Launching ChatGPT with Zack Kass, Former Head GTM at OpenAI

(Intro Narrator at 00:00:00) Today, we're talking to Zach Cass, former head of go-to-market at OpenAI, about the story of launching ChatGPT and how society can responsibly adopt AI. Thank you to DigitalOcean for sponsoring this episode. For simple cloud and powerful AI that's built to scale, visit digitalocean.com or just click the link in the show notes. You're listening to Joel Beasley, Modern CTO.

(Joel Beasley at 00:00:31) You've been in AI for a long time, and you were working at OpenAI and go-to-market. And now you're doing a lot of speaking and consulting, helping companies make this transition. Is that a fair overview, or am I completely off base?

(Zach Cass at 00:00:42) It's a totally fair overview. I realized that there's this massive, massive disconnect between how the industry itself perceived itself, how the industry saw what was coming, and how basically everyone else saw what was coming—from my mom to Jamie Dimon—and that a lot of the change is going to happen not from inside of these organizations. And so I have spent the last two years building an advisory business and also doing research at University of Virginia and Hong Kong University of Science and Technology, studying the future of work and trying to bring to bear 16 years of applied AI experience and also academic exposure now to everyone else, as I say.

(Joel Beasley at 00:01:33) That's pretty cool. And so walk me through a little bit of your professional background—your credibility of those 16 years in AI.

(Zach Cass at 00:01:41) Started my career at a company called CrowdFlower, which became Figure Eight. It was the first Scale AI, basically. It was started by two Stanford researchers who were search scientists at Yahoo and figured out that there was—I don't know if your listeners are going to be familiar with search science—but they went from precision to DCG and DCG, which was just a more accurate way of informing the indexes on how to rank search results, and they could do DCG up to, you know, 20 or 25 really effectively. And it was actually what gave Yahoo almost a performance advantage. But the only way you could do it is with humans at massive scale, and so Lucas built this sort of front page on Amazon Mechanical Turk.

(Zach Cass at 00:02:26) Again, I don't know how much you remember about this, but early, early crowdsourcing technology to build training data for the search algos. And we turned that into a company, and that was CrowdFlower and Figure Eight and spent a while there with him. As we watched sort of statistical machine learning gain more popularity, as we watched the models proliferate, the number of companies that could effectively use these models expanded, and then went to a company, Lilt, which was also started by two Stanford researchers using large language models to solve machine translation. And there, I started to sell basically—I was one of the first people to sell large language models, and we were doing it in sort of this workaround way where we were selling enterprise translation services powered by the machine translation efficiency. So it was AI-enabled services.

(Zach Cass at 00:03:24) Long before people were talking about it in that way, and did that for a while and then graduated. Just before OpenAI launched GPT-3, the head of applied called Lucas, my first boss, and said, "Do you know anyone who can sell AI?" And Lucas is like, "There are not many people who have done it, so here's one of the guys." And got to, you know, launch 3, 3.5, 4, ChatGPT, 4o, and some other, you know, just amazing products at OpenAI. And we sort of watched as the world—as I went sort of from a world where no one cared about AI to a world where, you know, no one didn't care about AI.

(Joel Beasley at 00:04:08) Yeah. I remember that time very clearly because it happened while we were doing the show. That is super interesting. So when you were doing the go-to-market stuff at OpenAI, I mean, was that hard or easy? Because, like, for example, if I create a teleportation business tomorrow, everyone's like, "Holy crap, teleportation," and they're all just going to flood to it.

(Joel Beasley at 00:04:28) Was it hard or was it easy?

(Zach Cass at 00:04:31) Both. So what I think is important—so the answer is both. The expectations when you're working at a company like OpenAI are exceptionally high, and so the amount of pressure that you put on yourself is basically all relative. Right? It's like when you talk to friends who grew up to very successful parents, it's like, "Well, of course, you had so much going for you. How could this not be a really pleasant childhood?" And it's like, "Well, yeah, it's because my parents expected me to outperform them, and they had already outperformed their parents and so on." And so when you work at these exceptionally high-performance organizations, the pressure you apply is all relative, and our ideas of growth wasn't, you know, 5 to 10 million. It was 20 million to a billion.

(Zach Cass at 00:05:14) Right? I joined OpenAI at 2 million in revenue and left at 2 billion in revenue, and that's work. You know? That's—there's a lot that goes into that, and one of the key elements is maintaining talent density. So you need to hire an exceptional team and not compromise, which is just one of these incredible challenges.

(Zach Cass at 00:05:34) The other thing, and I think more to the point that maybe you're asking, is what is it like to work at a company where it's so exciting that the market is pulling you? And what I would remind people is OpenAI had been building important products long before we built ChatGPT. DALL-E predated ChatGPT, GPT-3.5 predated ChatGPT, and we thought that we had cracked the knot with the science. And what's really interesting and the thing that I would remind any CTO listening and any CEO or head of product is that GPT-3.5 was released in June 2022. Right?

(Zach Cass at 00:06:11) So four months, or a little more, four months prior to the release of ChatGPT. And it's important to remember this because the concept behind ChatGPT was actually a chance to show people what the API was capable of, with the idea that more people would use it. We were sort of sitting around like, "Why aren't people flooding the API? Why is it still so niche?" And the model that ChatGPT was built on, GPT-3.5, was publicly available for four months. Right? It was a slightly dialogue-aligned model, but not much. And the major breakthrough was the application layer, and I tell this story over and over because it really serves as one of my three epiphanies at OpenAI that in fact the application layer just matters as much, if not more, than anything else. And it serves as this constant reminder to anyone out there sort of feeling like they're stuck not knowing what to build, which is that building a great customer experience remains king.

(Zach Cass at 00:07:13) Right? Building a thing that people can use effectively that doesn't require material behavioral change is really important. And that, of course, sparked a bunch of interest, but it still didn't sell itself. Right? You still need to actually do a lot of the work, which is why OpenAI has invested so much in the last two years into the go-to-market organization because it's not sufficient to just have good technology. You must build great people to sell and support it.

(Zach Cass at 00:07:40) And so the answer, you know, is yes. Yes, it was easier than—when people ask for go-to-market advice, I'm like, "Look, I can give you plenty, but not from OpenAI. That is an anomaly." But not because it was easy, just because it was so different in scale and surprise.

(Joel Beasley at 00:07:56) It was so—they, to many people, they created intelligence. You know, to the average person, "Whoa. We can have synthetic intelligence now."

(Zach Cass at 00:08:05) Yep.

(Joel Beasley at 00:08:06) And then there's a lot of pressure because if the average person—if the uncle, if the aunt, if the grandpa is playing with the technology, then it's easy for people to understand it, and then they can talk to their board and say, "Why aren't we doing this type of stuff at our company?" And one of the conversations that happens a lot in our CTO space is pressure from the board to adopt AI and to show some return on AI, which doesn't happen a whole lot. They usually let technology kind of operate, and they have products, and they need some returns and stuff. But this is a whole lot of pressure right now from board members and the average people to their CTOs about implementing this.

(Zach Cass at 00:08:45) We're observing the consumerization of technology in basically real time. I mean, we sort of talked about it for a while, but it really is happening at the edge now where state-of-the-art technology—in this case, synthetic intelligence. I like that term, by the way. But the unmetered intelligence is arriving first at basically a teenager's pocket. And, I mean, I wrote about this the other day that more and more world leaders I'm meeting—they can remain unnamed so they'll continue to be willing to tell me about these stories—are learning about the next big thing from their teen.

(Zach Cass at 00:09:15) For, you know, they go home to the dining room table and they look at their child's phone and they go, "What is that?" And this is the ChatGPT effect, and it's actually—we, you know, I don't know where we're going to go in this conversation, but I would—one of the cases I make for why building in the open is so important. You know, Sora came out and everyone—I got phone calls from press and other people—"How could they do this? They're flooding the market. It's slop. It's dangerous. It's scary." And I go, "Listen. There are two alternatives to this world. The first is that we don't build this at all, and I don't believe in that."

(Zach Cass at 00:09:48) I think that we should build technology and sort of pursue frontier, within reason, obviously, alignment and explainability. The other option is they build it in secret. They build it and they don't tell anyone, or they build it and they only tell a few Hollywood studios. Right? They sell it to the highest bidder.

(Zach Cass at 00:10:07) And what's amazing right now is we're building state-of-the-art at the edge, at the consumer edge, and it's turning—it's creating a very strange power balance because a bunch of CTOs who used to have early access to the best technology, or sometimes late to the party, because it's their board or their board's teenager who's discovering the latest, greatest thing on a YouTube video, on a GitHub thread. And that's—I got no issue with that, frankly. I mean, I think that, you know, I say this with sympathy for the CTOs who are getting tremendous pressure to build things that are maybe not appropriate to build, which we can talk about separately. But I think it's pretty cool that everyone gets to see in real time how—where the technology is going.

(Joel Beasley at 00:10:52) And I get it. I have empathy for both sides. You know, you see your team at the table using this new technology, and you were just in a meeting earlier where the software people are lagging behind.

(Zach Cass at 00:11:02) You know—

(Joel Beasley at 00:11:02) You're like, "Why aren't we doing—why aren't we getting some of this intelligence into our company?" Yeah. One question I was really curious to ask you is, do you ever get into the legislation side of conversations at all in your work?

(Zach Cass at 00:11:15) So, I'll answer that, and I also want to speak to something you just said, which is I think one of the funniest things about this moment. I mean, there's a—you have to pardon the crassness, but there's that teenage sex joke about, like, you know, "Everyone talks about it. No one's doing it." And AI can—the more things change, the more they stay the same, which is like it can be so performative. It is so easy now. The cost of creating value is coming down, but so is the cost of creating a little less than value. Right? So when I say that, what I mean is lots of people now are building these really cool demos that cost very little, that are really interesting to look at but aren't actually doing a lot because they don't change behavior, they don't actually meaningfully produce a new outcome, and so then you get these situations where the board is now staring at something that isn't even valuable but just looks good and they're like, "Why don't we have this?" Which is, I think, a lot of the demoware going on right now. And it's not to say that AI can't do incredible things, but so much of the change happens in a traditional people way.

(Zach Cass at 00:12:22) And the reality is that just takes an enormous amount of work. So my sympathy goes out to the CTO doubly because I think they are getting pressure that is often—sometimes unfair and misapplied. Even though this technology is super powerful, there's still this incredible misunderstanding of where it should be applied. As for the legislation question, I spent a bunch of time on this. I got really interested in sort of the last three years on this role that emerged—the responsible AI function and person—which sort of appeared overnight. A bunch of people had been studying ethical AI their whole lives, it seemed.

(Joel Beasley at 00:13:09) Of course. Just update your LinkedIn bio, and that's it.

(Zach Cass at 00:13:12) Yeah. So I got really fascinated, mostly because I wanted to see what these people were saying and what the grift was. And I'm not—I don't use "grift" totally pejoratively. I use it sort of tongue-in-cheek. But a lot of people overnight were like, "Look, I'll show you how to do responsible AI."

(Zach Cass at 00:13:29) And what I pieced apart is that there was a bunch of, you know, there's a lot of just ways to extract revenue from enterprises who were terrified of being irresponsible. Right? The alternative to responsible AI is irresponsible AI, and no one wants to be that, so everyone has to go hire responsible or ethical AI people. But what I try to do when I talk to CEOs—I spend a bunch of time on this and policy makers—is pick apart what I think ethical AI is. And first I remind people that there's a lot—there are a lot of laws in place in this world. And a lot of these laws are meant to and effectively govern what enterprises can do.

(Zach Cass at 00:13:59) And they have been applied sort of carefully across technology for a while, and a lot of these laws do not change. Right? There's just a bunch of ethical practices that don't change because of AI. But there are a few that we need to reconsider. And two of them, unfortunately, happen much further upstream—sort of, two of them happen really at the point of research where the models are built. One is alignment and one is explainability.

(Zach Cass at 00:14:20) Explainability can happen a little bit further down, and the third, bad acting is critical and happens at the point of application and the policy to protect us. And I talk a lot about basically these three and not much else. When I talk about legislation, I basically am like, "Look, there is a bunch of stuff you're going to be inclined to policy. You should avoid all of it that is performative, and you should focus on making sure the models have an appreciation for the consequences of their actions and are aligned to human interests. There are ways that we can do this. You should enforce explainability, especially when the models are being used in precision high-risk environments. Think, loan approvals or even small claims courts."

(Zach Cass at 00:14:54) What's important is not that the model gets it right every time. What's important is the model can explain itself. And then the third is bad acting. I don't think we fully appreciated that we are massively empowering a low-resource bad actor. We spend so much time worrying about the nation states and the Bond villains, and I am principally concerned about the psycho in the garage or even just the criminal who has figured out that they can do a whole lot more with a lot less. The cost of doing bad stuff is coming down.

(Zach Cass at 00:15:47) And so this is where I crusade, and we've had success on sort of all three bases, both at the state, federal, and international level. And I really try to encourage others to just talk on these terms because it would clarify a lot of this dialogue.

(Joel Beasley at 00:16:02) Real quick, before we continue, I wanted to give a special shout out to DigitalOcean for sponsoring this episode. DigitalOcean is cloud infrastructure that's simple to spin up, complete with integrated AI dev tools plus 99.99% uptime SLAs and industry leading pricing on bandwidth. We all know DigitalOcean. It's super reliable. It's been around forever.

(Joel Beasley at 00:16:24) And now with DigitalOcean's Gradient platform, developers can train, fine-tune, deploy, and scale AI workloads all in one place. DigitalOcean is reliable and affordable at any budget. Companies can save up to 30% on their cloud bill when they come to DigitalOcean. Use code DO25 and get $200 in free credits to get started. Simple cloud, powerful AI, built to scale.

(Joel Beasley at 00:16:49) That's DigitalOcean.

(Intro Narrator at 00:16:51) Now, back to the episode.

(Joel Beasley at 00:16:53) Can I share with you an idea and you could tell me if it's stupid, has merit, if it's kind of okay but has holes, and then can we discuss something?

(Zach Cass at 00:17:02) The preface makes it sound like it's already gonna be good.

(Joel Beasley at 00:17:05) It's already good. Yeah. It's good. Okay. So I was thinking, I'm really into health, right? And so I always look at the nutrition labels. I want to know what's in the food that I'm eating. I try to eat as close to real food as possible. And so I was like, this is great. All of these companies are legislated. They have to do this through the FDA, put the ingredients, and they appear in the order in which the largest quantity. Right? Then I go and I use, let's just not name brands. Let's just say every major brand LLM out there that a consumer can pay $20-something a month for. And they all have these obvious default master prompts that we can't see. They're not transparent and they're injecting ideology of how the model should operate. I'm not saying which ideology. They have an ideology of how, and it's decided in this back room at these companies and I can't see what I'm consuming. Now, I'm okay if one thinks very differently because of the culture or whatever. I'm okay that there's a bunch of models with a bunch of differences. What I'm not okay with is the fact that there's no transparency as a consumer for me to make a choice. I want to see the master prompt for each service and then choose which one I want because there might be some services based on the culture that I'm living in that doesn't align with my values.

(Zach Cass at 00:18:24) You're describing explainability. I mean, do, do, do—

(Joel Beasley at 00:18:27) This is explainability?

(Zach Cass at 00:18:28) This is explainability. So what you just described is, and you did it very, very thoughtfully, it is not a crazy idea. It's a super reasonable idea, and we should all require this. One of the things that people always ask me, Joel, is how do we reduce bias in the model? Now stop me if you've heard this before, but I have to take the opportunity to explain to people that bias is not actually a pejorative. Well, it's a pejorative, but it shouldn't be. The term bias itself is in fact neutral, and what I have to remind people is that all thinking machines, all humans, all thinking machines, all AI systems have bias, and they only work because they're biased. We only work because we're biased.

(Joel Beasley at 00:19:09) I think we're gonna be best friends. Oh, my gosh. The number of people I've had to explain that to is unbelievable. I was like, you cannot function as a living organism without bias. How do you live? How do you survive? You're gonna drink motor oil because you can't bias it against water? What is your deal?

(Zach Cass at 00:19:24) Insert Stepbrothers meme. Josh, you can do the Stepbrothers meme there.

(Joel Beasley at 00:19:31) What they're really saying is values, though, right? They're really saying—

(Zach Cass at 00:19:34) No. Sort of. Sort of. You almost hit the nail on the head there. And what you said was, are you gonna drink motor oil? No. You're gonna drink water because you know that you need water. There are some things that actually we know we need. Where bias is super interesting is we would be zombies. We would be basically functional zombies without bias, but we would perform the rudimentary actions that we are sort of hardwired to perform. And without being crass, you can sort of think of what they are. What is important to remember is that there are a bunch of things that we are born knowing how to do. Breathing is one. Now you don't need to be biased to breathe. You just need to have a brain that works. We sort of have these pre-wired ideas of living. Humans over time have sort of come to understand lots of things. What's important and remarkable about humans is that our brains have now filled with so much information that we cannot possibly unpack all the things that happen inside of our brains. And this is what I have to remind people, which is the problem in the world is not bias. The problem in the world is inexplicability.

(Zach Cass at 00:20:55) Humans are exceptionally inexplicable. In fact, not only are we inexplicable to each other, we're inexplicable to ourselves. A lot of the pain in living is not knowing why you feel a certain, is wrestling with a feeling that you cannot make sense of. It's not that you're sad, it's that, you know, your depression doesn't, why are you—people always say, I don't know why I'm sad, and it makes me even more sad. Right? Everything's going well, but I can't find joy, or I don't know why I'm so angry right now. Or I woke up really happy today. That's amazing. You know, it must be the astrology. Humans are inexplicable. There is not enough therapy and ayahuasca in the world for any individual, you or me or anyone else, to unpack all the things that we believe, that we—the wiring in our brain is so complex, and we store so—we store everything that's ever happened to us in there, and we could never access all of it. Moreover, Joel, even if we could, we would never admit it. We would never, ever tell another soul all the things we believe and why we believe them. There are just some memories that are so taboo that they should never be shared. And frankly, a lot of decisions we make are so sort of inappropriate from where they come from that it would be professional or personal suicide. Right? A judge would never say, I'm sentencing you to six months because I woke up on the wrong side of the bed this morning. I had a terrible, I didn't sleep well last night. I'm pissed off, and you get an extra six months. That happens all the time. That happens all—

(Joel Beasley at 00:22:29) Did you hear the study they did where they took the board of parole and they mapped the time? Did you see that one?

(Zach Cass at 00:22:38) Yes.

(Joel Beasley at 00:22:38) And they found out that they would deny the parole before lunch because they got hungry or in a bad mood.

(Zach Cass at 00:22:44) I have a friend who applied over and over to get a visa three times, and then they finally got a good—they finally found a good visa lawyer, and the visa lawyer said, actually, you're applying to the wrong office. That office is famous for rejecting. It is that bad. It's just, it's one of these situations where we live in a world that is unfair principally because we often can't explain it. It's not even that we think the rules are broken. Sometimes they are. Now I say all this because the reason that bias is a pejorative is, I think, our association with bias as something inexplicable. When a parent, you know, you'll be at a soccer game, or the scenes in the movies if you don't have kids, and some parent will say, oh, the coach put their son in because he's biased. It's like, that's true. That is exactly how bias works, but no one ever says, oh, that really kind person picked that woman up, or helped that woman with her bags because they're biased. Right? We never say someone is biased when they do something good. We only say they're biased when they do something bad even though most bias, it turns out, is actually good. Right? Most people do pretty good things throughout the day. We make mistakes, but a lot of our bias tells us to do really nice things, and we never use that. We always say, oh, you know, you're so kind, you're so thoughtful, you're so gracious, but it's never your bias to do good. Now, I say all this because I'm on this crusade, it sounds like you are, to remind people that bias is not a pejorative. It's not inherently bad. And what we live in is a world that is not filled with bad bias. We live in a world that is filled with bias we cannot make sense of. That is just not how we are wired, and frankly, I don't want to unpack all the things in my life. There are plenty of memories that don't need to be accessed. And so what we have an opportunity to do now with AI is actually to build an explicable world. We have an opportunity to build a world that makes more sense to us by building machines that are required to explain themselves, not because their decisions are perfect, but because we can make sense of their decisions. It is, the assumption should not be that machines will always get it right. The assumption should be that machines will always tell us exactly why they did something so that we can then action—

(Joel Beasley at 00:25:05) Yeah. Okay. I like that. It solves this—it's the same end result. That's right. Nutrition labels versus just asking it, hey, tell me what your default thoughts are.

(Zach Cass at 00:25:16) Joel, do you have kids?

(Joel Beasley at 00:25:17) I have three. Yeah.

(Zach Cass at 00:25:19) Okay. So parents will always say to me, I'm really worried about, you know, LLMs teaching my kids. And I go, that's interesting. Have you ever been to a school?

(Joel Beasley at 00:25:28) Yeah. Well, we homeschool, but—

(Zach Cass at 00:25:30) There you go. Okay. Enough said. I'm preaching to the—we are best friends. I'm preaching to the choir. My wife's a first grade Waldorf teacher, and she's like, listen, Rudolf Steiner has a pretty clear understanding of what should and shouldn't be taught, and still, you know, Waldorf teachers go off the reservation because of their biases. And when parents are like, you know, what is this LLM teaching me? I'm like, nothing different than what a teacher might. If you think ChatGPT gets it wrong, you should go into a classroom. Wow. And it's not because teachers are bad. It's just because the—

(Joel Beasley at 00:25:58) Correct. The job is very hard.

(Zach Cass at 00:25:58) And what you never get a chance to do as a parent is say, show me all the things you believe, show me all your values, show me all your genetic preconditions, show me all of your lived history so that I can know what it is you're biasing my child with. You don't get to do that. You meet the teacher once or twice at a parent-teacher conference and you go, oh, I like them, I think. Right? I guess. But with the models, and I'm not saying we should hand over teaching to LLMs. I'm not proposing that at all. I don't want to sound like I am, but I'm saying it gives us a chance to say, show us all the underlying decisions and information that led you to, or that informed this curriculum or this suggestion or this response. And that is all we can ask for from the machines that are never really gonna get—

(Joel Beasley at 00:26:51) Now, can we legislate that though?

(Zach Cass at 00:26:54) Yes. In fact, not only can we, we must. I am on a crusade that alignment, explainability, and bad acting are the three things that must be legislated and probably the extent to which we should legislate for now.

(Joel Beasley at 00:27:06) But how is the bad acting not already legislated?

(Zach Cass at 00:27:09) Well, we have not yet, so it's starting to be, but let me give you an example. The cost, the penalty for being caught trying to do mail fraud or wire fraud in the United States sort of scales on a per-action basis. So, obviously, Bernie Madoff got, you know, 500 years. But if you try to, if you get caught phishing someone on email, you can only be tried for the amount that they can prove that you were trying to phish them for. Now, the reason why, Joel, you can answer. Why is that flawed? Well, because whereas, historically, you had to do like Robert Redford, Paul Newman, The Sting in order to steal from people. You had to do these very elaborate financial fraud situations. And so you could basically say this person was caught doing this, and this is the extent of the damage. Phishing is exceptionally inexpensive. So we have to assume, we have to assume intent to sell or, you know, whatever the terminology in drugs are, which should carry with it a much bigger penalty because the cost to doing it is so low. And so we have to create this new artificial penalty that doesn't exist. So if you get caught scamming someone over email or catfishing someone, you get slapped on the wrist. Sometimes it's a fine. Sometimes, you know, some DAs will not try it. They'll figure it out on their own. I am advocating that in fact, if you get caught trying to steal from someone using technology, broadly defined, but deepfakes, et cetera, we should assume you're doing it to a lot of people. And we should assume that you are an incredible threat to the fabric of society and trust. And I advocate aggressively for very punitive policy to punish and, principally, to terrify people who consider using AI to tear the fabric of society. Because if we don't, I think it's gonna get really weird.

(Joel Beasley at 00:29:04) Dude, I did an interview with an insurance guy a couple years ago that does insurance for data loss and wherever they lock up your data, you know, and hold it hostage. And he was saying, you can buy these things like franchises on the dark web and just let them run. And then you just basically provide support for the, you're the customer service for all the people you're scamming and locking up their data. And it blew my mind.

(Zach Cass at 00:29:33) I think that we should try these people like, well, I don't want to lose followers. But I—

(Joel Beasley at 00:29:41) I'm losing followers today. That's what the engagement was. I was like, Zach, come on. Let's rub everyone wrong.

(Zach Cass at 00:29:47) I think that we should treat these crimes as crimes that tear the fabric of society. Like, these are not—these are—so Florida recently passed a law that I advocated for and helped with, that sort of basically says if you get caught stealing from a senior citizen in Florida using technology, it's an extradition law. You get extradited to Florida. And they're trying to put people away for up to 25 years for attempted financial crime. And my feeling is good. Like, do it. Let everyone find out that this is a scary proposition. It is not something you should consider doing. And if you're a bad guy in a room with other bad guys sort of going through the list of the bad guy ideas, pass over the one around deepfakes, pass over the phishing one. And we just, we do not want to devolve into a low-trust environment. So much of what makes, especially this country, but so much of what makes the world livable is knowing that you can trust when your phone calls, you know, when your phone rings. It's knowing that you can trust a text message from a friend. It's knowing that you can respond to an email safely. And if we lose that, it's gonna take a lot to get it back.

(Joel Beasley at 00:30:56) Yeah. I agree. I think we should make, in general, I'm gonna make a blanket statement. I would like to see higher penalties, but I'd also like to see the age change. I don't think you're an adult until you're 30. I mean, I don't think your brain, you might have been an early developer, but my brain didn't start really working till my late twenties. So I'm like, I think we should raise this—not popular, but not many people talk about this specific topic anyways. But because like, look, when if you're 13 and you're at home and you're screwing around hacking stuff, I don't think you should get 50 years, you know, but if you're 30 and you do that same thing, maybe because we need to know that there's serious penalties for doing it.

(Zach Cass at 00:31:38) It's funny. I take the other side. I think that younger and younger people are more and more capable. I'm not—I agree that your brain isn't developed, but I actually think that part of the problem is that we—I think that we should create more accountability for younger people. They're gonna start entering the workforce sooner.

(Zach Cass at 00:31:55) I mean, Alex Karp and Sam Altman have already done away with the college requirement for employment because they're like, young people are super capable. I mean, thanks to AI, largely. And moreover, the tools that they have—I mean, it used to be, you know, when you and I were coming up, we didn't have a lot of tools to learn how to cope, to learn how to program. So the amount of harm we could do is pretty limited. I mean, we smash mailboxes.

(Zach Cass at 00:32:14) You know, I'd throw the local golf flags into the nearby lake. That's okay. When people start, you know, doing things like chaining the other models to fish and then scam and doing it at scale, I mean, yeah.

(Joel Beasley at 00:32:32) There's an argument for that.

(Zach Cass at 00:32:34) Yeah. Scattered Spider. You know that group?

(Zach Cass at 00:32:37) The crypto theft ring run by, like, four 19-year-olds in Canada and the US. They stole whatever, $200 million. I don't know, man. You're gonna have to convince me those are not adults doing adult things.

(Joel Beasley at 00:32:54) Yeah. Yeah. No. I mean, look, you're exactly right.

(Joel Beasley at 00:32:59) No, no. I'm—that was not an incredibly strong belief that was held really tightly. That was just in general watching how stupid and undeveloped kids are. I'm around a bunch of eight-year-olds and stuff, dude. I'm like, your brains are—you are not capable of decision making. But I forgot that, like, yeah.

(Zach Cass at 00:33:17) It's just, it's a long day as a parent sort of reminding yourself that kids are not—yeah.

(Joel Beasley at 00:33:22) Right. So we homeschooled for the first several years. They're eight, six, and three. We homeschooled, and then we moved—we lived on a farm, and we lived out in the middle of nowhere. We homeschooled. And then we moved into this neighborhood about three months ago, and their friends, the neighborhood kids, go to school. So they approached us three months ago, and they're like, "Hey, can we go to school?" And I said, "Well, I gotta go meet the teachers first," because at their grade, they're with one person.

(Zach Cass at 00:33:46) I gotta know their biases.

(Joel Beasley at 00:33:48) Yeah. I did. And so I walked in. I walked in there, giving us the tour, and the teacher goes, "Do you have any concerns or questions?" And my daughter goes—she goes, "My daddy wants to make sure you guys aren't woke." And they laugh so hard. But where are you?

(Zach Cass at 00:34:06) Where are you? Are you—can you tell me?

(Joel Beasley at 00:34:08) Nashville.

(Zach Cass at 00:34:09) Yeah. That side of Nashville. They probably had a good chuckle.

(Joel Beasley at 00:34:12) They had a good chuckle. Yeah. And so we actually sat down and had a conversation with the two teachers. One for my son, one for my daughter. And we talked about their values. And obviously, we didn't see eye to eye on everything, but it was 80% there. Yeah. You know? And I was like, "Okay, I'm comfortable with my child spending seven hours a day with you."

(Zach Cass at 00:34:31) That's great. How's it going?

(Joel Beasley at 00:34:33) They love it. My son, the younger one, my son's like—he likes it, but if we wanted to go to Disney World, he'd be like, "We're going to Disney World." My daughter is like, "I don't even wanna go to Disney World. I just wanna go to school." She is incredibly smart. Wow. And she's really taken to it well. So—

(Zach Cass at 00:34:52) That's awesome, Joel. And sorry, what are their ages?

(Joel Beasley at 00:34:55) My daughter's eight. My son's six, and my third is three. So he's not in school right now. Yeah. What about you?

(Zach Cass at 00:35:01) My daughter set to be born in 18 days.

(Joel Beasley at 00:35:06) Okay. Dude, you're in it right now. You're on high alert.

(Zach Cass at 00:35:10) Yeah. So baby coming and all of this is super—I mean, it's so top of mind. And actually, the book—I wrote a book coming out January 13th and basically wrote it with my daughter. I started the process. We weren't yet pregnant. And then sort of as the book got closer and closer to the manuscript deadline, the pregnancy became more and more real. And then I just started thinking about everything from the lens of, like, what is this experience gonna mean for my child, whatever it was. And so I now sort of find myself advocating for lots of things that I think I'd believed, but—or known, I should say known, but now I believe.

(Joel Beasley at 00:35:48) I cared about nothing before kids. I would tell people I'm apolitical. I'm a technologist. I create things. I spend all of my energy creating products and businesses. I didn't get involved in any of the social stuff at all. And then I had kids, and then it just became unavoidable. Yeah. It's like—and you could kinda see how decisions and stuff between different parents and how their kids start to develop. And it's like, "Whoa, we're all raising our own little large language models." It's crazy. And we're all programming them. It's wild.

(Zach Cass at 00:36:19) On that point, I will say something that I think is unavoidable and hard for me to consider, but it's something I've thought about a lot, which is, you know, you and I are sort of this generation where the last generation where we didn't—we sort of grew up with the internet, but not high-speed internet. So what we learned, we savored. I remember going on the—dialing into that 56K modem and going to a web page and being like, "Wow. I just read—I am the—someone ask me something because I am the smartest man alive. Right? Grandma just sent me an email. I just read a web page about, you know, who knows what. I just spent an hour on eBaum's World. Ask me anything." And what I've realized in talking to young people is so much of what I attribute to the angst in the younger generation today is, of course, one, device addiction. I just think the screen is destroying lives in ways we don't yet appreciate or do, but haven't really priced in. But I actually think the other issue is we just know too much. It's like the large language models, you know, we've had these very powerful neural nets and they've developed quite a bit for a while, but all of a sudden, we've spent our entire lineage sort of knowing nothing. You know, knowing very little. Right? And then within, you know, in a matter of 30 years, we said to every child on Earth, "Watch this screen for eight hours. We'll watch everything. All the horrors of the world and everything else, and then tell us how you feel. And by the way, you know, why are you complaining? Go enjoy your day." And the consequences of this are remarkable, which is the model itself hasn't updated, but the RAG, the amount of retrieval information has. And so you don't actually have a system that is effective at parsing any other than a prior model. It is, you know, it is the equivalent of a GPT, but it's now being asked to make sense of so much more information. And so I just sort of attribute so much of the current state of angst to just the fact that people know too much. They could be just staring at all of the problems in the world despite the fact that things constantly get better everywhere all the time. I mean, with some—

(Joel Beasley at 00:38:31) Significantly, yeah.

(Zach Cass at 00:38:33) We are filled with so much dread because we never figured out, you know, how to retrieve the right information, and so we're filled with selection bias, confirmation bias, negativity bias, and my favorite, rosy retrospection, which keeps telling us that the past is better than it actually was. And I'm like—

(Joel Beasley at 00:38:51) It's not.

(Zach Cass at 00:38:52) Yeah. The problem here is that the kids know too much. Like, they just know too much. And we should start acknowledging that we overexpose them to the horrors of the world and then ask them why they were having a bad day. And that is something I think about all the time, and I do not know how to escape that. I don't know—I know the world my daughter will come into is full of wonder, and I think we'll be on a better path than the one we're on today. And I also have to wonder, you know, how she's gonna—how we're gonna manage the neural net that is gonna be overexposed to just bad stuff.

(Joel Beasley at 00:39:24) So what I've seen over the past eight years of being a parent and doing this show and all of this, for me, it just forces you to become a more involved, better parent. So I think the kids that have absentee parents—when we were a child, they could get into some trouble and stuff, but not like real addictive trouble. You know, it's like, it's really hard. It's so accessible. It's so, like, it's a Swiss Army knife. The screen can do everything. It can be the alcohol that you shouldn't get into, but it can also be the first aid kit that you need. It can be so many things, and it's so accessible and so cheap, and it's all through one device. So that means if you are wild and free and you don't have involved parents as a kid, you can really go far down the wrong path. And for us raising the kids, we've just been, you know, making up our own rules as we go with watching their stuff. We've got the—they're all on Apple devices, and we all have very—we're very involved with the restriction systems on the Apple devices. And we—you know, they have limits to their different categories of applications. Like, they can do their educational apps more than they can play some games.

(Zach Cass at 00:40:40) Seems really smart. I mean, the rules-based systems clearly are tough. When people ask, "How are you—" you know, there's always the gotcha. "What are you—how are you gonna raise your—you know, are you gonna give your daughter social media?" And I'm like, "Look, who knows what the device is even gonna look like by the time she arrives at that age?" I said, "All I can hope is that she loves the outdoors." The only thing—I really, really want to over-invest in the alternatives to a digital reality, and I'm sure that the technology is gonna carry her far. I'm just desperate to create a world where she is excited to be with friends and family in the outdoor spaces.

(Joel Beasley at 00:41:16) I can give you that answer right now.

(Zach Cass at 00:41:17) Tell me.

(Joel Beasley at 00:41:18) It doesn't matter what you tell them. They're just gonna do what you do, and they're gonna love it. They just wanna—you're the hero, and they want to do that. So if you're doing bad things, they're gonna copy you and do bad things no matter how much you tell them to do good things. If you do good things, they're gonna do good things even if you told them to do bad things.

(Zach Cass at 00:41:35) Well, good thing Adley is my wife then.

(Joel Beasley at 00:41:41) Yeah. So they're gonna do—is she in the outdoors?

(Zach Cass at 00:41:46) Exceptionally. Yeah. She's—I married my sort of spirit animal, as I say to people.

(Joel Beasley at 00:41:53) Oh, there we go.

(Zach Cass at 00:41:54) You're talking about a woman whose phone can die, and she won't notice for a day.

(Joel Beasley at 00:41:58) That's great.

(Zach Cass at 00:41:59) Loves camping, first grade Waldorf teacher, and elite athlete who is—you know, her idea of a late-term pregnancy activity is to play volleyball for a few hours.

(Joel Beasley at 00:42:13) There you go.

(Zach Cass at 00:42:13) Yeah, it's all I could have ever dreamed of, and I just didn't even—I had no idea how important it would be. I knew it was important, and then AI started to catch the pace it did. And now it's so unclear to me that, like, truly, the future currencies are physical spaces. I mean, there's just gonna be so much digital abundance, and we just—all we have to impart on our children a love for community and outdoors.

(Joel Beasley at 00:42:38) Yeah. Well, we, in COVID about three or four years ago, we sold everything. We bought an RV, and we did a year-long camping trip around the country.

(Zach Cass at 00:42:46) Wow.

(Joel Beasley at 00:42:47) So the kids got outdoors. They were playing in the national parks every day. It was—that's why we homeschooled originally was so we could do the RV thing.

(Zach Cass at 00:42:55) Favorite—this is, I guess, so reductive, but favorite moment or a special moment that you captured from that trip?

(Joel Beasley at 00:43:05) Yeah. The most special moment is we were on our way to Nebraska, and we had to stop in Tennessee. And it was, like, close to summertime or at summertime. And we were born and raised in Florida, so that's where we were coming from, which is hot every day and muggy every day. And so we get to Tennessee, and it's summer, and it's cool in the evening. And I said, "I wanna live here." We then proceeded to continue to travel for nine, ten months, and we ended up buying within an hour of that place that we stopped in Tennessee. Wow. Yeah.

(Zach Cass at 00:43:38) That is a special moment.

(Joel Beasley at 00:43:40) That was a special moment. Yeah. It was just a one-night stop while we were driving through, and we just went on this walk, and the people at the campsite next to us were cool, and it just had really good vibes, for lack of a better term, and—

(Zach Cass at 00:43:54) Cosmic intervention.

(Joel Beasley at 00:43:56) Yeah. 100%. Hey. Side question just for fun. What's the first piece of content you've ever put on the internet?

(Zach Cass at 00:44:03) Well, I mean, in a public domain or private domain? Because private domain obviously are emails, like early AOL email or AIM messaging.

(Joel Beasley at 00:44:11) Public post that you could track back down?

(Zach Cass at 00:44:13) I assume that it is the first Facebook post I ever did, and I think it was—if I recall, it was a photo in senior high school. I took a selfie on a traditional, you know, camera, digital camera, of me and four friends in a Lufthansa. Remember those three-four-three—well, I think they still exist. I don't go to the back of the plane much anymore. But those three-four-three crazy economy cabins in Lufthansa, where they just jam people in, I took a digital camera selfie. The whole row was my high school friends. Oh, nice. I love this photo. And I then posted on Facebook, and it was—I think I'm fairly certain the first photo I ever posted. I think I'd still be there. I haven't logged into Facebook in quite some time.

(Joel Beasley at 00:45:03) That is awesome. For me, it was 1996, and I had posted—I was trying to—I don't know if it was cheating or just being smart about my homework, but I had a report due on sea slugs. And a guy had a blog on sea slugs, and I posted a message to him asking him essentially the questions that were in my worksheet about sea slugs. And he responded two years later, but, uh, that's like—you could see Joel Beasley, you know. So I looked it up on the Wayback Machine and took a screenshot of it a few years ago because someone was asking me about it, and that's it.

(Zach Cass at 00:45:47) I was pretty active on the Tom Clancy Rainbow Six gamer board in eighth grade, 2000 or—yeah, 2000 and 2001. I was pretty active on the Tom Clancy Rainbow Six message boards. I don't really know why. I love that game.

(Joel Beasley at 00:46:10) Rainbow Six 3 was the best, of course.

(Zach Cass at 00:46:10) I loved that game. Yeah. I just remember telling everyone that I was gonna play Tom Clancy and that everyone needed to unplug the phones. I was like, "Listen. If the phone rings or if you pick up the phone, it's gonna ruin this whole thing. So just let—I'm logging—I'm going—I'm dialing into the modem." I do remember that. I remember all that.

(Joel Beasley at 00:46:31) I want to talk — I know we only have about fifteen minutes or so left — but I wanted to talk about the societal thresholds. I think you've written some white papers on this. Can you give me the brief overview of that?

(Zach Cass at 00:46:42) Societal threshold. Yeah, we've written a few white papers now that have gained some momentum. Unmetered Intelligence is sort of like the principle one, and we predicate the book on this idea of, you know, what if intelligence becomes a resource? What if we treat intelligence like a resource that is just declining now precipitously in cost and therefore increasing in abundance? And one of the questions that is born from Unmetered Intelligence is this idea of technological and societal thresholds. Technological threshold asks the question, what can a machine do? And the societal threshold asks the question, what do we want a machine to do? What are we willing to let a machine do? And societal thresholds and technological thresholds have always coexisted, and often with a space — what we call the adoption gap — between them. And sometimes that space has felt very pronounced, but basically never more so than it does today. And the reasons for this are sort of obvious. But before I go into it, I'll tell a story that I think explains societal thresholds because it makes it very real for most people. I was trying to figure out, like, what is the best way to explain societal thresholds, and we studied a bunch of stuff.

(Zach Cass at 00:47:51) And there are tons of really interesting examples: nuclear power, GMOs, chemicals like DDT, right? Things that we adopted and then actually rolled back over time. The elevator is my favorite, and the elevator story goes like this. Elijah Otis discovers a better way to move people up and down tall buildings. Previous to this, we'd been using rope and pulley systems in mine shafts and tall buildings. And in 1854, he introduces what he calls the People Mover, a safer way to move people up and down these buildings, and it comes with basically a fail-safe that catches the pulley if it breaks. So it's a second parachute, and it's empirically a much, much — it's this massive improvement that actually exists today in one form or another, this thing that if the mechanism breaks, it actually catches so elevators don't plummet. And he starts selling a bunch of these People Movers. At the time, it's not called an elevator. And the People Mover captures a bunch of attention from building owners and they buy a bunch. But within a few years, 1856, Elijah starts getting all these phone calls, and it's the building owners. They're saying, "Hey, no one's riding the elevator." "Why not?" "People are afraid they're gonna die still."

(Zach Cass at 00:49:12) So Elijah's met the technological threshold. He has built a machine that empirically is safe and effective at moving people up and down, but he hasn't met the societal threshold. He hasn't convinced people that they should use it. So he goes back to the drawing board and he comes up with three ways he thinks he can convince people to ride the elevator. The first is he puts a mirror in the back of every elevator, so when you walk in, you'd be distracted by your own image. The second is he puts music in every elevator, so when you walk in, you'll be entertained. And the third is he puts an operator in the elevator, a physical person in the elevator to give you a sense of human agency and control. And of course it works. It moves the societal threshold. People gradually start to adopt the machines, and then the rest is history. We don't think twice. I actually — I like the story particularly because I do think twice. I am terrified of crowded elevators. That's a separate psychological problem. But it is this, you know, it's this funny reminder that, like, just because something works doesn't mean that we want to use it. And when I tell people this story, they laugh. You know, they're willing to chuckle at it. And of course, then I ask them. I said, "Well, okay. What do you think our elevator is?"

(Joel Beasley at 00:50:18) Tesla's self-driving right now. I want them to release it. I want them to. I know they can. It's great.

(Zach Cass at 00:50:24) You nailed it. I mean, actually, I argue that of all the things that have been said about autonomous vehicles, not enough has been said because we're fixated on all the wrong things. We're talking about how well it works or how will we make the roads, you know, forward compatible, what will it do to the economy. And I'm like, no, no, no. You're missing the whole point, which is that right now, globally — but pronounced in the developed world — autonomous vehicles, Joel, poll at 25% approval. They are exceptionally unpopular. And actually, it appears that they used to be more popular. Now some of this is an Elon thing. Some of this is like this political flashpoint, and it's just like this bad confluence of events. But actually, in studying it, we sort of unearthed some other very interesting truths that I think tell the story about AI societal thresholds. How and why will we adopt AI systems? And the first is that humans — so the first reason that AI societal thresholds move slowly as told by the autonomous vehicles is that humans love control. We love control.

(Zach Cass at 00:51:33) And we understood this really well, having done a bunch of different things, but we went to Disneyland and we studied the ride Autopia. Have you taken the kids to Disneyland, by the way?

(Joel Beasley at 00:51:41) Yeah, we have.

(Zach Cass at 00:51:42) Okay. Disneyland in Anaheim.

(Joel Beasley at 00:51:44) Nope. Disney World.

(Zach Cass at 00:51:46) Okay. So Disneyland in Anaheim features a ride that was built in 1954 called Autopia. It is one of the seven original rides at Disneyland. It is a cement track on which children aged three through eight ride around in gas-powered go-karts about two miles an hour.

(Joel Beasley at 00:52:01) Okay. I know what you're talking about.

(Zach Cass at 00:52:02) They operate on a track literally like this wide, six inches wide, that gives them this much wiggle room so you don't have to turn the wheel. And Disney kind of hates this ride, and in fact, they are finally updating it. But the reason it took them so long, despite the fact that the ride is very smelly — it actually emits enough fumes that people working the ride have to change shifts regularly because it gives them headaches — it sounds bad, it looks bad, it's a cement track, it has no Disney affiliation. And the problem with this ride, as far as Disney is concerned, is that since 1954, its inception, it has been the most popular ride for children age three. And when you learn this, you appreciate the most remarkable truth that more so than the Matterhorn, more so than Space Mountain, more so than the teacups, kids want to drive a car. And it is this reminder that even a child who has no appreciation for what a vehicle is or what it offers them knows that controlling a machine is fun. And relinquishing that control is very hard for us. And this is just one of these incredible reasons that we are gonna struggle, even if you hate the monotony of your commute, to actually give up the steering wheel, or even if you appreciate how dangerous it might be. Second reason: technological thresholds are moving too quickly. So when we started doing the study, autonomous vehicles were actually not that good. Empirically, human drivers seemed to be safer. And in the sort of three years — in the time that we released the book — something amazing happened, which is that autonomous vehicles got really good. I mean, empirically better than human drivers. And we expected it to move the polls. Guess how much it moved the polls?

(Joel Beasley at 00:53:48) Little to none.

(Zach Cass at 00:53:49) Not at all. In fact, it appears that autonomous vehicles were more popular three years ago.

(Joel Beasley at 00:53:54) Yeah.

(Zach Cass at 00:53:55) And 25% of people now say that they are open to it, 50% say they are disinterested, and an incredible percentage say they would never — they would advocate aggressively against it. And this, to me, I used to think, "Oh, humans don't risk adjust well." That may be true, but actually, something else is happening here, which is most people just have no idea how good autonomous vehicles are.

(Joel Beasley at 00:54:18) That's what it is. So for me, I went down in December. My wife came out and I was like out in the garage and she goes, "Hey, did you hear that Full Self-Driving's out now?" Because I'd always talked about getting a Tesla or trying one because some of my business partners had them and I thought they were cool. And I was like, "For real?" She's like, "Yeah." She's like, "There's one in Franklin," which was like two hours from us. It's where we live now. She's like, "Do you wanna go look at it?" And I said, "Yeah, let's just go experience." Because, you know, I'm Modern CTO guy, right? I wanna sit in the car and let it fully self-drive me from A to B. That would be very cool and nerdy. So I did, and I walked out of the dealership with one that day because it was just — it was such a magical experience.

(Joel Beasley at 00:54:59) A white Model Y. Just a white Model Y. And it was, you know, $500 a month lease, and I was like, "Yeah, I'll take it just to, you know, experience." It's my favorite car. Yeah. I thought I would just have it. It would be, you know, one of the cars, and it's become my favorite car, hands down. I wish I could take it everywhere, you know.

(Zach Cass at 00:55:21) So what you're describing is this reminder that most people actually have no idea how good autonomous driving is because it's happened so fast. Outside of Pittsburgh and Phoenix, San Francisco, Los Angeles, there are just very few people who have ever been exposed to these things. And I experienced this in a very real way sort of orthogonally recently. I was at a medical conference espousing the promises of radiology, pathology in computer vision, talking about how we had made so much progress and it was getting so exciting. And when I was done talking about this, this doctor stood up, raised his hand, stood up in a huge room, and said, "Zach, I'm sorry to embarrass you." That's how he started the question. Said, "I'm sorry to embarrass you, but you're wrong." And actually, he took the time. He spent like a couple minutes sort of lambasting the technologist who had been talking about all the things that AI was gonna do, and none of us appreciated what the actual plight of the patient experience or the doctor was, and we should stop promising these panaceas. And he sat down never really having asked a question. And I had the sort of gall to respond and say, "Sir, I'm sorry to embarrass you, but you're wrong. The study that you have cited is actually a year old. And as of two months ago, radiology pathology is basically solved in all scans except the cardiovascular chest scan, which is hard for reasons we can't yet figure out. But limbs, abdomen, brain — GPT-4o is outperforming 95% of radiologists." And what I said was, "I don't blame you for not knowing. It is all happening too fast." And if a radiologist who, you know, has to — who's clearly at a conference, so he wants to learn more, who has to take regular boards, who has to stay up to date — doesn't know what a technology is capable of and how it is impacting his profession, what chance does the average person have to know? And the speed at which the technological threshold is moving is gonna serve as this sort of natural but very interesting rate limiter to the societal threshold. Now, I'll start by saying, Joel, before I tell you the third reason, this one is okay because I do think it gives us some appropriate pause. Like, it's not terrible to me that we don't immediately adopt everything. There are plenty of societal thresholds I argue, Joel, that should never move. And we can have that conversation another time. But the third reason that societal thresholds move slowly, as told by the autonomous vehicle — because of, you know, in the case of AI — is one that you can never unsee once you see it and is sort of, you know, funny. But humans have a remarkable tolerance for human failure, and we have no tolerance for machine failure.

(Joel Beasley at 00:58:10) And that's why I did this interview. When I read that line, I was like, we're doing this interview. Nice.

(Zach Cass at 00:58:17) It is this incredible thing that you cannot unsee, and it's why 1.3 million people can die on the global roadways every year, 95% by human error. And if a Waymo swerves into the wrong lane, people go, "How dare they? How dare they endanger us? Shut it all down."

(Joel Beasley at 00:58:36) Headlines for 72 hours. Yeah.

(Zach Cass at 00:58:38) Shut it all down. Like, drag them in front of Congress. Let's get to the bottom of these dystopian technocrats. Now I will say there are some features to this very strange anomaly — this very strange bias. There, I am using it. One of the features is that, like, our expectations for machines is so high that, like, things like flight work exceptionally well when it could be really, you know, questionable. We are like, "Listen, we're gonna inspect that plane every three months, and if there's any issue, it's coming out of the sky. You're gonna get fined." Buildings don't fall over for the same reason. We have exceptional code around the machines that we build that run our lives. And to such an extent that, like, any manufacturer of modern goods worth its weight offers like three-year warranties. I mean, like, people are like, "Oh, they don't make things like they used to." I'm like, "Yeah, they still make things pretty good." Right? I mean, like, there are a lot of laws that protect you from lemons and things that break that in a world where, like, they're manufacturing like a million of these, 2 million of these, and they still work really well. That's amazing to me. But it is fascinating to me that we are so willing to watch ourselves do incredible damage to each other as long as it's an accident and human error. And the marginal likelihood of a machine that could save, in our case, again, 1.3 million lives — the marginal chance of failure is just too great for us. And I think there are two reasons for this that we talk about in the book. One is humans have this association with mechanical failure as systemic failure, which is interesting. We believe that if one machine can fail, all machines will fail in the same way. And of course the opposite is true. If we can observe a machine fail, we can stop all machines from failing in the same way. And humans, as you well know, do not possess this feature. Right? Humans fail all day, every day in the same way over and over and over collectively and individually. And the other reason is that we have no empathy for machines, and I think this is fascinating, and I don't hate this one either. But we really do have empathy for each other. Right? Watching someone make a mistake, we always — you know, we all often imagine, or empathetic people, but many people will say, "Oh, it could have happened to anyone. I understand. You know, that's tough. Right? That's hard." Someone — you know, truck driver falls asleep at the wheel, and people are like, "God, we just asked too much of these truck drivers." Right? We rightfully have none for machines. We expect precision from machines, and we expect directionality from each other. And honestly, I think it creates a more human world. I prefer this to the alternative, but I also acknowledge that this is going to be one of these cases where so much good is going to be locked up behind a societal threshold because there is a marginal rate of failure that is just intolerable. And early adoption will happen on an individual basis, but it will take a long time to happen on the societal.

(Joel Beasley at 01:01:50) I agree. That's — I like the rate limiter part. It's — I always think about the human's ability to adopt things and how difficult it is for us to change behaviors as sort of like a safeguard for rapid advancement. Because right now, we're having breakthroughs, you know, let's say they were happening on a five-year basis. I feel like right now, they're happening on a quarterly basis.

(Joel Beasley at 01:02:15) You know, it's been speeding up in my professional career, and there's no reason for me to think that that wouldn't continue to stay in motion. And then in the next couple years, we're having breakthroughs on a daily basis that I can't even keep up with them. It's my full-time job. I get paid to do this, to try to keep up with this stuff, and I have a team. I get paid to do it, and I can't keep up. And then my mind is still blown on a daily basis. It's wild.

(Zach Cass at 01:02:39) Yeah, but I really do think that our societal thresholds will serve a critical role going forward. They will create an artificial rate limiter, and it will give us all a chance to sort of update to society in a more natural way. But it is gonna create some head-scratching moments where we go, "God, we would be a lot better off if we just, you know, if we allowed this or did this." And it's gonna be—the question, "Can a machine do something?" is just not gonna be that important in ten years. The question of "Do we want a machine? Do we let a machine do that thing?" is gonna be way more interesting.

(Joel Beasley at 01:03:18) Yeah. Well, luckily, we do have a biological limit in generations, like iPhones, right? Or the generations of humans will be more prepared to handle this. But we are—I'm in my late thirties right now. We are in the position where we're having an impact on this stuff for the next generations below us. So we have to at least think deeply about these things, do the best we can.

(Zach Cass at 01:03:42) Well, we're scratching the surface today, and I do think social media, like others, serves as a reminder that not all societal thresholds need to stay crossed, right? I think there are times where we can say, "Oh, whoops. Let's dial this back. Let's..." We get to live and learn, and we can acknowledge that we are still capable of making mistakes. And the humility to admit that is gonna allow us to actually undo a lot of wrong.

(Joel Beasley at 01:04:13) Let's get some rapid-fire stuff here as we start to wrap up. Are you cool with that?

(Zach Cass at 01:04:16) Fire.

(Joel Beasley at 01:04:17) Okay. You've got a book coming out in January 2026. Tell me about the book.

(Zach Cass at 01:04:22) Well, we talked about some of it today. It's the answer to my attempts to help people make sense of the future. And I think the world lacks a lot of nuance, especially when it comes to AI. And it's the best attempt I have to give people the good, the bad, and the ugly from honestly a biased lens of hope. I was born with an optimistic disposition. I studied history and found a lot of defense for it in computer science. And when you study those things, you learn that the world gets better all the time, and then you learn why. And the book is my attempt to help people imagine a future that is better, acknowledging all the things that can and may break.

(Joel Beasley at 01:05:06) And is it available for order today?

(Zach Cass at 01:05:08) It's available for preorder today. Thanks. You're leading the witness quite well. You can host the link, I'm sure, on your site. There—it's on Amazon. It's also on my website.

(Joel Beasley at 01:05:20) So people can search "Great Renaissance." Is that correct?

(Zach Cass at 01:05:24) The Next Renaissance: AI and Expansion of Human Potential. And I checked—it's doing pretty well on SEO and ChatGPT. So depending on your search preference, it will show up most places.

(Joel Beasley at 01:05:35) Yeah. And also, guys, I'd like to give a shout-out to Zach because, you know, when doing this interview and prepping for it, I got to watch some of the videos and stuff on your website and of you speaking and doing these conferences and this consulting stuff. I mean, you got a five-star review from Coca-Cola for consulting and helping them figure out the future with AI. So I believe you do that, right? You do consulting and you do speaking as well?

(Zach Cass at 01:06:00) Yeah. I have an advisory practice and a speaking practice. We have content—obviously, we have the book coming out. And then I have a company to be announced. I'm going back to operating, so I just finished a fundraiser for a company to be announced soon enough that I'm really excited about that aims to bring AI to everyone.

(Joel Beasley at 01:06:22) Oh, that's amazing. I—now I'm an interviewer, so I have seventy questions, but I'll...

(Zach Cass at 01:06:27) I mean, we can do a follow-up, and I'll—it brings to bear a lot of my experiences and is trying to help the Fortune 5,000, a lot of companies that I worry about getting left behind. There's a concentration of resources happening right now, and we're gonna try to help everyone else catch up.

(Joel Beasley at 01:06:46) That's interesting. Maybe we talk after the show. I've been having conversations in the past two weeks about how do we help companies in the five to fifteen million revenue range adopt this stuff, and a couple different people. So January 2026, the book comes out. And let's see. I also have—let's do one of these questions for advice for tech leaders. What should a CTO prioritize when integrating AI into their organization?

(Zach Cass at 01:07:15) It obviously depends on, you know, B2B, B2C. Are they building internally? Are they building externally? I try to remind companies: forward-compatible infrastructure. And forward-compatible infrastructure now is not just technological. There's organizational. There's cultural. Building a company that is amenable to change. The other thing I try to remind people is this idea of anchoring on your vision and mission and values. I think it's easy to be a candle in the wind when everything's changing. It's very easy for a CTO to pick up a new object just to satisfy the board with a shiny object. And I think it's critical now more than ever to say, "Listen, this is what we're trying to solve. These are the things that we alone can build, and these are the things that we must purchase." And acknowledging that that list may change, but our pursuits, our panacea, our utopia will not. And if you're amenable to the ways and means, if you're very open to how you accomplish something—which I've discovered a lot of CTOs are sort of religious in their practices, actually—be religious in your vision. Be religious in your mission and purpose and be very amenable to how you accomplish it. You will have a very good time.

(Joel Beasley at 01:08:26) Yeah. I love that. No, this is good. I like the way you think. It's zakkass.com. Is that correct?

(Zach Cass at 01:08:35) Zackkass.com.

(Joel Beasley at 01:08:38) All right. We got it. And we'll put links in the notes below. Thank you so much for doing this. We made a podcast. How do you feel?

(Zach Cass at 01:08:44) Great, Joel. I had a ton of fun. And as you said, "Are we—did we just become best friends?"

(Joel Beasley at 01:08:49) I think so.

(Zach Cass at 01:08:49) Yeah. There you go. Insert Stepbrothers.

(Joel Beasley at 01:08:54) Thank you so much for listening. And if you found this episode useful, please share it with a friend or colleague who you think would get value from it. And if you have topics that you would like to hear discussed on the podcast, either add me on LinkedIn or send me an email: [email protected]. Every time I get an email or LinkedIn message, it absolutely makes my day and inspires me to keep going.