All right. What I'm going to do today is we're going to have a debate between myself and Marcus Hutchins. If you don't know Marcus, he is probably the most famous malware researcher in the world. He's been on the front page of wired, and he has a dramatically different view of AI than I do. So I am in sort of the pro AI camp and believing that it is net positive and that it is extremely useful today, and also getting more useful and most importantly, that it's going to start affecting jobs in a very serious way. Marcus has called it basically, autocomplete does not believe that it's intelligence, and believes that the impact is dramatically overplayed, mostly by people trying to scam people and make money. So again, we are on pretty much opposite sides of this, but we are also both security people and, you know, fairly well known in the security space, and we've both done a lot and accomplished a lot. And this is really showing that you can have two people who are in kind of the same space, can have dramatically different opinions. And what turned out to be fantastic here, because Marcus and I have been friends for some amount of time now, and we had this debate previously in text and it was fine in text. I was a little worried that we're both a bit snarky online, and I was worried that we were going to be snarky to each other in this debate that turned out not to happen. It was a very positive experience, and I think we made our points. Well, not only that, but I think I made a number of points that he really enjoyed. He made a number of points that I really enjoyed. I feel like we both learned something from this, and we've already committed to doing a follow up at some point in the future, but this one is hopefully going to be useful to anybody who is kind of agnostic or really just kind of wants to hear strong points being made on both sides of the conversation and also in a civil way. So that's what we're going to do here. I do apologize a little bit for the audio quality. We actually did this on zoom, and we should have done it on Riverside because you can't actually do 4K in zoom yet. So anyway, the next one is going to be higher quality. But um, let's jump into it. This is a debate between myself and Marcus Hutchins on AI. I think it's wrong to call what is happening now with AI hype and to tell people not to worry. Um, because I think it is a significant threat to people's jobs and, um, overall, just like a massive thing, uh, in, in much more of a way than crypto was, I think crypto was a giant. Nothing. Um, I wasn't excited about crypto at the time. I didn't see why why it was like everyone was taking it so seriously. But I see this as very different. Um, and so what? I see people like yourself or someone like another person. I have a lot of respect for, uh, Christopher Hoff, who I've been friends with for a very long time, actually, for him, like 15 years or something. 20 years. And he's frequently putting out similar stuff to you, whereas like, look, what are you guys talking about? This thing is like, not anything. It's going to be just like crypto. It's going to be gone soon. It's massively overhyped. So, um, when I see you posting those things, I haven't responded to you and I'm sure you're, like, itching to respond to me as well. Uh, we haven't done that because we planned on doing this. And this is a better, I think, a better way to start that off. But, um, I would respond to Chris Hoff and he would respond back and it would get kind of like adversarial, snarky or whatever. Um, but I mean, all that to say that I, I think that position is steering people in a, in a bad direction. It's telling people not to worry. And I'm not saying like panic. I'm saying this thing is real and we have to get ready for it. I think it's wrong to call what is happening now with AI hype and to tell people not to worry. Um, because I think it is.

All right. What I'm going to do today is we're going to have a debate between myself and Marcus Hutchins. If you don't know Marcus, he is probably the most famous malware researcher in the world. He's been on the front page of wired, and he has a dramatically different view of AI than I do. So I am in sort of the pro AI camp and believing that it is net positive and that it is extremely useful today, and also getting more useful and most importantly, that it's going to start affecting jobs in a very serious way. Marcus has called it basically, autocomplete does not believe that it's intelligence, and believes that the impact is dramatically overplayed, mostly by people trying to scam people and make money. So again, we are on pretty much opposite sides of this, but we are also both security people and, you know, fairly well known in the security space, and we've both done a lot and accomplished a lot. And this is really showing that you can have two people who are in kind of the same space, can have dramatically different opinions. And what turned out to be fantastic here, because Marcus and I have been friends for some amount of time now, and we had this debate previously in text and it was fine in text. I was a little worried that we're both a bit snarky online, and I was worried that we were going to be snarky to each other in this debate that turned out not to happen. It was a very positive experience, and I think we made our points. Well, not only that, but I think I made a number of points that he really enjoyed. He made a number of points that I really enjoyed. I feel like we both learned something from this, and we've already committed to doing a follow up at some point in the future, but this one is hopefully going to be useful to anybody who is kind of agnostic or really just kind of wants to hear strong points being made on both sides of the conversation and also in a civil way. So that's what we're going to do here. I do apologize a little bit for the audio quality. We actually did this on zoom, and we should have done it on Riverside because you can't actually do 4K in zoom yet. So anyway, the next one is going to be higher quality. But um, let's jump into it. This is a debate between myself and Marcus Hutchins on AI. I think it's wrong to call what is happening now with AI hype and to tell people not to worry. Um, because I think it is a significant threat to people's jobs and, um, overall, just like a massive thing, uh, in, in much more of a way than crypto was, I think crypto was a giant. Nothing. Um, I wasn't excited about crypto at the time. I didn't see why why it was like everyone was taking it so seriously. But I see this as very different. Um, and so what? I see people like yourself or someone like another person. I have a lot of respect for, uh, Christopher Hoff, who I've been friends with for a very long time, actually, for him, like 15 years or something. 20 years. And he's frequently putting out similar stuff to you, whereas like, look, what are you guys talking about? This thing is like, not anything. It's going to be just like crypto. It's going to be gone soon. It's massively overhyped. So, um, when I see you posting those things, I haven't responded to you and I'm sure you're, like, itching to respond to me as well. Uh, we haven't done that because we planned on doing this. And this is a better, I think, a better way to start that off. But, um, I would respond to Chris Hoff and he would respond back and it would get kind of like adversarial, snarky or whatever. Um, but I mean, all that to say that I, I think that position is steering people in a, in a bad direction. It's telling people not to worry. And I'm not saying like panic. I'm saying this thing is real and we have to get ready for it. I think it's wrong to call what is happening now with AI hype and to tell people not to worry. Um, because I think it is.

But when you say it's staring people in the direction of, uh, I guess the, for a lack of a better term is harmful. What do you think the harm is there? So if I say now, this isn't my personal position, but let's say I say AI is completely useless. Like beyond useless. It does nothing of value whatsoever. What is inherently bad about people believing that.

But when you say it's staring people in the direction of, uh, I guess the, for a lack of a better term is harmful. What do you think the harm is there? So if I say now, this isn't my personal position, but let's say I say AI is completely useless. Like beyond useless. It does nothing of value whatsoever. What is inherently bad about people believing that.

Uh, they will not, uh, think about their own skills and their own capabilities and their own value and how that is potentially going to be impacted by AI if they don't see AI as a threat because it's a nothing burger, and it's basically all hype and marketing and a bunch of rich people trying to get rich, uh, even more rich, then they will ignore it. So to answer your question. The threat, the risk of that is them ignoring something that is potentially harmful to their career.

Uh, they will not, uh, think about their own skills and their own capabilities and their own value and how that is potentially going to be impacted by AI if they don't see AI as a threat because it's a nothing burger, and it's basically all hype and marketing and a bunch of rich people trying to get rich, uh, even more rich, then they will ignore it. So to answer your question. The threat, the risk of that is them ignoring something that is potentially harmful to their career.

But I think you could make that argument for, uh, we use the the example of blockchain or crypto specifically. You could have made that same argument for crypto. We could have said, uh, this is going to be this future big, impactful technology, which is something many people did say. They were like, everything is going to be blockchain, crypto. There were people arguing that Doge was going to replace the US currency. And you could have made the argument back then that if you don't go out right now and spend hours and hours of your time learning blockchain, you're going to be unemployed. You're not going to be able to, uh, make it in the future. So we're sort of trying to come up with this reason for why AI is different from blockchain. Now, I'm not saying it's similar. I, I think most people will agree large language models have infinitely more usefulness than blockchain did. But then it still comes down to what are we thinking that where do we think that this technology is going that is going to have such a big impact that people need to be paying as much attention as possible? And like, what even is the level of attention people need to pay? Like, do I just need to know, okay, ChatGPT is a thing and here's what it can do. Or do I need to spend hours and hours of my day learning how to use ChatGPT in order to avert whatever problem we're going to say is going to be the thing in the future.

But I think you could make that argument for, uh, we use the the example of blockchain or crypto specifically. You could have made that same argument for crypto. We could have said, uh, this is going to be this future big, impactful technology, which is something many people did say. They were like, everything is going to be blockchain, crypto. There were people arguing that Doge was going to replace the US currency. And you could have made the argument back then that if you don't go out right now and spend hours and hours of your time learning blockchain, you're going to be unemployed. You're not going to be able to, uh, make it in the future. So we're sort of trying to come up with this reason for why AI is different from blockchain. Now, I'm not saying it's similar. I, I think most people will agree large language models have infinitely more usefulness than blockchain did. But then it still comes down to what are we thinking that where do we think that this technology is going that is going to have such a big impact that people need to be paying as much attention as possible? And like, what even is the level of attention people need to pay? Like, do I just need to know, okay, ChatGPT is a thing and here's what it can do. Or do I need to spend hours and hours of my day learning how to use ChatGPT in order to avert whatever problem we're going to say is going to be the thing in the future.

Yeah, sure. I mean. Yeah. Well, first of all, I mean, ChatGPT is just one app from one group, but I think I think I see what you mean, just AI in general. Um, no, I mean, what I advocate is people focus more on first principles thinking and like, understanding their own opinions and forming their own opinions. Uh, because what I'm trying to get to is like, what comes after the job replacement, which is like more human to human interaction. So I think people need to be more, um, thoughtful internally about who they are and what they are and like, what they're about actually having opinions. I think too many people who are in danger of being replaced by this technology, um, simply had a job executing tasks that someone else gave them, and that didn't require that much intellect, but it required enough intelligence. And this is another point we can debate. Like, is AI actually intelligence? Um, but it has enough intelligence for all these hundreds of millions of knowledge worker jobs. It has enough intelligence required that you couldn't automate it. So which is why they have the job today, and that if the bar is getting to your question, if the bar for AI raises above that level. So we could do basic intelligence tasks, which is, you know, 90% or 99% of knowledge worker, uh, Capabilities is included in that. Then those jobs go away. Hundreds of millions. I don't know the actual numbers for knowledge worker total employment, but I would say that vast, vast majority of those jobs go away and they are left not doing anything. And it's not like they could just pivot to the next thing, like previous technology trends, because, um, those, those pivots that they will make likely are also taken over by that.

Yeah, sure. I mean. Yeah. Well, first of all, I mean, ChatGPT is just one app from one group, but I think I think I see what you mean, just AI in general. Um, no, I mean, what I advocate is people focus more on first principles thinking and like, understanding their own opinions and forming their own opinions. Uh, because what I'm trying to get to is like, what comes after the job replacement, which is like more human to human interaction. So I think people need to be more, um, thoughtful internally about who they are and what they are and like, what they're about actually having opinions. I think too many people who are in danger of being replaced by this technology, um, simply had a job executing tasks that someone else gave them, and that didn't require that much intellect, but it required enough intelligence. And this is another point we can debate. Like, is AI actually intelligence? Um, but it has enough intelligence for all these hundreds of millions of knowledge worker jobs. It has enough intelligence required that you couldn't automate it. So which is why they have the job today, and that if the bar is getting to your question, if the bar for AI raises above that level. So we could do basic intelligence tasks, which is, you know, 90% or 99% of knowledge worker, uh, Capabilities is included in that. Then those jobs go away. Hundreds of millions. I don't know the actual numbers for knowledge worker total employment, but I would say that vast, vast majority of those jobs go away and they are left not doing anything. And it's not like they could just pivot to the next thing, like previous technology trends, because, um, those, those pivots that they will make likely are also taken over by that.

I, I understand that line of thinking of, okay, it's not like the printing press or, uh, early machine learning where I can just go into something that is not covered by the current models. But I think the idea of this just sweeping replacement, where llms are just going to take out 90% of the workforce, I don't personally believe that that is rooted in reality. I think, uh, we're basically extrapolating with, uh, it's almost like a new car, right? If I, If I've just invented the car, it only goes 15mph. And then if I do some tweaks to it, I spend like a couple of months doing some tweaks. It goes a little bit faster. And now we could say that if I doubled the speed in three months than in six months, I can double the speed again. And then you can just keep extrapolating, extrapolating, extrapolating. So at some point this car will break the speed of light. Um, whereas I think that's what we've been seeing with Llms is the LLM technology was very new and it had a lot of issues in the beginning. And that gave the illusion of a lot of very fast progress. We weren't actually seeing these models rapidly getting like better and better and better. All that was happening is we made a new technology that was a little bit, uh, like finicky. It was it had problems. And we slowly ironed out some of those problems, which made it look like it was advancing at this rapid pace. But it seems now that it is sort of capped out. Um, I would say the leap from GPT two to GPT three and 3.5 was pretty groundbreaking. But then as I look at the newer models, when we go from 3.5 to whatever they've called it now, like 001, I don't even know what the naming convention is anymore. Um, we haven't really seen any improvements. We're just seeing very slight increases in, oh, it's better at coding now. It's a little bit better at math. And what it looks like is happening is they've made the base large language model, and they've refined that almost as much as they can. And then they're building these sub models to then make up the slack where it can't do math or it can't do programming or whatever other areas it struggles. So I think what we're actually going to see is we're going to see large language models as a technology just cap out like they're just going to hit a wall. And then what we're going to see is them start building out sub models that make them better at a specific job. Now that would just be normal technological evolution like I as a malware researcher, I got into malware analysis when it was already machine learning automated. There was still enough space in my industry where I didn't need to become an expert in machine learning. I didn't even need to touch those models. I still to this day, have no idea what the virus classification machine learning models do, because there was still enough space for me to operate in that field as a human, and I don't see large language models as any different from that. I think what we're going to just see is people making sub models that make, say, malware analysis better. And it's like, okay, maybe I can't do manual virus classification anymore, but I can do this other thing and I think it's going to move slowly enough as they do that, that people can just pivot into not even different fields, but into different, different like roles in the same field where a lot of their skills are transferable. But rather than doing something that maybe is very pattern recognition and is very, uh, lm Them automatable they could move into something that takes the same skills but actually requires more critical thinking.

I, I understand that line of thinking of, okay, it's not like the printing press or, uh, early machine learning where I can just go into something that is not covered by the current models. But I think the idea of this just sweeping replacement, where llms are just going to take out 90% of the workforce, I don't personally believe that that is rooted in reality. I think, uh, we're basically extrapolating with, uh, it's almost like a new car, right? If I, If I've just invented the car, it only goes 15mph. And then if I do some tweaks to it, I spend like a couple of months doing some tweaks. It goes a little bit faster. And now we could say that if I doubled the speed in three months than in six months, I can double the speed again. And then you can just keep extrapolating, extrapolating, extrapolating. So at some point this car will break the speed of light. Um, whereas I think that's what we've been seeing with Llms is the LLM technology was very new and it had a lot of issues in the beginning. And that gave the illusion of a lot of very fast progress. We weren't actually seeing these models rapidly getting like better and better and better. All that was happening is we made a new technology that was a little bit, uh, like finicky. It was it had problems. And we slowly ironed out some of those problems, which made it look like it was advancing at this rapid pace. But it seems now that it is sort of capped out. Um, I would say the leap from GPT two to GPT three and 3.5 was pretty groundbreaking. But then as I look at the newer models, when we go from 3.5 to whatever they've called it now, like 001, I don't even know what the naming convention is anymore. Um, we haven't really seen any improvements. We're just seeing very slight increases in, oh, it's better at coding now. It's a little bit better at math. And what it looks like is happening is they've made the base large language model, and they've refined that almost as much as they can. And then they're building these sub models to then make up the slack where it can't do math or it can't do programming or whatever other areas it struggles. So I think what we're actually going to see is we're going to see large language models as a technology just cap out like they're just going to hit a wall. And then what we're going to see is them start building out sub models that make them better at a specific job. Now that would just be normal technological evolution like I as a malware researcher, I got into malware analysis when it was already machine learning automated. There was still enough space in my industry where I didn't need to become an expert in machine learning. I didn't even need to touch those models. I still to this day, have no idea what the virus classification machine learning models do, because there was still enough space for me to operate in that field as a human, and I don't see large language models as any different from that. I think what we're going to just see is people making sub models that make, say, malware analysis better. And it's like, okay, maybe I can't do manual virus classification anymore, but I can do this other thing and I think it's going to move slowly enough as they do that, that people can just pivot into not even different fields, but into different, different like roles in the same field where a lot of their skills are transferable. But rather than doing something that maybe is very pattern recognition and is very, uh, lm Them automatable they could move into something that takes the same skills but actually requires more critical thinking.

Yeah, yeah. I mean, I, I definitely agree there's going to be pivots. Um, still possible. Unfortunately, I think those pivots are going to be mostly available to the really sharp, exceptional people who can pivot up to something much higher up in the stack, uh, which requires a lot more critical thinking that the AI can't do yet. Um, you had the analogy there of speed and approaching the speed of light, uh, a car approaching the speed of light. So here, here's what I think might be a fundamental difference, uh, in our opinions here, um, the speed of light that you're talking about, I agree it would be hard to say. Went from 30mph to 35mph. And next year we're going to be at the speed of light. That's a crazy statement. The difference is the speed of light that we're trying to get here, um, that I'm talking about is actually approaching. Cannot do the work of a knowledge worker. An average knowledge worker on earth. Okay. And so my argument to you is that that bar is not high at all. So it is basically and this is something we talked about in text. Um, right in the previous version of this, this discussion, it's like the average knowledge worker. They, they clock in and clock out. And I'm talking about mean, you know, like, uh, average just average across the globe. It's like, hey, got some new emails, respond to the email. Uh, the email says, hey, go, um, make a list of all the documents that are related to this project. Okay, cool. Here's your list. They send out the list. Um, someone says, hey, I need you to write a report on how our finances are doing. So what do they do? Is they pull a wiki, uh, article, they review this other document, they talk to Sarah in this other department that takes them 4 hours or 5 hours or two days or however long it takes, and they write a four page report, and then they take that report and they put it in an email and they send it out to the company. They are doing work. They're being paid 50 grand, 100 grand for this work. And I'm claiming to you that this is a low bar, not for I in the future, but for I, we already had a year ago, in some cases two years ago. So my concern here is not about the technical definition of I. I don't give two shits about it. People are going to argue about, oh, AGI means this technically, specifically, I don't care. All I care about is impact to humans. So I'm claiming that the AI that we have a year ago already could do if it were properly orchestrated. A lot of the jobs that I just described. So customer service, um, responding to emails, setting up like a lunch meeting, just just basic tasks like that that could not be automated with regular tech, but can be automated with with this tech.

Yeah, yeah. I mean, I, I definitely agree there's going to be pivots. Um, still possible. Unfortunately, I think those pivots are going to be mostly available to the really sharp, exceptional people who can pivot up to something much higher up in the stack, uh, which requires a lot more critical thinking that the AI can't do yet. Um, you had the analogy there of speed and approaching the speed of light, uh, a car approaching the speed of light. So here, here's what I think might be a fundamental difference, uh, in our opinions here, um, the speed of light that you're talking about, I agree it would be hard to say. Went from 30mph to 35mph. And next year we're going to be at the speed of light. That's a crazy statement. The difference is the speed of light that we're trying to get here, um, that I'm talking about is actually approaching. Cannot do the work of a knowledge worker. An average knowledge worker on earth. Okay. And so my argument to you is that that bar is not high at all. So it is basically and this is something we talked about in text. Um, right in the previous version of this, this discussion, it's like the average knowledge worker. They, they clock in and clock out. And I'm talking about mean, you know, like, uh, average just average across the globe. It's like, hey, got some new emails, respond to the email. Uh, the email says, hey, go, um, make a list of all the documents that are related to this project. Okay, cool. Here's your list. They send out the list. Um, someone says, hey, I need you to write a report on how our finances are doing. So what do they do? Is they pull a wiki, uh, article, they review this other document, they talk to Sarah in this other department that takes them 4 hours or 5 hours or two days or however long it takes, and they write a four page report, and then they take that report and they put it in an email and they send it out to the company. They are doing work. They're being paid 50 grand, 100 grand for this work. And I'm claiming to you that this is a low bar, not for I in the future, but for I, we already had a year ago, in some cases two years ago. So my concern here is not about the technical definition of I. I don't give two shits about it. People are going to argue about, oh, AGI means this technically, specifically, I don't care. All I care about is impact to humans. So I'm claiming that the AI that we have a year ago already could do if it were properly orchestrated. A lot of the jobs that I just described. So customer service, um, responding to emails, setting up like a lunch meeting, just just basic tasks like that that could not be automated with regular tech, but can be automated with with this tech.

Uh. Well, you see, I kind of disagree with that because I've unfortunately had to interface with LM based customer service agents. And they do have the the flaws that both of us are familiar with with LMS. They are not deterministic. If I give it the same question 50 times, it does not have the same response 50 times. Like depending on, uh, I don't know, the, the star alignment, how the electricity is going that day, who knows. It comes out with a different response to each, um, to each input. And that is fundamentally a problem because if you need a machine to do a deterministic task like finance reports or customer service, you cannot have a model that comes out with completely different answers every single time you ask the same question. And that is one of the major limitations that has been, uh, currently holding back large language models is that by their very nature they are not deterministic. So all of those tasks you are talking about are tasks which, uh, when you say knowledge task, you, you're kind of, I think, sort of talking about tasks that don't require critical thinking. They don't really require much intelligence. They're just doing like tasks that will be the same every time you do them. And if the large language model cannot even do tasks the same way multiple times, how can we then go and automate all of those jobs?

Uh. Well, you see, I kind of disagree with that because I've unfortunately had to interface with LM based customer service agents. And they do have the the flaws that both of us are familiar with with LMS. They are not deterministic. If I give it the same question 50 times, it does not have the same response 50 times. Like depending on, uh, I don't know, the, the star alignment, how the electricity is going that day, who knows. It comes out with a different response to each, um, to each input. And that is fundamentally a problem because if you need a machine to do a deterministic task like finance reports or customer service, you cannot have a model that comes out with completely different answers every single time you ask the same question. And that is one of the major limitations that has been, uh, currently holding back large language models is that by their very nature they are not deterministic. So all of those tasks you are talking about are tasks which, uh, when you say knowledge task, you, you're kind of, I think, sort of talking about tasks that don't require critical thinking. They don't really require much intelligence. They're just doing like tasks that will be the same every time you do them. And if the large language model cannot even do tasks the same way multiple times, how can we then go and automate all of those jobs?

Yeah, I think I think it's fairly easy. I mean, I've been doing this for a long time. Everyone I know who's building apps here, it's fairly easy to actually give somebody, uh, give an AI a list and say, you can only pick from these, like, this is the thing that was solved, you know, two and a half years ago, probably before that. Um, but you can make a system fairly deterministic. Um, and keep in mind, like customer service is already horrible. Everyone already hated customer service because it was so bad. Before I came out. Right. So it's not like customer service is the high bar, and we're seeing if we can get there. No, it's a horrible bar. Like, everyone hates it already. And oftentimes it has the exact same characteristics you were talking about, where it's like you're not getting good answers. The good answer is actually somewhere there in the knowledge base. But somehow this tier one or tier two person you're talking to, they didn't find the answer. Um, so a couple of, you know, objections to that. I don't think, um, current execution of a lot of these jobs is at a high level. And that's why I think it's possible to beat it. The other thing is, um, you're not actually when your boss says, hey, go make a write a report for this thing that is fundamentally a non-deterministic thing because the problem is always different that the human is given. So it definitely does require intelligence, right? You can't give that to a script because the script doesn't have like an if, then it doesn't have a deterministic thing to go look up the thing because the problem is always different. And this this is another thing that you and I talked about before, which I thought was super interesting, is like, what actually is intelligence and how much deviation from a standard slotted thing does it require to be called intelligence. So I would argue if someone is thrown a giant batch of emails and a giant batch of documents and they're like, hey, Chris. Hey, Sarah, um, go write a report on this. And I need it by today at 4:00. No one person or two people or no combination of humans is going to make the same report. They're going to make vastly different reports based on how they choose to go through the documents, based on how they choose to form the thing. How long is the report going to be? If you ask a thousand people to make a report, they're all going to be different, just like with an LLM. So and the other thing is just because that doesn't require like PhD level invention or creativity that still requires human intelligence. No, no tech prior to AI could do that job.

Yeah, I think I think it's fairly easy. I mean, I've been doing this for a long time. Everyone I know who's building apps here, it's fairly easy to actually give somebody, uh, give an AI a list and say, you can only pick from these, like, this is the thing that was solved, you know, two and a half years ago, probably before that. Um, but you can make a system fairly deterministic. Um, and keep in mind, like customer service is already horrible. Everyone already hated customer service because it was so bad. Before I came out. Right. So it's not like customer service is the high bar, and we're seeing if we can get there. No, it's a horrible bar. Like, everyone hates it already. And oftentimes it has the exact same characteristics you were talking about, where it's like you're not getting good answers. The good answer is actually somewhere there in the knowledge base. But somehow this tier one or tier two person you're talking to, they didn't find the answer. Um, so a couple of, you know, objections to that. I don't think, um, current execution of a lot of these jobs is at a high level. And that's why I think it's possible to beat it. The other thing is, um, you're not actually when your boss says, hey, go make a write a report for this thing that is fundamentally a non-deterministic thing because the problem is always different that the human is given. So it definitely does require intelligence, right? You can't give that to a script because the script doesn't have like an if, then it doesn't have a deterministic thing to go look up the thing because the problem is always different. And this this is another thing that you and I talked about before, which I thought was super interesting, is like, what actually is intelligence and how much deviation from a standard slotted thing does it require to be called intelligence. So I would argue if someone is thrown a giant batch of emails and a giant batch of documents and they're like, hey, Chris. Hey, Sarah, um, go write a report on this. And I need it by today at 4:00. No one person or two people or no combination of humans is going to make the same report. They're going to make vastly different reports based on how they choose to go through the documents, based on how they choose to form the thing. How long is the report going to be? If you ask a thousand people to make a report, they're all going to be different, just like with an LLM. So and the other thing is just because that doesn't require like PhD level invention or creativity that still requires human intelligence. No, no tech prior to AI could do that job.

So I'd actually would like to go back to the customer service example, because I think I would argue customer service is horrible for the exact same reason that the solution you proposed would be horrible. They are given these limited scripts of answers where if customer asks this, then you must answer that and they cannot go beyond their script. So if I come up with a unique problem that requires some kind of critical thinking, although the human on the phone has intelligence and is capable of critical thinking, they are not allowed to go outside of my script. And essentially you would run into the same problem with trying to do that with a large language model is you've just given it a list of canned answers it's not allowed to go outside of. And you're going to get the exact same results where there is no critical thinking. We can get the argument that I don't think I can think like full stop. But you're essentially just recreating the same awful system where it's just it's just parroting back canned responses. And I'm trying to ask my question, and it doesn't match up to some canned response that they have. So they're just repeating the same closest line that they have in their script. And then we both leave the interaction completely annoyed. And, um, I think that's exactly what is going to happen with large language models. And it's already happening. They're just feeding their dogs or their scripts into the LLM, and it's just making a worse version of a customer service representative, because not only does it still have the canned responses, but it doesn't have the critical thinking to go, oh, maybe I should escalate this to my boss.

So I'd actually would like to go back to the customer service example, because I think I would argue customer service is horrible for the exact same reason that the solution you proposed would be horrible. They are given these limited scripts of answers where if customer asks this, then you must answer that and they cannot go beyond their script. So if I come up with a unique problem that requires some kind of critical thinking, although the human on the phone has intelligence and is capable of critical thinking, they are not allowed to go outside of my script. And essentially you would run into the same problem with trying to do that with a large language model is you've just given it a list of canned answers it's not allowed to go outside of. And you're going to get the exact same results where there is no critical thinking. We can get the argument that I don't think I can think like full stop. But you're essentially just recreating the same awful system where it's just it's just parroting back canned responses. And I'm trying to ask my question, and it doesn't match up to some canned response that they have. So they're just repeating the same closest line that they have in their script. And then we both leave the interaction completely annoyed. And, um, I think that's exactly what is going to happen with large language models. And it's already happening. They're just feeding their dogs or their scripts into the LLM, and it's just making a worse version of a customer service representative, because not only does it still have the canned responses, but it doesn't have the critical thinking to go, oh, maybe I should escalate this to my boss.

Yeah, no, it's an interesting point. I, I like that argument. Um, I would say the best argument I have against that is that, um, companies have been trying this customer service thing for decades, Like, uh, human customer service with the different levels. And they have tried like millions of different things. They've tried to give people autonomy. They've tried to not give people autonomy. The system they came up with is, look, follow this script, use this document. So I would argue to you that the reason that we see this in the industry is because it's the thing that works best. If what worked best was to just hire people who are somewhat smart and give them free rein, then that's what companies would be doing. Uh, and I don't think they're doing that because it doesn't work. Now, I would argue that as you move into tier three, tier four or whatever, whatever the structure is for these, um, like more senior customer service people, I feel like that thing you're describing is exactly what's happening. They are given free rein because all the known answers have already been tried, and now they're, um, now now they're going to have to use critical thinking.

Yeah, no, it's an interesting point. I, I like that argument. Um, I would say the best argument I have against that is that, um, companies have been trying this customer service thing for decades, Like, uh, human customer service with the different levels. And they have tried like millions of different things. They've tried to give people autonomy. They've tried to not give people autonomy. The system they came up with is, look, follow this script, use this document. So I would argue to you that the reason that we see this in the industry is because it's the thing that works best. If what worked best was to just hire people who are somewhat smart and give them free rein, then that's what companies would be doing. Uh, and I don't think they're doing that because it doesn't work. Now, I would argue that as you move into tier three, tier four or whatever, whatever the structure is for these, um, like more senior customer service people, I feel like that thing you're describing is exactly what's happening. They are given free rein because all the known answers have already been tried, and now they're, um, now now they're going to have to use critical thinking.

But I think a lot more jobs than you maybe estimate are they fall into that category of like, sure, a lot of it is automatable, but the reason they need a human in that role and not a script or a machine or whatever we want to call it, is because there is some aspect that requires critical thinking, um, uh, like the one I will push back against endlessly until my last breath is software engineers. I think there is no point in any of the AI's evolution, at which point you will be able to replace software engineers. And I think the disconnect that happens is a lot of the non software engineers or even the coders, the people who just don't fully understand software engineering don't understand the difference between writing code and designing software. So they're looking at ChatGPT and Cursor and Gemini or whatever, and they're looking, oh, I can put in a query and it spits out all of this code. And they're like, that is what coding is. So like that very easily could replace software engineering. But because they're not a software engineer, they don't understand the amount of critical thinking and design decisions that actually go into the software. I think, uh, a lot of software engineers will say that writing code is like 10% of software engineering. But then if you're not a software engineer, you're looking at that, that a small percentage that the AI, okay, sort of can do that and then ignoring the rest. And I think the same is true for every role. Um, until you actually have like spent a year in any given job, you don't really understand how much is automatable versus how much actually does require a human with critical thinking. So my personal belief is, I think for the large, like the overwhelming majority of all roles, there will never, ever be a point in which AI replaces that person. It might be able to make the person faster. You might be able to take one person and have them do work at a slightly more productive rate, but I don't think there is any point in which I actually starts replacing like, like large swaths of the job market.

But I think a lot more jobs than you maybe estimate are they fall into that category of like, sure, a lot of it is automatable, but the reason they need a human in that role and not a script or a machine or whatever we want to call it, is because there is some aspect that requires critical thinking, um, uh, like the one I will push back against endlessly until my last breath is software engineers. I think there is no point in any of the AI's evolution, at which point you will be able to replace software engineers. And I think the disconnect that happens is a lot of the non software engineers or even the coders, the people who just don't fully understand software engineering don't understand the difference between writing code and designing software. So they're looking at ChatGPT and Cursor and Gemini or whatever, and they're looking, oh, I can put in a query and it spits out all of this code. And they're like, that is what coding is. So like that very easily could replace software engineering. But because they're not a software engineer, they don't understand the amount of critical thinking and design decisions that actually go into the software. I think, uh, a lot of software engineers will say that writing code is like 10% of software engineering. But then if you're not a software engineer, you're looking at that, that a small percentage that the AI, okay, sort of can do that and then ignoring the rest. And I think the same is true for every role. Um, until you actually have like spent a year in any given job, you don't really understand how much is automatable versus how much actually does require a human with critical thinking. So my personal belief is, I think for the large, like the overwhelming majority of all roles, there will never, ever be a point in which AI replaces that person. It might be able to make the person faster. You might be able to take one person and have them do work at a slightly more productive rate, but I don't think there is any point in which I actually starts replacing like, like large swaths of the job market.

Yeah. Interesting. Um, so let me give you a counterpoint. And this this is anecdotal, of course, but I have a close friend who is a cardiologist, like actual MD actually sits in the, you know, the doctor's office every day. He also happens to be a bug bounty hunter and like, a programmer. So he he is, uh, you know, using whatever tools are approved at his particular job. And, um, he is now finding as of like this is this is six months ago. This is a year ago. Um, the analysis that this, this thing is doing is as good or better in some cases than the thing that he was going to write. So this is at the cardiologist level. Not not even like a nursing level. This is really, really advanced stuff. So it's it's taking all the inputs that came from like the patient and also from his notes. It's combining all that together with, of course, the knowledge of the model itself. And then, you know, giving out recommendations, recommending the actual drugs. And so my thing is like if this thing is actually doing the job and it's not just one person, right? A lot of doctors are actually saying that the output is matching their capabilities as well. We've already seen this with analyzing moles. We've seen this like in lots of different places. Um, and that brings me back to this one other point that you made earlier, which I thought was really interesting, um, how they're going to hit a wall. The Elon's hit a wall and they basically start building subsystems. This is exactly how I define AGI. I don't think AGI is going to be some breakthrough, generally intelligent thing. I think that is more of an AI breakthrough than we've actually seen. I don't think it's happened yet. What I'm arguing is going to replace the average knowledge worker is actually one of these, uh, Frankenstein systems where the LLM is like an orchestrator, and it's basically spinning up these individual things to go and accomplish individual tasks like write the email, organize the events, organize the conference, uh, summarize this or whatever. So it's going to have at its disposal dozens or hundreds of these little workers. But the problem is that you're going to buy this thing, let's call it a synth worker or whatever the stupid AI company is going to be called. You buy, you buy synth worker, synth worker shows up to work on Monday. It goes to the onboarding meeting. It talks to the manager. It talks to all its team members. It reads slack, it reads the wiki. And it starts working just like a regular employee. And it produces the work of a regular employee. And I would argue in a lot of cases, even better. Um, but if you look under the hood, it's not some all intelligent AI system. It's actually a whole bunch of tricks. It's like this big agent orchestrating a whole bunch of other ones. But functionally, what you end up with is this new AI named Julie actually does the work. And guess what? We didn't hire a human to do that job.

Yeah. Interesting. Um, so let me give you a counterpoint. And this this is anecdotal, of course, but I have a close friend who is a cardiologist, like actual MD actually sits in the, you know, the doctor's office every day. He also happens to be a bug bounty hunter and like, a programmer. So he he is, uh, you know, using whatever tools are approved at his particular job. And, um, he is now finding as of like this is this is six months ago. This is a year ago. Um, the analysis that this, this thing is doing is as good or better in some cases than the thing that he was going to write. So this is at the cardiologist level. Not not even like a nursing level. This is really, really advanced stuff. So it's it's taking all the inputs that came from like the patient and also from his notes. It's combining all that together with, of course, the knowledge of the model itself. And then, you know, giving out recommendations, recommending the actual drugs. And so my thing is like if this thing is actually doing the job and it's not just one person, right? A lot of doctors are actually saying that the output is matching their capabilities as well. We've already seen this with analyzing moles. We've seen this like in lots of different places. Um, and that brings me back to this one other point that you made earlier, which I thought was really interesting, um, how they're going to hit a wall. The Elon's hit a wall and they basically start building subsystems. This is exactly how I define AGI. I don't think AGI is going to be some breakthrough, generally intelligent thing. I think that is more of an AI breakthrough than we've actually seen. I don't think it's happened yet. What I'm arguing is going to replace the average knowledge worker is actually one of these, uh, Frankenstein systems where the LLM is like an orchestrator, and it's basically spinning up these individual things to go and accomplish individual tasks like write the email, organize the events, organize the conference, uh, summarize this or whatever. So it's going to have at its disposal dozens or hundreds of these little workers. But the problem is that you're going to buy this thing, let's call it a synth worker or whatever the stupid AI company is going to be called. You buy, you buy synth worker, synth worker shows up to work on Monday. It goes to the onboarding meeting. It talks to the manager. It talks to all its team members. It reads slack, it reads the wiki. And it starts working just like a regular employee. And it produces the work of a regular employee. And I would argue in a lot of cases, even better. Um, but if you look under the hood, it's not some all intelligent AI system. It's actually a whole bunch of tricks. It's like this big agent orchestrating a whole bunch of other ones. But functionally, what you end up with is this new AI named Julie actually does the work. And guess what? We didn't hire a human to do that job.

Um, so I guess the first thing I'd say is that is not the definition of AGI that most people are going with. Um, but it is a fair point. Um, a lot of people have this idea of, uh, if we give the llms enough data, at some point, they start learning for themselves, and at some point they can do all of these things at the Submodels are doing, um, which I think most likely we both agree that that's that's horseshit. That's that's never happening. Um, but then the problem is when it comes to customizing these models, um, now you need a custom model for everything. You, uh, you need your model to identify tumors. And that is no different from where we've been like. That is what we've been doing for the past. Uh, I don't know. Probably longer than I've been alive, like 30 years, maybe more. And that hasn't replaced humans. Um, I would go back to my own personal example, which is machine learning based malware detection. Those machines have massively improved, uh, malware detection. Like, uh, back in the old days, you would have to open up a virus. You would have to manually determine that this is actually malicious code, and then enter a signature into a database. And can you imagine, like a human trying to do that with billions and billions of new files every single day? It's impossible. Um, so we built machine learning models to do that. But then the question is, well, how many malware analysts got laid off as a result? And the answer is there's actually more malware analysts today than there was in those days before the machine learning based virus classification. And the reason was the classification could only do so much, and you still needed people to check false positives. You still needed people to write reports, and there was always a place for humans and it wasn't even different skills like, uh, my job has been automated three times now, and in none of those three times have I had to, uh, even learn like, some massively different skill set. I've always been able to pivot, and I think that is all we're going to see with that is like, okay, maybe our cardiologist, I can automate a lot of that. So the cardiologists end up doing something else. I can't say what else they might end up doing because that's not my field and I don't know anything about it. But I do know that in my field, we have just been automating every single thing. We've been building sub models for malware classification. We've built some models for, uh, security alerts. We've built uh, sub models for analyzing code. Um, and we built up all of these sub models. But there's two problems there. The first is there is no evidence of it ever actually replacing humans. In fact, we've, uh, the industry has grown by a factor of ten since these machine learning models started coming out. And the second thing is they're very expensive to produce and maintain. So you're not actually getting the thing, which is the AI companies are trying to sell people on, which is, oh, you can get rid of employees, get rid of Greg, get rid of Simon, get rid of Sandra, and just keep all their salaries. Just pay $20 a month for this AI subscription. And now you're saving all of this, all of this money. But then when it comes to the actual who's maintaining these models, who's updating them for the latest, uh, whatever it is they're doing, because, um, like, obviously the technologies that they're working on are always changing. So someone has to maintain these models and they're phenomenally expensive to run. Like any kind of pattern matching model requires insane amounts of computational power. And what's actually just happening is you're taking those salaries, and I don't think you're even cutting them. You're actually spending more running these models or paying to have the models run than you would be having employees do it. It only makes sense when you're operating at scale, where employees physically like there aren't enough employees to do this thing. Like, imagine if you had to hire, uh, humans to manually categorize a billion malware samples. Like, it just wouldn't be possible. Uh, but when it actually comes to, hey, let's replace a human with this AI, uh, it usually ends up just working out more expensive. And the only reason we haven't started seeing that with Llms yet is because they're being subsidized by VCs. We don't see how much. Uh, I'm just going to say ChatGPT, because that's the one most people are familiar with. Um, we don't really see how much ChatGPT costs to actually operate. We don't know how much they're spending on acquiring the training data, training the model, how much computational power is being used, how much electricity is being used. And then when you think about an actual human being, it's just some cells. They need a little bit of water, a little bit of glucose, the amount of money it costs to operate a actual human being is almost zero. The only reason it's not zero is because capitalism, we've made houses so expensive, so now they've got to pay rent and we want to make a profit selling food. So now they've got to buy, uh, like this $0.01 banana is now a dollar. Um, so we basically just increased the cost of running humans to the point where these machines now look compatible. But then when you try and actually replace the humans, you're just going to run into the exact same problem, which is these machines are expensive. They may not need to pay rent. Actually, technically they do need to pay rent. You need a data centre, but they may not need to eat. They may not need water, they may not need plumbing. Uh, but they need a billion, trillion dollars worth of GPUs, a ton of electricity, a ton of cooling. And I actually don't think we are going to reach a point where we can make, uh, like computational based intelligence more cheaper than the equivalent human intelligence. I think right now we're in, uh, I call it sort of the Utopia era. Even though I don't feel like we're in Utopia, I think everything is terrible. I hate what's going on with AI. It makes me miserable. But it reminds me of, um, uh, in LA, we had this thing called, uh. Well, it's actually it's quite widespread now, but we had this program called the Bird Scooter, and it was actually the very first bird scooter was in LA, and I was there for the pilot program. It's like, okay, free public transport. You pay like maybe a dollar at most. You can go anywhere, you can leave the scooter anywhere you want. And I was like, this is incredible. Like, this is going to revolutionize transport. I can go from point A to point B, I don't need to find parking. It cost me a dollar. Like this is perfect. But what was actually happening on the back end is VCs were pouring billions and billions of dollars into this system to try and, like, build out this company. And what was happening is people were getting hit by cars. They were throwing the scooters into rivers and oceans, and they were actually losing more money, uh, to running this program than they were ever going to make. And I think we're seeing the same thing start to happen with AI is it's being funded by VCs, but it's there's not really yet any certain path where it's even going to be profitable. Like they are currently operating at a loss, and they're selling us on this idea that at some point you will be able to run like a genetic AI or whatever for less than the equivalent of an employee. But I actually don't think that's going to happen. I don't think we're going to reach a point where computers are cheap enough to run, that we can replace humans with these, uh, with these AI agents. And I think the more that we start, uh, going from this idea, which is never going to work of a single AI that is just able to do everything to building out sub models, I think that's just going to it's going to compound the cost. Uh, we're just going to get more and more and more cost to the point where eventually we'll come to the realization that actually, it's cheaper to just hire humans.

Um, so I guess the first thing I'd say is that is not the definition of AGI that most people are going with. Um, but it is a fair point. Um, a lot of people have this idea of, uh, if we give the llms enough data, at some point, they start learning for themselves, and at some point they can do all of these things at the Submodels are doing, um, which I think most likely we both agree that that's that's horseshit. That's that's never happening. Um, but then the problem is when it comes to customizing these models, um, now you need a custom model for everything. You, uh, you need your model to identify tumors. And that is no different from where we've been like. That is what we've been doing for the past. Uh, I don't know. Probably longer than I've been alive, like 30 years, maybe more. And that hasn't replaced humans. Um, I would go back to my own personal example, which is machine learning based malware detection. Those machines have massively improved, uh, malware detection. Like, uh, back in the old days, you would have to open up a virus. You would have to manually determine that this is actually malicious code, and then enter a signature into a database. And can you imagine, like a human trying to do that with billions and billions of new files every single day? It's impossible. Um, so we built machine learning models to do that. But then the question is, well, how many malware analysts got laid off as a result? And the answer is there's actually more malware analysts today than there was in those days before the machine learning based virus classification. And the reason was the classification could only do so much, and you still needed people to check false positives. You still needed people to write reports, and there was always a place for humans and it wasn't even different skills like, uh, my job has been automated three times now, and in none of those three times have I had to, uh, even learn like, some massively different skill set. I've always been able to pivot, and I think that is all we're going to see with that is like, okay, maybe our cardiologist, I can automate a lot of that. So the cardiologists end up doing something else. I can't say what else they might end up doing because that's not my field and I don't know anything about it. But I do know that in my field, we have just been automating every single thing. We've been building sub models for malware classification. We've built some models for, uh, security alerts. We've built uh, sub models for analyzing code. Um, and we built up all of these sub models. But there's two problems there. The first is there is no evidence of it ever actually replacing humans. In fact, we've, uh, the industry has grown by a factor of ten since these machine learning models started coming out. And the second thing is they're very expensive to produce and maintain. So you're not actually getting the thing, which is the AI companies are trying to sell people on, which is, oh, you can get rid of employees, get rid of Greg, get rid of Simon, get rid of Sandra, and just keep all their salaries. Just pay $20 a month for this AI subscription. And now you're saving all of this, all of this money. But then when it comes to the actual who's maintaining these models, who's updating them for the latest, uh, whatever it is they're doing, because, um, like, obviously the technologies that they're working on are always changing. So someone has to maintain these models and they're phenomenally expensive to run. Like any kind of pattern matching model requires insane amounts of computational power. And what's actually just happening is you're taking those salaries, and I don't think you're even cutting them. You're actually spending more running these models or paying to have the models run than you would be having employees do it. It only makes sense when you're operating at scale, where employees physically like there aren't enough employees to do this thing. Like, imagine if you had to hire, uh, humans to manually categorize a billion malware samples. Like, it just wouldn't be possible. Uh, but when it actually comes to, hey, let's replace a human with this AI, uh, it usually ends up just working out more expensive. And the only reason we haven't started seeing that with Llms yet is because they're being subsidized by VCs. We don't see how much. Uh, I'm just going to say ChatGPT, because that's the one most people are familiar with. Um, we don't really see how much ChatGPT costs to actually operate. We don't know how much they're spending on acquiring the training data, training the model, how much computational power is being used, how much electricity is being used. And then when you think about an actual human being, it's just some cells. They need a little bit of water, a little bit of glucose, the amount of money it costs to operate a actual human being is almost zero. The only reason it's not zero is because capitalism, we've made houses so expensive, so now they've got to pay rent and we want to make a profit selling food. So now they've got to buy, uh, like this $0.01 banana is now a dollar. Um, so we basically just increased the cost of running humans to the point where these machines now look compatible. But then when you try and actually replace the humans, you're just going to run into the exact same problem, which is these machines are expensive. They may not need to pay rent. Actually, technically they do need to pay rent. You need a data centre, but they may not need to eat. They may not need water, they may not need plumbing. Uh, but they need a billion, trillion dollars worth of GPUs, a ton of electricity, a ton of cooling. And I actually don't think we are going to reach a point where we can make, uh, like computational based intelligence more cheaper than the equivalent human intelligence. I think right now we're in, uh, I call it sort of the Utopia era. Even though I don't feel like we're in Utopia, I think everything is terrible. I hate what's going on with AI. It makes me miserable. But it reminds me of, um, uh, in LA, we had this thing called, uh. Well, it's actually it's quite widespread now, but we had this program called the Bird Scooter, and it was actually the very first bird scooter was in LA, and I was there for the pilot program. It's like, okay, free public transport. You pay like maybe a dollar at most. You can go anywhere, you can leave the scooter anywhere you want. And I was like, this is incredible. Like, this is going to revolutionize transport. I can go from point A to point B, I don't need to find parking. It cost me a dollar. Like this is perfect. But what was actually happening on the back end is VCs were pouring billions and billions of dollars into this system to try and, like, build out this company. And what was happening is people were getting hit by cars. They were throwing the scooters into rivers and oceans, and they were actually losing more money, uh, to running this program than they were ever going to make. And I think we're seeing the same thing start to happen with AI is it's being funded by VCs, but it's there's not really yet any certain path where it's even going to be profitable. Like they are currently operating at a loss, and they're selling us on this idea that at some point you will be able to run like a genetic AI or whatever for less than the equivalent of an employee. But I actually don't think that's going to happen. I don't think we're going to reach a point where computers are cheap enough to run, that we can replace humans with these, uh, with these AI agents. And I think the more that we start, uh, going from this idea, which is never going to work of a single AI that is just able to do everything to building out sub models, I think that's just going to it's going to compound the cost. Uh, we're just going to get more and more and more cost to the point where eventually we'll come to the realization that actually, it's cheaper to just hire humans.

Okay. I like this argument. I find that interesting. I don't believe it's true. My intuition is that it's not true. And the reason for that is just the the amount that the, um, the costs are falling. And the fact that I think we're so bad at what we're doing right now with artificial intelligence. I think there's so much slack in the rope that I think we're going to end up getting, you know, 99% or whatever, some, some dramatic amount of the cost that we're paying now just will keep falling out and falling out. But I would say that's an empirical question. And we'll just have to like see how that works out. Because I do agree with you that there is a the current state is like a state of confusion because you don't know how much it's being propped up. So so I like your argument, and I definitely think the there will be an effect from what you're saying. Um, I, um, I want to I want to touch on this, this whole concept of intelligence itself in like, because I feel like, um, we were trying to define it before, uh, when we had the previous conversation, we were trying to define like what, uh, what exactly do we mean by it? So I think the way that I defined it before was, um, or at least the way that I wanted to define it now, which which I think is roughly similar, the ability to solve everyday human problems using knowledge about the world. So I think that I try to hit it from a bunch of angles, and I think that accounts for why an average knowledge worker can't be automated. Um, and I put every day in there because it's basic stuff like, should I break up with my boyfriend? Um, and keep in mind, just just the case of, like, how useful this stuff is. I think the current numbers are OpenAI or ChatGPT has like a billion users per day. So I feel like for everyday problems, It's if we're using that as a as a metric doesn't that to me indicates that it's very useful. Um, but but I guess that's a red herring because you're, you're already acknowledged that it's useful. Um, but I've got some examples of, like, things that ChatGPT is really good at. So write an essay about a book that you read. It could already write pretty good essays. Oh, and by the way, this is unrelated to the debate, but we actually blew by the Turing test because most people can't tell the difference between AI and humans at this point, because it's already. I just thought it was an interesting piece of trivia that that used to be the gold standard for AI, and now no one even cares that we passed it. Um, managing.

Okay. I like this argument. I find that interesting. I don't believe it's true. My intuition is that it's not true. And the reason for that is just the the amount that the, um, the costs are falling. And the fact that I think we're so bad at what we're doing right now with artificial intelligence. I think there's so much slack in the rope that I think we're going to end up getting, you know, 99% or whatever, some, some dramatic amount of the cost that we're paying now just will keep falling out and falling out. But I would say that's an empirical question. And we'll just have to like see how that works out. Because I do agree with you that there is a the current state is like a state of confusion because you don't know how much it's being propped up. So so I like your argument, and I definitely think the there will be an effect from what you're saying. Um, I, um, I want to I want to touch on this, this whole concept of intelligence itself in like, because I feel like, um, we were trying to define it before, uh, when we had the previous conversation, we were trying to define like what, uh, what exactly do we mean by it? So I think the way that I defined it before was, um, or at least the way that I wanted to define it now, which which I think is roughly similar, the ability to solve everyday human problems using knowledge about the world. So I think that I try to hit it from a bunch of angles, and I think that accounts for why an average knowledge worker can't be automated. Um, and I put every day in there because it's basic stuff like, should I break up with my boyfriend? Um, and keep in mind, just just the case of, like, how useful this stuff is. I think the current numbers are OpenAI or ChatGPT has like a billion users per day. So I feel like for everyday problems, It's if we're using that as a as a metric doesn't that to me indicates that it's very useful. Um, but but I guess that's a red herring because you're, you're already acknowledged that it's useful. Um, but I've got some examples of, like, things that ChatGPT is really good at. So write an essay about a book that you read. It could already write pretty good essays. Oh, and by the way, this is unrelated to the debate, but we actually blew by the Turing test because most people can't tell the difference between AI and humans at this point, because it's already. I just thought it was an interesting piece of trivia that that used to be the gold standard for AI, and now no one even cares that we passed it. Um, managing.

That one was actually passed the with the very early GPT. Um, I thought I.

That one was actually passed the with the very early GPT. Um, I thought I.

Think it might.

Think it might.

Have. Yeah. Because I mean, so as, like a, like putting my, uh, like my philosophy hat on instead of my scientist hat on. I have never, ever seen the Turing test as a reliable metric for anything. Because essentially, for anyone who doesn't know what that is, it is just you put a machine and you put a person, uh, like basically concealed from the person they're talking to. And the question is, can that person distinguish between a machine or a human conversation? But the bar for that is very low, because your average person will spend like 30 hours arguing with a troll bot on Twitter, not realizing that that is a Python script. Um, so I've always felt that Turing test is more a test of humans lack of ability to distinguish a real human, rather than, uh, AI being successful at what it does. Now, granted, I will concede that AI is, or at least Llms are very, very good at imitating human like conversation. I will give them that. But there is a huge difference between imitating human like conversation and human intelligence, which is why I believe that people aren't, uh, that they're not seeing it passing the Turing test as this amazing feat because it's not showing that the AI can mimic human intelligence. It's showing that it can mimic human conversation, which is a very different thing.

Have. Yeah. Because I mean, so as, like a, like putting my, uh, like my philosophy hat on instead of my scientist hat on. I have never, ever seen the Turing test as a reliable metric for anything. Because essentially, for anyone who doesn't know what that is, it is just you put a machine and you put a person, uh, like basically concealed from the person they're talking to. And the question is, can that person distinguish between a machine or a human conversation? But the bar for that is very low, because your average person will spend like 30 hours arguing with a troll bot on Twitter, not realizing that that is a Python script. Um, so I've always felt that Turing test is more a test of humans lack of ability to distinguish a real human, rather than, uh, AI being successful at what it does. Now, granted, I will concede that AI is, or at least Llms are very, very good at imitating human like conversation. I will give them that. But there is a huge difference between imitating human like conversation and human intelligence, which is why I believe that people aren't, uh, that they're not seeing it passing the Turing test as this amazing feat because it's not showing that the AI can mimic human intelligence. It's showing that it can mimic human conversation, which is a very different thing.

Yeah, that makes sense. I think we agree there. Um, so what about this definition that I'm using? Um, and I can't remember the exact definitions that we used before. We had a slight disagreement there. But what do you think about this? The ability to solve everyday human problems using knowledge about the world. So it's kind of general. Right. And it's general in the sense that a knowledge worker could do it. It's general in the sense that AGI would be able to do it. AC obviously would be able to do it. But I've got examples here managing someone's daily schedule, writing performance review for an employee, reading and answering emails, writing code for an application at a company. Um, by the way, I completely agreed with, um, your your differentiation between writing code and doing software engineering or software architecture. Completely agree with you there. I think that's a big thing that coding is missing out on. Um, writing a requirements document based on conversations, writing a work status update for your boss plan, a four day trip to Switzerland. So these are my examples of like everyday human problems that are nowhere near free, like automation could possibly even try to do versus, um, with AI, these are kind of like trivial. And I would argue that these are the types of things that are going to make up a human replacing AI in the workforce.

Yeah, that makes sense. I think we agree there. Um, so what about this definition that I'm using? Um, and I can't remember the exact definitions that we used before. We had a slight disagreement there. But what do you think about this? The ability to solve everyday human problems using knowledge about the world. So it's kind of general. Right. And it's general in the sense that a knowledge worker could do it. It's general in the sense that AGI would be able to do it. AC obviously would be able to do it. But I've got examples here managing someone's daily schedule, writing performance review for an employee, reading and answering emails, writing code for an application at a company. Um, by the way, I completely agreed with, um, your your differentiation between writing code and doing software engineering or software architecture. Completely agree with you there. I think that's a big thing that coding is missing out on. Um, writing a requirements document based on conversations, writing a work status update for your boss plan, a four day trip to Switzerland. So these are my examples of like everyday human problems that are nowhere near free, like automation could possibly even try to do versus, um, with AI, these are kind of like trivial. And I would argue that these are the types of things that are going to make up a human replacing AI in the workforce.

I'm not sure I agree because, um, uh, the the point on, uh, I mean, it goes back to your general knowledge worker point, the question of can it replace a average worker? Um, and we've talked about that already. Uh, but I don't think that is the definition of intelligence. For me, the definition of intelligence is not the ability to perform a task that wasn't performable by previous automation. It is the ability to work with incomplete information. Like if I were to, I don't know. Let's say I build you a puzzle and I take away some of the pieces and let's say I take away five of the pieces. Um, and I give you the puzzle with the, uh, without those pieces that I took away, you could probably at some level redraw the rest of the puzzle. Right? You could use your, uh, critical thinking and your intelligence to see. Oh, okay. I don't have all the pieces here, but I can see this is clearly a photo of a waterfall. Right? It's that ability to work with. Not what we know, but what we don't know. That is my definition of intelligence. Um, because I think I think it gets a little confusing because at least in the early schooling era, they're not really teaching you intelligence. They're teaching you knowledge. Uh, the teacher will just give you a fact, and you memorize that fact, then they'll give you a test and you repeat back the fact. And that is the metric that a lot of people are using for large language models. Yeah, they're doing like bullshit, like, oh, can it do the SATs? Can it do the bar exam? Like, of course it can do the bar exam. It's a database of all the answers. It can just regurgitate the answers. Now, my definition of intelligence would be okay. What happens when we put things that, like, we ask it to do things that aren't in its database, which is admittedly very hard for large language models because, well, they have all of human knowledge pretty much in there. They have like every book, every movie, every web page. So it's very hard to find a task where the AI cannot just use basic pattern matching and be. It isn't in their data set. And this is where I say AI is not in any like any definition of the word intelligent. Um, and the way I would assess this is by product. Um, if we ask the AI to do something that it already has in its database. It's very easily going to be able to do it because it already knows the answer. But then we have a lot of problems, like we as humans have a lot of unanswered questions. We have mathematical problems that have not yet been solved. We have incomplete scientific theories. Now, uh, I like to go to Einstein for this example, because Einstein lived before computers. Uh, in his day, he could read as many books as he could read. He couldn't Google things. He couldn't control F through a book. He would have to read a book or a scientific paper. And back then I think they were published in like, actual physical journals. So he would have had to gone and got a scientific journal and read through it to learn a little bit about a thing. And with as little like, I'm not going to say Einstein wasn't knowledgeable, but with as little knowledge as he had access to, he was able to create theories like special and general relativity, which revolutionized physics. Now, in physics classes today we work with a lot of his theories. And your average like, not very bright college student can can work with Einstein's theories because we already have the complete picture. He's already come up with them. He's already given us the answers and we can now work with them. But what made Einstein so amazing is he did not have the answers. He came up with them initially. So, um, kind of my point there would be, okay, so Einstein could come up with this amazing theory with very, very limited access to information. We have these machines that have, I don't even know how much trillions and trillions and trillions and trillions of data points like every book, every scientific paper, absolutely everything is in their knowledge database. But like, what have they come up with? If I ask it for a theory that unifies classical physics and quantum physics, it just shits out some existing theories that someone else came up with, and if it had any intelligence at all, with the sheer amount of knowledge it has access to and the sheer amount of computing power it has access to, I would expect it would be able to complete at least one of those theories. Like the fact that it just has so much knowledge, yet seems to have done nothing novel whatsoever. And in my opinion says that not only does it lack intelligence, it isn't intelligent at all.