weird, uh, um, I, I didn't, I mean, I never, I don't, I don't know what I expected. All I knew was going to be, it's going to be crazy. It is continuing to be crazy. Um, and I think, to, like,
[01:21:46] B: the main thing is, you don't want to pay attention to the mainstream kind of narratives. That kind of stuff is just, like, a waste. It's better to use the technology, then
[01:21:53] A: Yeah.
[01:21:53] B: you know what's true, and talk to other people are using the technology, because, like, most people who are talking are not using it. You can tell because they have level zero perspectives, and it's like,
[01:22:02] B: if I can predict, if an LLM, if an LLM can emulate, simulate you and predict everything that you're going to say, then maybe you haven't gone that deep, right? Like, what is your actual unique perspective that the, like, that no one else is, like, seeing, because you actually went in and used the technology? And you might say it's not that good at this, but, like, you have to discover that for yourself, you know?
[01:22:24] A: Can I get your reaction to one more kind of prediction or it's kind of choosing what you think the outcome will be?
[01:22:29] A: of this technology in a decade.
[01:22:32] A: And this is going to reference back to some work you did pre-Clubhouse.
[01:22:36] B: Hmm.
[01:22:37] A: You supported the Andrew Yang.
[01:22:39] A: campaign?
[01:22:39] B: Oh,
[01:22:40] A: and
[01:22:40] B: yeah.
[01:22:40] A: Very, very high agency move. People were not on the Yang Gang as early as you.
[01:22:45] B: Did I didn't tell my parents I joined that campaign until after I took the job and moved to New York?
[01:22:50] A: that's so badass um but Andrew Yang like is still remembered for like one of his primary policies surrounding UBI and his prediction though was much more regarding the automation of blue collar jobs and the automation yeah automation of driving
[01:23:06] A: And
[01:23:06] B: Yeah,
[01:23:07] A: truck
[01:23:07] B: truck drivers,
[01:23:07] A: drivers,
[01:23:08] B: factory
[01:23:08] A: yeah.
[01:23:08] B: automation,
[01:23:09] B: the idea that, okay, actually, it's not that.
[01:23:12] B: Like, you go to a factory, it's not like immigrants are taking all the jobs in a factory, but that, because that's, like, a particular narrative that one side tends to harp on.
[01:23:20] A: What's that?
[01:23:21] B: Uh, but you look at the factories, like, filled with robots, and, like, the plan is to fill it with more robots. I talk to, like, robotics companies out here, and I see what they're doing. It's like, oh my God, the, like, factory of the future is a guy sitting at a desk orchestrating 25 robots, and then you send a guy to go fix one robot every once in a while,
[01:23:37] B: and they have robots of all sizes, like, mostly arms and stuff. I think the misnomer is that the robots will all be humanoids. It's like, that's kind of, like, a, not a red herring, it's just, like, that's, that's the flashy thing, because Terminator,
[01:23:48] A: Mm
[01:23:48] B: but
[01:23:48] A: -hmm.
[01:23:48] B: the practical thing is, like, arms on wheels and treads, because actually, like, a humanoid size robot is not strong enough to build a fire truck. You need, like, it's like the Tony Stark robots, you know, the arms
[01:23:59] A: Yeah
[01:23:59] B: that are, like, building the suit. Like, he doesn't have a bunch of robot humanoids doing the suit assembly. It's like a very different kind of the strength and the precision. So robots, uh, yeah, so factory automation. And he was basically making the case that factory automation, and, like, if you think about the most common jobs, he was like, if you just take all the most common jobs in the country and then how likely are they to get automated, and maybe the mistake we made was we were fixated on the
[01:24:25] B: the US, we, well, no, it's not that we're fixing, it's just that, like, we didn't expect Chat GPT and language models to come online as fast as they did. So we were thinking about the robots that we saw. But actually what's interesting is that ever since Chat GPT came out, so up until Chat GPT, like, we were the only crew saying the letters UBI, and so I would say we introduced it to the discourse. Since Chat GPT, I don't say these letters.
[01:24:51] B: Everyone else does. Like, other people, and they, they're like, oh, like, maybe we're going to need a UBI. I'm like, funny, you use those letters, dude. Like, they thought we were insane for even suggesting it. And I think there's other versions of this idea that are, like, more palatable to different sides of the political spectrum, where it's like, no, we're just going to invest in every single citizen, because we believe in you and your own agency, and that you know what's best for you. And that, like, instead of creating a patronizing system that says you, you deserve this particular benefit because we think that's what,
[01:25:19] B: it's like, no, we're going to allow you the agency to decide what you, your family needs, and to spend the money as you see it. And that, like, from your birth, you might, you might be allocated an investment by the government. There's different ways to brand these ideas, and then I'm not too
[01:25:34] C: That
[01:25:34] B: because
[01:25:34] C: was great branding by the way you should be VP of Yang 2028 please
[01:25:38] B: Yeah, maybe. But it's just like, yeah, that's a lot of what I do, is wordsmithing. Um, but it's just like, it's not zero sum. It's not me, right? It's not,
[01:25:48] B: heck, it's not my demise at your expense. This is, like, a mindset of abundance. But the question really is, like, um, we're like, uh, I mean, I don't know. It's like, on one side, I hope that things get cheaper, but everything is so expensive. So, like, you could say that, yeah, yeah, it's going to do all this stuff, but, like, if I don't see it, like, it's not, it's not real. It's like we're just talking about hypotheticals. Um, I do think that, like,
[01:26:14] B: We should have ideas and experiments at the ready for helping people ease the transition,
[01:26:21] B: because it is likely that things move faster than people's ability to adapt, and
[01:26:26] C: Yeah
[01:26:26] B: make it easier for people to transition to whatever. Like, it's just going to be rocky.
[01:26:33] B: And I don't think we should make it harder than it needs to be.
[01:26:39] C: Yeah, right now, your techno optimism usually fires me up.
[01:26:43] B: Yeah.
[01:26:44] C: I am curious, though, do you feel like when we see the GDP potentially double in
[01:26:49] B: Yeah.
[01:26:49] C: the next five to 15 years due to these technologies,
[01:26:53] C: do the rich just get richer and does late stage capitalism play out in greater?
[01:26:59] A: greater inequity? Does it lead to
[01:27:02] B: mm
[01:27:02] A: the
[01:27:02] B: -hmm this
[01:27:02] A: Yeah, you know,
[01:27:03] B: is
[01:27:03] A: where
[01:27:03] B: this
[01:27:04] A: I'm going
[01:27:04] B: is
[01:27:04] A: Yeah.
[01:27:04] B: A common, like, sentiment that people have, that, like, uh, I, I, okay, uh, I do have a feeling that, like, the, the gap kind of increases. But on the other hand, I think, like, would I rather have money or the ability to wield these technologies?
[01:27:27] B: And I would rather have the ability to wield these technologies.
[01:27:29] A: Hmm
[01:27:29] B: And I keep thinking, it's like, even if you have money, it doesn't guarantee you your ability to wield this stuff.
[01:27:36] B: And, like, you just can't turn money into wisdom. You will be able to turn it into compute and turn that into action, but
[01:27:41] A: Mm-hmm.
[01:27:42] B: taking the right actions. If you can make anything, if anyone can make anything, then it actually matters what you make, because now there's a surplus of, like, creative, no, production capabilities. And part of that is in the taste conversation. Part of that is into, like, just opportunity cost of your attention, your time.
[01:28:02] B: So, I think that there will be definitely like people in a couple companies will probably accumulate a lot of value more quickly.
[01:28:16] A: Mm-hmm.
[01:28:17] B: But also the other thing is like if the ability to wield this technology is potentially more important than having money right now.
[01:28:27] B: Then it might be, anyone has that advantage of, like, being able to wield the technology. Like, it could be someone in, on the other side of the planet that you don't have a common language with, it's just better at using this than you. And then the language barrier isn't as much of a big deal, because models and agents translate everything. So there are new play styles being unlocked that are just unclear. So I would not bet on any, like, typical argument, because, like, same thing, if you look at, like,
[01:28:56] B: before the internet, we didn't have creators in this new form. And now it's like, well, is the creator actually, like, the, like, is the creator more important today than ever before, in terms of what they, what kind of, like, unit economics they can, they can, they can move?
[01:29:15] B: Um, also, I kind of hope every, if, if we have a surplus of production, I hope everything gets cheaper. Yeah.
[01:29:20] A: Yeah, it's just like, along the way, probably the amount of solutions that will be created will, will scale as well. As, like, if we get to a point where that problem in a decade seems like a curious problem, it's like, there's probably a lot of things that were solved that we don't even realize could be solved now
[01:29:34] B: hundred
[01:29:34] A: because
[01:29:34] B: percent i mean
[01:29:35] A: of
[01:29:35] B: there's
[01:29:35] A: all the technological
[01:29:36] B: like a
[01:29:36] A: advancement
[01:29:36] B: lot of that is, like, the surplus of intelligence that just hasn't yet been applied to a problem space where previously maybe you couldn't,
[01:29:42] B: couldn't attract the talent to solve the problem. But now that's, there's no, like, you have an alternative labor force that doesn't have career aspirations, AI
[01:29:49] A: Mm
[01:29:49] B: agents
[01:29:49] A: -hmm.
[01:29:49] B: so now you can just turn dollars into cognition applied to any problem space, even if that wasn't a sexy problem space. And so that creates new opportunities for solving problems,
[01:29:59] A: things like fixing the grid, fixing, like, the operations of, like, a large multi-agent system, like the city, like the company, the country. A lot of that we can just make more efficient using AI, which is obvious to me, but, like, it's going to take longer, for sure. And then the, the, oh man, I have one last thing to say there. No.
[01:30:17] B: One more banger line.
[01:30:18] A: Man, I'm forgetting. Oh man, I lost the train of thought. Yeah.
[01:30:23] B: I think it comes back to you.
[01:30:24] A: 
[01:30:24] B: Feel free to pull it up.
[01:30:25] A: Yeah, cool.
[01:30:26] B: So I'd love to.
[01:30:28] C: go through a few philosophical questions
[01:30:29] B: Sure.
[01:30:30] C: and just kind of, yeah, more broad general questions.
[01:30:35] C: One is I think a lot of my ideas right now about the world,
[01:30:40] C: I see them as hypotheses because I feel like I'm young and dumb and I don't really know what I'm doing and I don't think I can call a lot of my ideas philosophies quite yet.
[01:30:48] C: You have tested a lot of your life hypotheses over the last kind of decade since UCLA.
[01:30:55] C: I'm curious what hypotheses in your 20s you feel like you would reach as philosophies now.
[01:31:01] C: Basically, ideas you had about the world that you feel like are very true.
[01:31:09] A: Okay. I mean, yeah, so in my early 20s, I, because I started interning when I was like 16, and I got, like, corporate finance internships, and early on in my career I was like, wow, a lot of the world, this is 20,
[01:31:23] A: it's 2011. It's like, wow, a lot of the world is just, like, people, like, corporate world is people moving paper from one filing cabinet to another. And
[01:31:31] C: Hmm.
[01:31:31] A: I was like, this is insane. I got video games that move information more efficiently than this company, Fortune 500 companies. I was like, man. Then I was like,
[01:31:38] A: clearly the workplace needs to be digitized. People made billions of dollars digitizing the workplace. And then it was like, after, the next wave after that was, like, so I was, at that point I was a finance analyst. Like, I was a business analyst, but then I was at a startup, and I was like, wow, like, many of the questions I have about the business are not reflected in the numbers of the finances. Like, what is our, like, I was working at Chow Now, an online ordering for startups, online ordering for
[01:32:05] A: restaurants as
[01:32:06] C: Mm-hmm.
[01:32:07] A: a startup, and I was like, wow, it's, like, not just about the dollars in and out. It's actually about, like, do people use the app? Do they like the app? Or what features are they using? Like, what's the most popular restaurant item, right? And, like, I was like, a lot of this is data that we have that we haven't turned into insights, and I'm limiting myself in problem solving if I only think about dollars in and out of the business and, like, how much we pay sales, sales reps. Like, the finance problems are just a small subset of, like, all analytical problems in a business. Up
[01:32:34] A: operational problems product problems so then I became a data analyst I was like now actually if I just get good at analyzing data everyone will give me their data within the business and then I will have a more complete picture of like the business not just how much money comes in and goes then you after that it was like the well obviously there was like the high-speed analytics analytics
[01:32:59] A: optimize databases. So I was basically, like, compounding on my, my, like, theory that, like, okay, if things get digitized, now there's more data. Now we're collecting more data, then we need to slice the data more faster. And then you have, like, Snowflake build billion dollar company off of, like, analytics optimized databases. Then language models, and I was like, oh, now we can, we have the steam engine of cognition. Like, the actual action of analyzing can be also delegated to the computer. And this comes from just working with computers
[01:33:28] A: all my life playing video games and having my parents who worked on semiconductors and just not betting against computers and thinking about the real world as like okay this is now a job that we need to put into the computer because this is no longer the best way for us to be spending our time and that has been a journey of like just betting on computers another thing is probably betting on networks like I just fundamentally think that life is multiplayer
[01:33:57] A: and it's multiplayer co-op. If you play video games, it's like, we are, yes, some of it is PvP, but actually, like, you bring, you pick allies, and then you attack life as, like, a team over the course of a time, like a very long time horizon. It's not that, it's like, it's actually even more important to build connections after an outside work, because you never know. Like, like, like, work is just one alliance, but there are so many alliances that actually transcend that
[01:34:24] A: at, over the 10-year time frame, people that I used to work for early in my career that now come and ask me for advice, right? But now it's like, eventually you become peers. Like, yeah, even your, even people that you beef with at the office, eventually you realize, wow, we actually were more similar than anything. And then, like, years later, I'm like, uh, so now I'm more like, okay, actually everyone I meet and work with is a potential ally in this, like, multiplayer game of life. And this was very helpful when I was unemployed, because I was like,
[01:34:50] A: I was like,
[01:34:51] A: I'm not alone.
[01:34:51] A: Actually, I have a bunch of friends that I can balance these ideas off of that will hear me and then help me understand if I'm insane or onto something.
[01:34:58] A: And that signal from the network allows me to make judgment calls that otherwise would not be possible.
[01:35:04] B: Mm
[01:35:04] A: And-hmm. planting the seed of the technology across the network allows you to get more signals from the network as to like,
[01:35:09] A: oh, this is a very good technology.
[01:35:11] A: This is how we should be using it. And then getting that back into my new thing is like, instead of other people texting me.
[01:35:18] A: I would rather them like be in a kind of group chat environment where they can ask each other and so now you the network is unlocked to each everyone within the network and so I think the cultivating networks that are that are like well curated and aligned towards some some goal of like increasing the collective agency of the group is very valuable especially when the world is changing so quickly.
[01:35:41] C: Right.
[01:35:41] A: That's beautiful. Yeah, no, that makes a ton of sense. Then, and the other thing is just play with tactic, play with the tools. Just play with it. Like, play with it. Don't just view this as, like, I gotta use this for work. Then, like, play with, understand it. Um, and, like, learn for yourself. Don't, like, there just isn't a textbook for some, like, a lot of this stuff.
[01:35:59] A: And waiting for someone to write the textbook is like definitely not right.
[01:36:03] A: So go chase the insight, find it, find,
[01:36:06] A: you know, turn the card over,
[01:36:07] A: find the insight for yourself and, and do that by meeting people that are also doing the same thing.
[01:36:12] B: Hmm.
[01:36:13] B: Hmm.
[01:36:14] B: Okay.
[01:36:14] A: Yeah.
[01:36:14] B: That's, that's gorgeous.
[01:36:16] B: I have two more questions.
[01:36:17] B: Cool.
[01:36:17] B: One is
[01:36:20] B: Or a friend actually who recently quit his job as an aerospace engineer in the desert.
[01:36:26] B: He's one of the smartest people I know.
[01:36:27] A: Okay.
[01:36:28] B: And he called me last week after a day in Replit and was,
[01:36:34] B: I've never heard this guy seem more high.
[01:36:36] A: Yeah.
[01:36:36] B: He was like, he drank four celsius.
[01:36:38] A: Yeah.
[01:36:38] B: It was sick.
[01:36:39] A: Dude,
[01:36:39] A: I drink two monsters a day.
[01:36:42] A: It's a problem,
[01:36:43] A: bro.
[01:36:44] A: Hey,
[01:36:44] A: let me get that monster sponsorship.
[01:36:46] A: First hacker sponsored.
[01:36:47] B: It's a muscle memory for, that's Monster good, that's Monster good. But he
[01:36:51] A: Yes, I've been, he's there. I've been there.
[01:36:52] B: Now he's, yeah, he's now entering this kind of weird period where he's like, yeah, I, you could go
[01:36:56] A: have this red pill moment of, like, oh my God, what can't I do? Yeah.
[01:36:59] B: Exactly. And he could go get a job, or, and, like, he knows that he has the capabilities. He was excellent at his last one. He left on his own accord, and he has another startup that's interested in hiring him. But right now he's like, do I move in with mom and dad? Do I just grind this technology? Do I create, I've a couple months of rent.
[01:37:16] B: How should I be thinking about this?
[01:37:17] A: Hmm.
[01:37:18] B: And I thought if you could give a little bit of advice to him,
[01:37:21] A: Yeah.
[01:37:21] B: he's wickedly smart.
[01:37:23] A: Yeah.
[01:37:23] A: Okay.
[01:37:25] A: Um, I, uh, okay, so, like, I'm not gonna give financial advice or anything like that, but I will say, like, um, this is, this is something that, if you can and you have buffer room, you should allocate your entire energy, attention to, while you're, while you have that opportunity. Most people will not have the opportunity to allocate their entire attention to it. And if you have buffer room, you should. If you're intrinsically motive, if you're, like, deep, he's like, oh shit, like, he
[01:37:52] A: he doesn't need to be told to do this he would otherwise do this like with his free time then like he's going to go further and you're going to go further in that thing that for you is very much fun like if it's fun for you and other people for it's for them it's work and they have to drag themselves out of bed they're never going to compete with you on this and so you will find you will pursue this to the extent of like your own energy and then you start drinking drinking Red Bull because you realize you the only thing stopping you is that you go to sleep at the end of the day the agents waiting for the next
[01:38:18] A: the next job, right? But I would say, like, Replet is great. It's a great, um, it's a great tool to start with. I would also suggest picking up Cloud Code and Codex from Open AI. They're complimentary to the Replet thing. You can use Replet to deploy and build the first version of your thing, but you can use Cloud Code and Codex to just go way further, because they're just very powerful agents. I think being good at using those tools
[01:38:42] A: is this new role called the AI engineer.
[01:38:45] A: And I think it's a very valuable role of the future.
[01:38:47] A: So if you're willing to aim for that next role,
[01:38:51] A: that next game of like orchestrating software agents to create things,
[01:38:55] A: I think it's not a waste of time.
[01:38:57] A: And it is a generally useful solution.
[01:38:59] A: Still go back to aerospace and apply this superpower to that problem space, and they may be happier and excited to work with you, because you spent your time uniquely devoted to cultivating this new superpower. So that's my opinion. Um, if it's, if it's your passion, and you already see the, it seems like he's already taken the pill. The question is, like, how much of, like, the current world do you optimize for versus, like, plan for the next one? And that's more of a personal thing. Um, but
[01:39:29] A: I don't think it's as, I don't think it's that risky.
[01:39:32] B: All that crazy.
[01:39:32] A: Yeah.
[01:39:32] B: Yeah.
[01:39:33] B: Love you, Brian.
[01:39:33] A: Yeah.
[01:39:34] B: And last question, something that I ask all my guests.
[01:39:37] A: Yeah.
[01:39:38] B: You can't answer time.
[01:39:40] A: Yeah.
[01:39:40] B: The question is, what is the ultimate scarce resource?
[01:39:45] A: Yeah, this is easy.
[01:39:47] B: Hit it.
[01:39:47] A: It's attention.
[01:39:50] A: It's attention.
[01:39:51] A: I think this is like,
[01:39:53] A: I haven't even figured,
[01:39:55] A: like I'm at 1% of understanding this.
[01:39:57] A: Because if I understood this, I would have been doing podcasts a long time ago. I would have been creating more media around what I've learned a long time ago.
[01:40:04] A: But that's just one aspect of attention.
[01:40:06] A: I think especially human attention.
[01:40:08] A: I mean, there's AI agents. They have their own new attention vector,
[01:40:12] A: which you can go for.
[01:40:13] A: But human attention is like the thing that doesn't increase.
[01:40:18] A: That I think about. Like, it's like, there's only 24 hours in a day. We sleep for a couple hours, and then everyone is trying to get in front of you. It might be some brand, it might be a social product, it might be, like, a creator. It might be, there's just, everyone's trying to get your attention. That's the thing that you just can't create more of. It's, like, finite. And then, um, it's like, you consuming, like, is you directing your own attention.
[01:40:41] A: In my case, like, I can apply my attention to a single problem for three months in a row. I think I'm very proud of my ability to just focus. Um, so being good at directing your own attention is key, because otherwise we're competing with a world that is trying to, like, draw us in. And, uh, so, like, you want to have agency over your own attention, especially with all these other things competing for it. And then also, like, understand that, like, it is good to consume too. Like,
[01:41:09] A: there are things you, that are valuable to consume. Like, allow your, your, like, I think that playing video games is actually very good for me, and that's not a mainstream idea, but, like, for me it's, like, a form of critical thinking in a sandbox, where it's like, I can learn risks without having to, like, there's no real world impact to, like, me going all in in a video game. But then I learn more about my personality. I'm like, okay, when it, when it, when it calls for it, I can, I can put everything on the line in a simulated environment. These, like, exercises of, like, simulation allow me to explore that. So I think even that
[01:41:37] A: form of attention, which a lot of people think it's a waste of time, is not actually a waste of time. It's a critical thinking and high-speed decision making and, like, strategic thinking. There's an outlet there. Um, I think that other things, like when you scroll social, part of understanding why someone is popular is, like, understanding how good they are at keeping people's attention, how they think about presenting information. I have so many friends here at the studio that are
[01:41:59] A: really good content creators, good at taking a complex topic and distilling it in short form video. And I'm like, man, I'm like, takes me, like, four hours to communicate something like that, but you've figured out how to do it in 10 seconds. That's very, so being able to wield, being able to create stuff that, like, efficiency, efficiently makes use of other people's attention is also very powerful. So, um,
[01:42:24] A: And I don't mean that in a good or bad way.
[01:42:26] A: I think that it's just like finite and
[01:42:28] B: Mm-hmm.
[01:42:29] A: worth getting good at thinking about.
[01:42:32] B: It makes a ton of sense.
[01:42:33] B: Speaking of which,
[01:42:34] B: where can people find you, now that you are trying to get a little bit more attention
[01:42:38] A: Yeah,
[01:42:38] B: on some
[01:42:38] A: yeah.
[01:42:39] B: of the incredible things you're doing?
[01:42:40] A: I'm getting better at this.
[01:42:42] A: AI agents have gotten my website to a pretty good place now.
[01:42:45] A: Made in Repl.it and polished by Codex and Cloud Code.
[01:42:49] A: So it's been a...
[01:42:50] A: So you can find me on a lot of platforms.
[01:42:52] A: I'm Parth Intelligence.
[01:42:54] A: And at Parth Intelligence on a lot of platforms. Go to my website, which is, you could be one of my only viewers on my website on any given day. I got, like, average of, usually it's like, oh, one user. It's like, oh, it's me. And then I'm like, oh, one user in Mexico is like, oh. But that's why I gotta, like, do more content, because, like, you wouldn't know who I am if I didn't show up somewhere on social. So, yeah, parth.club is my website. It's getting better. I'm putting more content. More of my mind is going there.
[01:43:21] A: Um,
[01:43:22] A: yeah,
[01:43:23] A: Instagram,
[01:43:23] A: Parth Intelligence,
[01:43:24] A: parth.club.
[01:43:25] A: I'll LinkedIn.
[01:43:26] B: Yeah,
[01:43:26] A: I'm
[01:43:27] B: I see
[01:43:27] A: pretty
[01:43:27] B: your LinkedIn post.
[01:43:28] A: active on LinkedIn.
[01:43:29] A: That's actually probably where my largest audience is.
[01:43:31] A: Parth last name fire emoji.
[01:43:36] A: That's which I can't say why you
[01:43:38] B: Very high agency.
[01:43:39] A: can say it's high agency. It's actually like
[01:43:42] A: it's like a trick for the language models. Yeah.
[01:43:44] B: Yeah.
[01:43:45] A: you
[01:43:46] B: So now when they cite the one
[01:43:49] A: know sometimes
[01:43:49] B: of your
[01:43:50] A: you
[01:43:50] B: twins,
[01:43:50] A: get an email sometimes
[01:43:50] B: they're like,
[01:43:50] A: you get an automated email. It's like, hi Parth, fire emoji. I'm like, all right, bro, do you even know who I am? Like, like, is this some LOLM? But now
[01:43:57] B: that's funny.
[01:43:58] A: that I've said this, like, it might, they'll be like, oh shit, I gotta, you
[01:44:01] B: Yeah,
[01:44:01] A: know
[01:44:02] B: the LLMs will figure out they'll read the transcript.
[01:44:04] A: i'll
[01:44:04] B: Yeah.
[01:44:04] A: have to change my game a little bit. Yeah, yeah.
[01:44:06] B: Well, Parth, thank you so much for taking the time. This was incredible.
[01:44:09] B: Well,
[01:44:09] A: Yeah,
[01:44:09] B: yeah,
[01:44:10] A: curiosity.
[01:44:10] B: curiosity is at an all-time high right now. I want to run through a wall and just explore these tools.
[01:44:15] B: So thank you so much, man. Really, really appreciate you.
[01:44:17] A: Thanks for having me. And I'm sure we'll do this again because there's so much that still I haven't even scratched the surface on.
[01:44:22] B: I love that.
[01:44:23] A: Yeah.
[01:44:23] B: Yeah.
[01:44:23] B: Thank you, Parker.
[01:44:24] A: Yeah.
[01:44:26] B: If you made it to this outro,
[01:44:27] B: thank you so much for listening.
[01:44:29] B: Genuinely trying my best to get better at this and make it as educational and entertaining for you.
[01:44:34] B: So please let me know how I can improve.
[01:44:36] B: I freaking love feedback.
[01:44:37] B: You can find me on LinkedIn,
[01:44:39] B: YouTube,
[01:44:39] B: Instagram,
[01:44:40] B: at Cole Hume.
[01:44:41] B: And I post frequently on my sub stack.
[01:44:43] B: So if you like to read a little,
[01:44:44] B: check out Young Smart and Battling Broke there.
[01:44:47] B: Until next time, smile at strangers and trust your curiosity.
[01:44:51] B: Thank you so much.