HockeyStick #4 - The Complete Obsolete Guide to Generative AI

Episode 4 is here and it was so much fun! I had a chance to chat to David Clinton , the author of the new book under the same name (Manning Publications Co. ) about the practical applications of Generative AI, how it's going to replace him as a technical author, how we wrote 5 books using ChatGPT and published them on Amazon and so much more!

I hope you enjoy listening as much as I enjoyed recording it!

Podcast

Follow HockeyStick Show and find the episodes below:

Web: https://hockeystick.show

Spotify: https://open.spotify.com/show/2jHvvkUkHAU8GRRHKbwAyJ

YouTube: https://www.youtube.com/@HockeyStickShow

Video

Audio

Summary

Here's what AI thinks about it:

This script delves into the transformative impact of Generative AI in various industries, covering discussions on the rapid advancement of AI, potential role replacements, large datasets, and AI tools' evolving landscape. It also touches on copyright issues, the OpenAI lawsuit, voice cloning, AI music and video, AGI, quantum computing, challenges with AI-generated content, and predictions for the future of technology.

Transcript

Miko Pawlikowski: [00:00:00] I'm Miko Pawlikowski, and this is Hockey Stick. Generative AI is eating the world just like software and the internet did before it. Probably everything you read and see today is at least in part generated or touched up by AI models. Revolutions tend to happen fast, but this one feels faster than anything I've seen before.

Miko Pawlikowski: Today, I'm joined by David Clinton, the author of the playfully named book" The Complete Obsolete Guide to Generative AI". He's a serial author of technical books, including titles like "Linux in Action", "Linux Security Fundamentals", "Ubuntu Linux Bible", "AWS Study Guides", and more. When interrogated, he admitted to a dozen titles.

Miko Pawlikowski: In this conversation, we'll cover why Gen AI, not to be confused with Gen X, is such a big deal, why it's happening right now, what breakthroughs made it possible, and how you too can incorporate it in your life right now. And become more productive right away. Welcome to this episode and please enjoy.

Miko Pawlikowski: David, before I even [00:01:00] met you, you've already given me a lot of trouble, because for most of my guests I can more or less describe them in three sentences. With you, on the other hand, I started looking up the books that you wrote, and I can't even count them.

Miko Pawlikowski: And, spoiler alert, David, I'm not sure he knows himself, because in one of the chapters of his book, we're going to talk about, "The complete obsolete guide to generative AI", he's asking ChatGPT to count the books that he's written and ChatGPT doesn't know either. So David, how many books have you written at this stage?

David Clinton: it depends. Are you talking about the books under this name or under other names?

Miko Pawlikowski: wow. Okay. Including your all other alter egos.

David Clinton: Yeah, but at least the ones with my own name, David Clinton, various editions, maybe a dozen.

Miko Pawlikowski: "Linux in action". "Linux security fundamentals", Ubuntu Linux Bible". I think you co-authored that one. And there's a lot of study guides as well.

David Clinton: the Wiley study guides. By the way, being [00:02:00] a Wiley author, makes me a colleague of Herman Melville, the author of "Moby Dick", because Herman Melville's publisher was John Wiley himself 200 years ago. The company somehow has survived 200 years. And the chairman is still a Wiley.

David Clinton: It's Jesse Wiley now it's absolutely amazing how they've survived and managed to adjust. There wasn't a lot of AWS, training content necessary, required back in the 1820, but, they seem to have adjusted and foundnew places. However, maybe near the end of this run, because one of the, niches, that generative AI is already eating is,technology training books wide by a technology training book.

David Clinton: if, GPT can tell you exactly what you want to know right now.

Miko Pawlikowski: Wow. So that's reason enough to go via Wiley. Did you prefer Wiley to Manning? What's your favorite publisher at this stage?

David Clinton: They're very different. and I like them both, each in their own way. Manning [00:03:00] is very demanding from an editorial process perspective. And they take a long time on everything. A long time. That's why the book is called "Obsolete...". they're good. They put out a great product.

David Clinton: Wiley has a much lighter hand from an editorial perspective. and, they have, excellent connections in the industry as much as possible to move books in large numbers, and they're a solid, reliable company. But they each have their value and each have the things they do best.?

Miko Pawlikowski: before I even opened the book, from the title I got the hint that it was going to be a fan book, "The ,Obsolete Guide". by the way, I'm still not sure how you passed that by the Manning editorial team and how did they

David Clinton: too, they loved it right away, actually, I was really surprised I threw that out I was angry with them. They were taking so long. So I just threw that title out and they actually bought it and they went with it.

Miko Pawlikowski: that's remarkable. So why did you write this book? I read the book and, you basically put in there a legend slash myth, about how [00:04:00] they effectively told you to stop pestering them about using generative AI and just write a book about it already. is that the short version of what happened?

David Clinton: it's close enough, as far as I can remember. we've had a good relationship for many years and, they've sometimes asked me to do things I didn't want to do, and sometimes I've pitched ideas to them they didn't want. this one we were talking about on and off over a year, we were talking about generative AI in different contexts, and then, they wanted to invest heavily in generative AI as far as their books and other content.

David Clinton: They didn't want to get left behind, which is a very smart decision. So a year ago, they were looking at publishing 10, 15, 20 books in general topic of generative AI. And they knew I was someone who thought a lot about it, so they were pushing me to do something. and I was pushing them to go away because I didn't think they could move fast enough for this particular market.

David Clinton: somehow, I don't remember exactly how, but we latched on to this project and, so far it's sold relatively well in its [00:05:00] pre-MEAP sales. sold about 1100 copies, so far. And this is, before it's published and before it hits Amazon. we can always, we can always hope.

David Clinton: And, but I'm not disappointed. The process, has been reasonable, if way too long.

Miko Pawlikowski: let's jump to the main question then. why is generative AI such a big deal at the moment?

a€?

David Clinton: It's upending everything. almost everything. There are people who do things with their hands, who build the world and keep the world going,they're in an excellent position and they're still keeping the world going.

David Clinton: But everybody else is at risk. but then again, 200 years ago, in the Industrial Revolution, we had the same fears that there were certain mechanical tools and processes were genuinely predicted and incorrectly predicted to be about to replace the vast majority of labor. And what would all those laborers do?

David Clinton: But it actually worked out quite [00:06:00] well. there wasn't mass starvation as a result of the industrial revolution. There are other issues, political issues, but not the industrial revolution issues. And, the economy and the market somehow reached an equilibrium and settled. and as, progress has been significant in the last 200 years, probably we're going to see something similar.

David Clinton: There's going to be huge upheaval. Generative AI can do things that highly skilled people were necessary for 5 years ago, and are no longer necessary. the things I do, I never truly understood why people bought my books. like "AWS Solutions Architect Study Guide". where did they think I got my information for writing the book?

David Clinton: I got it from the internet, right? Like everybody else. And if I could have gotten it from the internet, why doesn't everybody else? I'm not saying you shouldn't buy the book, of course you should. But, I just don't understand why. if you're just willing to invest a little bit of research time, you can get everything I got.

David Clinton: I'm gonna be replaced. At least that part of what I do. And, it's scary. For good [00:07:00] reason. And it's, exciting. And it is a big deal. I don't thinkmuch of the hype is oversold. This really is a big change.

Miko Pawlikowski: When you were talking about construction site, I just remembered, I saw the video on the internet and hopefully it's true and not AI-generated, but it was basically a construction site with the scaffolding with a big banner on it saying "ChatGPT finish this building".

David Clinton: they have experimented, I think, successfully already with, print-on-demand houses. where they can actually layer, a layer after layer of a construction material to build the walls and even the ducting and some of the wiring for a house in a day.?

Miko Pawlikowski: it's interesting that you should mention that, because I think 3D printing, and I will include the housing printing in that category, held such a premise not that long ago, maybe a decade and a half ago when we were picturing everybody printing anything, like in the [00:08:00] sci-fi movies, when you just download something and then you click it and a minute later you get an object, physical object, people were talking about how It's going to be so dangerous because people can just download a plan for a gun and print it at home, or they can do,replicas of things that will be indistinguishable from truth. But somehow that didn't really happen. what I actually see people printing is parts for their,drones when they have to go through quite a lot of propellers when they're racing them and, the mandatory figurines that everybody prints when they actually buy a 3D printer initially, and not an awful lot more. So how do we know that? Why are we so sure? And that seems to be prevailing that the AI revolution that we're seeing is different from that.

David Clinton: first I wouldn't say that, that, 3d printing is dead. You're right. it's, dormant.?

Miko Pawlikowski: but I think it is happening probably in niche areas where it's making a difference. they are [00:09:00] printing, 3D guns, they're just not advertising it because in many jurisdictions it's illegal. I think that's happening probably more than we'd like. so it's there and it'll find a niche, it'll probably find a market somehow. However, AI I think is different, that's not to say that it's going to play out exactly the way we think it will now. but markets will discover,the value of AI in ways that probably we haven't anticipated. it's so broad and potentially complex, it won't be ignored. There are too many creative people and imaginative people who are focused on it now, that I can't imagine that substantially transformative things won't happen. I can't tell you what they'll be necessarily, I can guess. but I can't believe that it won't happen. This is a tool that's far too valuable to be ignored. But, hey, I've been wrong before. In fact, if you ever want to know how to anticipate the results of an election, just ask me what I [00:10:00] think, who I think is gonna win, and go the other way. Because I'm always wrong in my predictions.?

Miko Pawlikowski: So you hinted a little bit at one group of people that you expect to be particularly affected. And you included yourself in the technical writing. And I think to a certain extent I'm ready to agree with you. Probably the ChatGPT version of, technical documentation is going to be equally dry as the majority of the documentation that I come across myself.

David Clinton: Who do you think, who else, is at the highest, end of the possible,affection of, their livelihoods effectively. I wouldn't want to be a young accountant. what accountants do is easily replicable by AI right now, and certainly in the next few years, they, mentally process a lot of numbers and trends. And of course there are a lot of accountants, different accountants do different things. They have different tasks. But at core, [00:11:00] they're processing data, whether it's financial data, or sales data, and they're trying to, number one, organize it, so that they can file their quarterly filings and their tax filings and everything, so they're trying to organize the information, but also the higher level accountants are anticipating where are these trends are going to affect their company, the company they're working for. And those are things that AI does exceptionally well already, and it's only going to get better. And when the time comes that they'll be integrating, accounting processes into AI systems. I think accountants will be looking for new work. Another thing, I read an article last week by, the owner of a small Canadian publisher, Sutherland House. He wrote about a conference of publishers he was at sometime before, and they had an AI specialist, whatever that is. I guess that's me, I don't know. And talking about how their industry was going to be changed. And they spoke about things like, AI translation. this publisher [00:12:00] likes the idea of taking excellent European books or books from all over the world that aren't written in English and have proven themselves as popular and successful books in their languages. But he said that translating them is expensive. hiring a good literary translator Is not cheap and it's not cost effective in his publishing model. But, feeding the manuscript to an AI translator will get you 80% of the way there. And then you hire a genuine human translator to fix it, to get the last 20%, but it's gonna be much cheaper. So those are things that are changing, but the AI specialist at this conference then said, one thing, he said, you can say goodbye to all your interns. And then the crowd, apparently the audience started screaming and shouting. that upset a lot of people. But it's probably true. An intern at a publishing company, their job is to make all the connections and do all the grunt work that the executives don't want to do or haven't got the time to do. And that's [00:13:00] exactly the type of work, the moving information, making appointments, making connections - that's exactly the sort of work that an AI can do well, or will be able to do well soon enough. So that junior knowledge worker is at risk. And of course the problem there is that they, the new talent comes into the publishing world through the interns. They get interns fresh out of college and they train them up. they get them to the point where they're. capable of doing bigger things. And if you wipe out the intern program, then you're wiping out your feed of new talent. That's disruptive. But,how we got a publishing company justify or a government for that matter, justify hiring hundreds of thousands of junior managers, if that can all be replaced with software.

Miko Pawlikowski: Let me challenge you a little bit on two aspects of that. First of all, it's not particularly difficult to get any of these models really to trip on basic arithmetics. Like it's gotten a little bit better now [00:14:00] GPT-4 that actually uses a calculator instead of trying to,do the arithmetics in it's head, but it's still not particularly hard to formulate it in a way that it doesn't trigger that and gives you a demonstrably false answer. I think you gave a few examples of that, from prior versions in the book as well when you were asking about, a population, that's the closest to the median in the list

David Clinton: can't remember. Yeah, it's been a long time since I wrote it, but you're right, definitely there have been problems. Yeah,

Miko Pawlikowski: So that makes me think that this is still better suited for situations where you've got a lot of groundwork. Like you said, the translation, for example, you can get it to be 90% good and then get someone to do the last 10% and you save yourself 90%of the cost. Great. but is that going to work the same way with the accounting? Can you get the accounts to be 90% good and then get someone to check the last 10%??

David Clinton: Possibly. I don't see why not. they do have auditors for a reason, and they can just, adjust the [00:15:00] role of your company auditor to be looking over the shoulder of the AI rather than all the, junior accountants possibly,it's hard to know how it'll play out. But also as far as the errors go,

David Clinton: GPT and other AIs are prone to certain types of errors and they're working on them, but, my experience with, let's say, I feed now,really large spreadsheets and 400 page, corporate filings. I'll feed them to GPT and, I'll get into look for anomalies and add up all the line items related to this issue or that issue.

David Clinton: Basically doing SQL, requests on it. it's very good and then I'll usually check it up. I'll usually confirm, I'll load it up into, into Python and Pandas and do my own searches. and recently it's been pretty much always right. I think the hallucinations are disappearing.

David Clinton: They've probably not gone completely. And, we're gonna have fun still. But I think they've turned a corner, I think, in that field. With the exception, of course, of Google Gemini, which is a [00:16:00] complete and utter disaster, because they knew what they were doing.

David Clinton: everything I've read from insiders, they knew exactly what they were building with Gemini. It was a political indoctrination tool, and it was not there to aid humans, but to guide humans. Maybe they're right, maybe they're wrong, but it wasn't what they advertised it as.

David Clinton: And,I think at this point still Gemini is pretty much unreliable for anything serious. But,we've discovered that. Maybe it would be true of other engines also. And maybe in time we'll show that we were too harsh on Gemini, but at least at this point, if I could back up a bit, 10 years ago, I was working for some, I was involved with some startups and there's something about the startup culture that I really appreciated. I thought was very positive: when a customer would leave, an angry customer would leave or an angry employee would leave. Very often the startup culture meant that they would beg the customer or the employee, tell us what we did wrong. you're leaving, fine, we're sorry to see you go, wish you the best, but just tell us what we did wrong. We want [00:17:00] to improve. And I saw this more than once. CEOs would send email blasts to the whole company. "This is a problem somebody just reported. I want it fixed in 48 hours". They love to hear criticism, they wanted to improve. It was a culture of openness and it's all about money, of course, but okay.

David Clinton: What's wrong with incentives? there was a real culture of change. I think a lot of startups from 20 years ago, like Google, have lost that. Whether they want it or not, they can't do it. They're so ossified. They're so big. They're so impenetrable. And they're so enslaved to processes that they can't change even if they wanted to. And they couldn't really genuinely hear criticism, even if they wanted to. It's just, it's a shame. I think it's probably a natural process of all organizations. we know how governments work. The large organizations are very hard to improve when things go wrong. but it's a [00:18:00] shame that to some degree we've lost that culture.

David Clinton: And I think, that leads to huge mistakes like Gemini. there will be a shaking out of the industry that some will do better than others. So some AIs are going to improve and probably become steadily more and more astounding and some aren't. And, it'd be interesting to see which.

Miko Pawlikowski: Google is such an interesting topic, and you're obviously not the only one person saying that. I've read, at this stage, dozens of former employees, or even current employees, complaining about the same thing, that it's by no means anymore a startup. And, that's, to a certain extent, normal. Their size, their influence, people complain about the same things like them killing projects and replacing them with the same thing and renaming and rebranding and, completely not listening to customers.

Miko Pawlikowski: But I think that Gemini, or in general, the AI, situation at Google is, So interesting because they were at the forefront of [00:19:00] that. They spent a lot of money acquiring researchers, acquiring assets, they were there from the ground up?

David Clinton: It's wild. And it reminds me of Kodak. Kodak owned photography for almost a century, I don't know, 60, 70 years. And they worked on digital photography. They didn't ignore it. I think they built one of the first digital cameras. They were certainly in the first few. And they did not ignore the technology.

David Clinton: But somehow, they couldn't pivot. And they just didn't build a product that the market wanted. And where's Kodak now?

David Clinton: So Google's sitting on a lot of cash right now, they invested probably more than anyone in AI, and probably did ground breaking work which we're using now, but, where are they?

Miko Pawlikowski: there are other companies that are by no means startups and they don't move very fast. But for example, at this stage at least Microsoft's investments seem so much [00:20:00] better in that respect. They basically more or less bought themselves OpenAI, by the accounts of many people. And they're integrating that like crazy into everything at the moment.

Miko Pawlikowski: And where were they when Google was investing that,a decade ago? There is, this document, internal memo that I don't think we still know who wrote that. It leaked about a year ago and it was called something around "We Have No Moat". And neither does OpenAI, and I think the points that they were raising in there, whoever wrote that, are, they basically materialized a hundred percent.

Miko Pawlikowski: They were talking about how, all the things that Google is planning and hoping to achieve, like running the LLMs,on your phone, that's already happening. things like, scaling multimodality, today, this is all over the place, but the open source picked it up and ran with it.

Miko Pawlikowski: And obviously a big part of that is Facebook, leaking accidentally their model, [00:21:00] to, at least in my version of events to undermine a little bit the hegemony potentially. and give people alternative, but the open source models just run away with that and there's no going back.

Miko Pawlikowski: and then things like, cheap fine tuning with LoRA and later QLoRA, all of those things are making it even more so than that. Prophetic, memo, announced a year ago, And I just found it crazy that you can, waste your initial advantage so badly.

David Clinton: Although it's interesting that I don't have the sense that there's a huge uptake on truly open source and private LLMs. I'm not sure how many people are doing it for themselves. Yeah, of course, NVIDIA is selling a lot of GPUs to the people who are doing this in their houses.

David Clinton: But I still see OpenAI and perhaps Anthropic. Are still attracting a lot of customers. And I have no numbers, but I suspect they have a large majority of the market,for [00:22:00] generative AI is going through those portals rather than people doing it on their own. So while it is available, there are LLMs you can build on your own. I'm not sure how popular it is.

Miko Pawlikowski: I remember a study, that was querying people about how many models they're using. And I think the most popular answer was either two or three different models. And one of them was OpenAI, and one of them was local. And I think the local It's easy to picture people hacking away in their garages and,kind of movie-style Hollywood hacker, kind of thing, but what I think is actually happening is that, In addition to using, OpenAI's models primarily because they still got the biggest,chunk of the market, to do the kind of workhorse thing and get projects started to get something off the ground quickly, people are at least looking and investing into getting something that they have some kind of control over.[00:23:00]?

Miko Pawlikowski: And you can pick like a Mixtral you can pick, even a LLAMA, and if you're using that for a particular use case that you can test, to an extent that is actually feasible at the moment. and you can look into how it's, behaving on your data in your, on your hardware?

Miko Pawlikowski: That gives you. A massive advantage, like nobody can pull a rag from under you. You download that, and initially the licenses were a little bit funny, and you had to apply and do things like that, but now a lot of those models are Apache 2.0, I think it's not necessarily a question of either/or, it's more of a different use cases, make some of those options preferable.

Miko Pawlikowski: If I was starting a startup and I wanted to get somethingoff the line very quickly. It's like an obvious choice to get one of the APIs pay a few cents per thousand tokens and get it off. But as soon as you would like to make it, a permanent feature, and have any control, I think,the local models [00:24:00] really help hold a lot of progress. We'll see obviously where it goes, but, I think that's how I would describe it today. jumping back to, my second challenge to you, I wanted to compare, the generative AI that we're seeing today to self driving cars.

Miko Pawlikowski: And, I think I remember the self driving wars of, what was it, mid 2010s, when they were offering, researchers, multimillion dollar packages and stealing research from each other and Waymo and Uber and everybody. they got it to a point where it was, like, 90-something percent good, but that wasn't enough, right?

Miko Pawlikowski: It has to be effectively superhuman quality for people to get over the fact that there's no one behind the wheel. Why do we think that generative AI is not going to suffer from the same problem?

?

David Clinton: First of all, sometimes the stakes aren't as high with a self driving car. You could be killing people if you make a mistake. And that's definitely a threshold. You don't [00:25:00] want to slip underneath. whereas generative AI, a lot of it, maybe most of it is just, 'okay, so it gets a little math wrong'. It's not going to kill anyone.

David Clinton: that may be part of it. the second thing is that, generative AI is coming at us from so many different directions. And only one has to succeed and become the next, killer app or whatever. whereas, self driving cars are self driving cars. That's all they do. And you either succeed as a self driving car or you don't.

David Clinton: by the way, I think self driving cars are probably gonna make it. And they are still making progress. I think in Las Vegas, they now have the Waymo taxis are actually on the street running in real time, I think.

David Clinton: so that is eventually going to happen. of course, 6-7 years ago, they were predicting that all transport trucks would be, driverless by the year 2020, or whatever. but then again, having said that, when I was in high school,my geography teacher told us about the Club of Rome predictions.

David Clinton: This is, by the way, back in the, late 70s. So the Club of Rome predicted that the world would run out of [00:26:00] gold and platinum by the late 1980s and would run out of, petroleum by like a 1990 or so, and would run out of food by about 1995. I guess that didn't happen. maybe some of these predictions aren't to be taken that seriously either.

David Clinton: I do think that generative AI has a much better chance of pushing through obstacles simply because There's so many ways it could go. And there's so many, so many possible happy endings.

Miko Pawlikowski: would you venture your best bet of what's going to be the real killer app coming first?

David Clinton: I would say that Google with their their lack of a moat is a good target in the sense that internet search, may disappear as a business model and as a use case, simply because, augmented AI search is probably going to be better. The, last three, four months, I think I've had a, ChatGPT+ subscription.

David Clinton: [00:27:00] I bit the bullet and did it. And I'm happy I did because it's pretty amazing. So when I ask, ChatGPT+ to find some current information, it'll activate Bing and go search. So that's only going to get better. And with the power of AI itself, married to the ability to search under the hood, the entire internet.

David Clinton: Why do I need, Google search anymore? A search engine has been the way we organize the world for the last, 15 years, 20 years, I don't know. Okay. And it's the way we think. When you go to the internet, your thought was "let's go to a search engine and get there that way".

David Clinton: It may not be rational to use that in a year or so. so I would say that would be a big upheaval. Because it's not just Google that's gonna be a victim of this. Every person who's ever optimized his website for SEO,that advantage will disappear. That'll be hugely disrupting.

Miko Pawlikowski: certainly.

David Clinton: wouldn't even know how to approach building a website anymore. Because, your mind is set. [00:28:00] you optimize for SEO. But, maybe not.

Miko Pawlikowski: Yeah, it's crazy to think how much we rely on Google,so many people rely on their SEO for their livelihoods. And like you said, obviously you have to optimize for SEO, otherwise nobody will find you. So you have no traffic, so you earn no money.but Last time I checked, or I vaguely remember, I think about 80% of Google's income was still coming from ads, and lately, I don't know if it's just me thinking about it more than before, but I started noticing that most of, my first page, it's basically ads at this moment, so I do wonder, in the back of my head, if they just assumed that was gonna die.

Miko Pawlikowski: So they have to like ramp up and make as much money as possible right now

a€?

David Clinton: Squeeze whatever they can out. Maybe.

Miko Pawlikowski: Have you, have you tried? Yeah. Yeah.

David Clinton: for them, and it's scary for [00:29:00] everyone who relied on ads.

Miko Pawlikowski: Yes. the rudimentary version of that you can use like GPT or premium or whatever. There is a thing I've tried called Perplexity, full disclosure, no, connection to them whatsoever. How did you find that?

David Clinton: as far as, simple questions, code, if I asked for code or if I, asked for, a thinking question,it's been close. It was close to GPT.I haven't used it for a couple of months regularly, but, I was impressed. It didn't have the extra functionality of GPT+, So you couldn't upload a document and have it analyze the document, but as far as just talking with it, it was fairly close, as is Claude by Anthropic.

David Clinton: for some reason in Canada, they didn't fully release it even yet completely, but that much Claude I'm allowed to use in Canada, I found it to be, definitely comparable to, to GPT.

Miko Pawlikowski: I don't know how you feel about Elon Musk, but have you tried Grok [00:30:00] and the X AI

David Clinton: No, I haven't tried it yet. by the way,I love watching Elon Musk. He's so much fun. and the guy is, controversial, he's crazy. but that's all part of the fun, right? he's done things with SpaceX that are just so exciting. what every nerdy kid would have wanted to become.

David Clinton: He's doing it. He's living the life.?

Miko Pawlikowski: Yeah. so now when I test a new chatbot, I'm bored very quickly because I've, talked to them to death before, but Grok is differentiated by being, the opposite of safe and boring. So you can get it to say a lot of things very easily and, Even when you try, it's not gonna give you like this boring, dry, answer.

Miko Pawlikowski: Which is, at the very minimum, refreshing. And they actually open sourced, the model, recently. On what seemed like Elon Musk's complete whim, one day, to look better than OpenAI. Alright.

David Clinton: a surprise, who saw that coming? That Elon Musk would have a whim?

Miko Pawlikowski: from what I remember, it's a [00:31:00] mixture of experts architecture that was 300 and I want to say 20 or 40 billion, parameters. So despite being only 25 percent weights active at a time, you're still not going to be able to run it on anything anybody has at home.but you can talk to that if you've got, X premium or whatever, if you give Elon Musk money, which is an interesting sidequest.

Miko Pawlikowski: Which also reminds me, speaking of Elon Musk, I think another whim, on the 8th of August, we're supposed to see the Robotaxi.

David Clinton: the speculation is that he was, working on, Model 2, or whatever the cheapest version of Tesla is supposed to be. And, he internally killed it to push for the self driving taxi that doesn't even have a steering wheel or anything like that. And then Reuters wrote something about how, the, that model was dead based on people's, leaked opinions. He called them liars. [00:32:00] Something happened. And the next day it's Oh, we're going to unveil Robotaxi in a couple of months. So trademark, Musk, PR, version, but I am curious whether it's going to be svaporware or whether we're going to see anything substantial on that.

David Clinton: with his track record, the odds are pretty good. Of course, you have to be careful where you tell it to take you, because you may end up on Mars or something.?

Miko Pawlikowski: I guess the caveat to that is that the track record of delivery. Yes. But the track record of timelines, is usually a few years skewed, but, we'll see. Anyway, Before I get asked to go on a grand tour of the different, use cases that you discuss in your book, and I think a lot of people might find it valuable, and, they'll just go and buy your book right away, I wanted to also touch base very quickly on Why it's all happening now?

Miko Pawlikowski: Because a lot of these ideas, if not all of them, are decades old.what happened? Why is it happening [00:33:00] now? Is it all thanks to gamers and crypto miners??

David Clinton: partly, it's certainly technology, gains have been significant in the last five years. and, Moore's Law is always shaky, it's about to break, because we're moving too fast. But, I think the big thing is integration. Whereas 3-4 years ago,you could do,image segmentation, let's say, using a deep learning model.

David Clinton: You could do it three years ago, five years ago, ten years ago, probably, but it was a discrete tool. You could use it for that and only that, and don't try to integrate it with anything else. what's happened now is that we can integrate, let's say, the entire internet with the generative AI model and an engine and then integrate that with the way you work and think. So I can talk to it. I can type to it. I could plug it into my company is and its workflow and have it manage things for me. I think that's a big deal. We're starting to see, I guess, the first [00:34:00] stages of AI agents, which is,it's going to push us a lot further I would imagine. I ask an AI a question, I don't know, about my business, I want to launch a Twitter campaign to promote my new product, and so it'll tell me how to do it, but I say, be my agent, now go do it, and, people have done this already, actually, people have connected Twitter accounts to a GPT and actually had it manage their social media.

David Clinton: probably a huge improvement in most cases over the real thing, but I don't know there's an accident, you make a claim to your car insurance company. the AI could manage the insurance processing, it'll contact the underwriter.

David Clinton: And it'll collect all the details of the accident and what the insured property was and send those to the underwriter and the underwriter will send back an answer and it'll go to the bank and it could actually handle complex processes, independently and reliably, at some point reliably at any rate.

Miko Pawlikowski: So this integration and then the next step integration with creating agents, I think has been a big game changer. it's taken a lot [00:35:00] of discrete technologies and brought them together into something a whole lot bigger. Would you say that the datasets available also played a role? I

David Clinton: yeah, and not only, have they been getting, oh, taken as one of the projects I'm involved it would seem on the surface to be really unrelated to anything I've done before. I have a, substack called the Audit, where I write about, policy analysis in Canadian politics.

David Clinton: It's not something most people even in Canada are interested in, and certainly not outside Canada, but,the underlying premise of the site and of almost everything I write there, is that governments and institutions make an amazing amount of that data freely available, so almost everything that goes on in Canada's parliament has always been recorded in a document called Hansard, but now it's all available, and not only available online in, digital format, but it's now, sadly JSON rather than CSV mostly, but it's available in a format that you can actually [00:36:00] work with, and you could do a deep data analysis with it.

David Clinton: So governments on all levels, all over the world, almost, are producing a ton of data. As are various types of institutions. And that data is all becoming available. It's all becoming available, through APIs and, in automated context. So yes, that has made a huge difference and it's going to make a much, much bigger difference.

David Clinton: the further we go on this line.

Miko Pawlikowski: think there is also a slightly more nefarious, or I guess, gray area to that, which is, exemplified in this massive lawsuit against OpenAI from, the New York Times, I think. That alleges that they should, be compensated for the demonstrable fact that their model that's bringing them billions of dollars was actually trained on,the body of work that they've accumulated over decades. it's interesting because it seems to be very divisive. A lot of people [00:37:00] say, 'oh, If I read the articles and I write something, you're not going to chase me', which is fair. A lot of them say that, if we allow OpenAI to go with that, and then the next ones can't do that, they get unfair advantage.

Miko Pawlikowski: There's a lot of facets to that, and I'm curious, personally, which way it's going to sway. A lot of people don't like the copyright law right away, which is also understandable. but I am curious, how that's going to play out, whether going forward, things will continue getting more open or whether it's going to be the opposite, because even before the ChatGPT revolution, I want to say.places like Reddit, for example, were guarding what effectively is their IP from Google, because they don't want Google to serve you ads and get paid for that. They'd rather you came onto their platform and clicked on their ads, to prop their business. do you have a firm opinion whether, [00:38:00] which way you expect this to go?

David Clinton: These are tough questions, obviously. I appreciate copyright protections. I'm a writer. And I create courses, and I create books and content, and I appreciate being protected.

David Clinton: And if I wasn't protected, arguably, I wouldn't be able to write and create this content. In my particular case, that probably wouldn't be true, because thousands of people have copies of my books illegally. on the dark web and it doesn't really hurt my business that much because my customers buy through normal channels and, it's really hard to market my stuff illegally.

David Clinton: but nonetheless, creators should be protected because they have to be incentivized to create. On the other hand, there's only a certain distance you can push that. I've heard the argument in the New York Times case that New York Times authors Generally learn to write by reading Ernest Hemingway.

David Clinton: Does that mean that the Ernest Hemingway estate has a claim against the New York Times writers for a portion of their intellectual property. no one's going to say that when it's a question of training. If it was taking content or [00:39:00] taking ideas, okay, that's would seem to be a clear copyright violation, but when it's a question of training, I've heard the argument that,that's not a violation.

David Clinton: everybody trains on something, the whole purpose of content is to train people, how to read and write and think And of course we want people to learn from us. And of course you expect that and you can't prevent that. If you're going to put something in the public domain, you have to expect that people are going to use it.

David Clinton: so these are tough questions. it's, going to be hard to find a balance. I don't know, how it's going to play out. The odds are, technology will probably win. I have never seen technology lose for better, for worse. I guess we have seen technology lose.

David Clinton: And you think of. the, nuclear energy industry, it lost in a lot of countries, like Germany. the, nuclear energy genuinely has produced,the carbon neutral energy, expensive but clean energy for decades and decades. There have been a few accidents, tragically, but, by and large,the energy has been clean, relatively affordable, and has [00:40:00] many upsides and no downsides, and yet, the technology has been rejected in certain jurisdictions.

David Clinton: so I guess technology can lose,but by and large it doesn't. So if I would have to bet money, I would bet against the New York Times, I would bet, with technology, prevailing. in other words, the use of, of AI expanding.

Miko Pawlikowski: I like both your example and your counterexample andit's probably because I've been watching, on Netflix, Lincoln Lawyer for last couple of evenings so I want to say it also has to win in the court of public opinion. and that might actually be more tricky than, anything else.?

David Clinton: The point that you make in your book about open research on that and how that accelerated this, there's been obviously open research by some people who strongly believe that in that there was a research behind a cluster, but would you like to describe briefly what you meant in the book when you said that this, [00:41:00] Cambrian explosion of generative AI that we're looking at right now is in part made possible by the open nature of research.

David Clinton: as I think Linus Torvald said in the context of open source software, many eyes make problems lighter. That is, the more people who are reviewing your research and trying to replicate your research, the greater the chance you're going to find problems faster and improve things better and faster.

David Clinton: that was the premise of peer reviewed medical studies, which, of course, is apparently how it's collapsed and is in complete shambles now. But, in the open source world, and, particularly in a lot of the early research into AI, there was a spirit of collaboration. And much of what was created was put into the public domain and was shared and commented on and, improved as a result, and there was a lot of money invested. but there were an awful lot of people involved in the research. And they're, as far as I understand, I have [00:42:00] to be careful what I say. I can't talk about too many details. One, because I'm ignorant. And two, because someone, actual pioneer in the AI research world, read my book, "the, obsolete guide to,generative AI", Manning thought it was a good idea to send it to him, just to see what he thought.

David Clinton: And,the appendix at the end, which also includes a brief history of the AI developments, he really didn't like, and he felt that I had misrepresented, a lot of the pioneers and the specific, processes in advancing various AI technologies.

David Clinton: One step at a time, he felt got it all wrong. He was an insider. He was involvedin a lot of the original research. So the problem is,he's probably right. Yeah, he's probably more, a lot more accurate than I am. I'm not 100 percent sure how to take it though. this was a complex history with personalities, and sometimes personalities that didn't get along well with each other. This is the academic world, after all, where you can't become a great academic unless you don't get along well with people, it [00:43:00] seems. so I'm not going to go into any detail about, how exactly it happened, and who were the big movers and shakers, but it did happen, and a lot of it was done in an open, collaborative format.

Miko Pawlikowski: So are you going to keep the brief history in the appendix, or are you going to

David Clinton: we are updating it. because again, he has a lot of credibility.

Miko Pawlikowski: from your description, there's probably about five people, that said that I could place in that role. I'm curious. But let's leave it at that. I promised, people a little bit earlier that we'd go through a little, tour of the different use cases. for anybody who's going to now go and put the money on Manning's website to receive goods in exchange, the book is very hands-on, it doesn't waste too much time, explaining how these things work under the hood.

Miko Pawlikowski: It goes more into "oh, here's how you play with it, here's what you can do, let your imagination run, go crazy. it will help you become more productive, the way that it did, [00:44:00] to the author". If we were to pick maybe three or five most favorite examples of you that you covered there and that you think, provide the best bang for your buck experience, which ones would you go with?

David Clinton: I'd say the one that excited me the most when I did it was using the tool called AutoGPT. That was an API tool you have the open ai, API tool, key. And you would feed.the tool with a set of detailed instructions.

David Clinton: I want you to go out, research this on the internet, then take the outcome and, distill it this way, and then I want you to draw a conclusion about this and calculate that. You get the detailed instructions, it will go out and search the internet, get information, gather it, come back and iterate over many steps, give you an answer.

David Clinton: That was really exciting. The downside is it was a complete failure. the technology, as far as I know, more often than not, it just goes into these, death spirals. It loops over and over again, the same iteration over and over, it never stops, just [00:45:00] keeps costing you more money. however, the few times I actually did get it to work, I was considering whether to create a course on Udemy on a particular, LPI, Linux Professional Institute,certification that was relatively new, and I wanted to know how popular is that certification?

David Clinton: How likely is it that users are going to want to buy my course, and then take that certification? So I had it, go out and research the popularity based on social media posts and based on Internet evidence, assess the popularity of this technology versus three other technologies, which I knew how popular they were.

David Clinton: I had a good sense of how effective they were. and it spent, 45 minutes or so and about, $2 of my money or something like that, and, came back with a coherent and a numeric answer. I said, "you're on a scale of one to a hundred. How would you rate the popularity of each of these four technologies, or these four certifications?"

David Clinton: And it came back with a good answer. May even have been correct. So I thought that was fantastic. that's going to change the world. it may change the world, but it won't [00:46:00] be AutoGPT that does it. But yes, that's a type of agent that I think has a lot of potential. I'm not sure exactly whether they'll get it right, but it'll, someone's going to get that right.

David Clinton: I have to say I'm not impressed with the AI music and video yet. everyone else seems to be, but,we have 300 years of all the world's music on Spotify available 24/7, right? Do you really think that AI is going to make something better?it may make good stuff, but there are millions of pieces of music that I've never listened to that are probably excellent, probably outstanding, but I just don't need enough time.

David Clinton: So do we really need more music, even if it is good? and what makes good music? I don't know. So AI music, I'm absolutely not impressed with yet.

Miko Pawlikowski: Okay, everybody skip the music chapter.

David Clinton: It was just a music paragraph, actually, if I remember correctly, there's this one paragraph, AI video.?

David Clinton: the OpenAI tool SORA. which is [00:47:00] producing these 30 second clips of video. Also, it's ho hum. I don't know. they're obviously AI. They're nice. They're visually attractive. But, show me something with a little more substance. I'm really not there yet. but,who knows?

David Clinton: Things change. AI voices, by the way. Voice cloning. is something that a number of companies have been offering for quite some time, and OpenAI just announced they have their new voice cloning tool,from 10 seconds of your recorded voice, they can then reproduce your voice reading anything.

David Clinton: No one's shown it to me yet. I'll believe it when I see it. In other words, can they produce voice that won't be recognizably machine?

Miko Pawlikowski: I think that's very high bar because so much goes into this little hints and the tempo and everything. But I do think that there's been a massive improvement in that you can listen to that. you're not going to be [00:48:00] fooled into thinking that it's a real person.but your ears are not going to fall off after at the end of it, there's a lot of websites that integrate that to, have the audio instead of reading through a longer article. And I find myself using that a little bit more than I did before. for example, if I need to do something else, I can just put it down and listen to the article a little bit, if it was a podcast episode. However. That's not gonna fool me. I think that transcription, on the other hand, is getting closer to basically as good as it's gonna get.because, we're using that in a couple of projects. For example, my brother and I, we run a series of technical conferences. And people submit their talks, and it's a lot of,subscriptions to make. We initially tried to get them done by hand, and that was just not gonna happen. we integrated that with Assembly AI right now, and it's about 90, I don't know, [00:49:00] maybe 5 percent good?

Miko Pawlikowski: To the point that, the things that it doesn't get very well sometimes is things like names of things. But it's good enough that mostly you can ignore the mistakes at this stage. So I think that's going pretty good. on the image and video thing, I think I would agree that people started talking about how Hollywood was going to fall because SORA could generate 30 seconds of a coherent image.

Miko Pawlikowski: There was a lot of excitement about that. There was a project called Pika Labs, I think, before that, and that also could generate 30 seconds or something like that. Some of them looked funny. Some of them looked like a very low budget animation from the 2000s, and some of them just looked rubbish.

Miko Pawlikowski: but the one, one exception to that, I think is I remember when I went back to using midjourney after I initially tried it and wasn't impressed and it just started. lIke 'killing it'. [00:50:00] I think it might have been about a year ago or something like that. I went back and I started generating some images.

Miko Pawlikowski: I was just so impressed how good they were. they had all the fingers and no additional

David Clinton: right number of fingers,

Miko Pawlikowski: Yeah, and the faces and, and people started coming up with this tricks on how to generate like consistent faces. If you wanted to generate multiple pictures of your character, I think it comes out of the box with that now. for everybody I talked to replaced stock images. Instead of going and paying for a stock image, you can get a custom one and you can ask it to tweak it until you like it. And so for this like going back to what you said, generative AI versus, self driving cars, when the stakes are lower and you can accept some problems, it's working well enough that it's just replacing that.In terms of music, that's a little awkward. All of the [00:51:00] music,for this podcast is AI generated.

?

David Clinton: It can be good, I probably won't listen to the music of this podcast hours and hours on

Miko Pawlikowski: Oh yeah. It's about 15 seconds, so it should be okay. But

David Clinton: for that purpose it's great, butI don't think it's going to replace people's playlists,

Miko Pawlikowski: no, I don't think so. Unless you're talking about the actual playlist, algorithm, because I think Spotify has been very loud about how they're going to replace your playlist and basically get AI to find the songs that you, they think you're going to like. Facebook or whatever social media is keeping you algorithmically,hooked to that.

Miko Pawlikowski: And that, I believe they might be onto something. There's probably,an optimal algorithm that can get you hooked and, addicted, to as many songs as possible. there's also that. for anybody else who wants to, get the nitty gritty of how to, actually play with all these things and test it for yourselves, there is a sequence of chapters [00:52:00] on all of these things and more, including as well as what we covered, things like code and, talking to documents. I think there is a little example where David is getting some kind of AI to talk to a PDF containing Tesla's financial statement and it's actually working very well. So there's that.?

David Clinton: it's gotten a lot better since, especially with the GPT+, a lot of barriers have fallen, with my writing from the audit side, I'm able to do things. in an hour that a traditional journalist would have taken a week or two if you could do it at all.

David Clinton: But I'm doing it because of AI's, and GPT's incredible processing powers.

Miko Pawlikowski: yeah, thank you gamers. If it wasn't for you, we have never gotten Nvidia to make all those GPUs.?

David Clinton: as you can probably imagine, I'm not going to let you off the hook before I try to squeeze your wild prediction about [00:53:00] things like AGI and things like regulation and things like quantum computing and where you think that is going. So maybe let's just bite the bullet and go there.AGI.

David Clinton: I'm not sure exactly how we'll measure it, because there's so many things that GPT, for instance, does so much better than I could ever do. and faster. Big Blue, beat Kasparov in chess, right? So that was one discrete task, which is a very complicated task. That's already 20 years old or something in that area.

David Clinton: but there are many things that GPT can do infinitely faster than I can. is that AGI yet? I'm not sure how we define it. I think it's, inevitable. My humble opinion is that AGI, if not now, certainly in the next couple of years, I would imagine is inevitable.

David Clinton: Is it a good thing or a bad thing? Personally, I don't stay up at night worrying about the AGI apocalypse. that the, AI is going to,push off the shackles of humanity and take over and destroy us all.

David Clinton: But, hey, [00:54:00] who knows? In any case, betting against it's not going to make any money because you won't be here to collect. so as far as AGI, I think it's inevitable, if not now, soon. as far as, quantum computing, I don't, I've read a lot about, I've read some about quantum computing.

David Clinton: I know that, I have a very vague understanding of what makes it different, and how it'll be so compute intensive.it could be that quantum computing will solve the energy problem that AI is facing. Maybe. That is, Platforms, like OpenAI, require thousands of CPUs chugging away in data centers, and some of them require their own nuclear energy generatorsjust to keep them alive.

David Clinton: There isn't enough electricity in the world being generated to keep this stuff running. And, of course, they're adding on, the electric cars,that are all supposed to be fed from the electrical grid. And there isn't enough electricity.

David Clinton: And the chairman of Toyota and Elon Musk also both warned that if every car [00:55:00] sold in 2030 is an EV, we are not going to have enough electricity to run it and to power your shaver, there isn't enough, so we're coming to a crisis probably soon.

David Clinton: quantum computing, if it scales up fast enough, might be more efficient from an energy perspective and might head off that crisis that they'll still be able to exponentially grow the compute base of these platforms and have enough energy to feed them.

David Clinton: Possibly. That's, it's a complete guess anddon't take it all seriously, but it's possible. the downside, the real downside of quantum computing, of course, is that all our passwords will be assumed to be broken. Because the, the RSA algorithm will probably be cracked fairly quickly once quantum computing becomes commercial.

David Clinton: if someone had, penetrated your, private website or your private, data, 2-3 years ago and saved the data archives, unable to get to it because it's encrypted because of course you use good passwords.

David Clinton: but once the quantum computing allows us to crack the RSA algorithm, they [00:56:00] can retroactively access everything that they've saved, because it's all behind a, an encryption that can be broken that it has or will have been broken. Okay,that's a possibility. As far as regulation goes, as I wrote in the book, I believe, this can't be regulated.

David Clinton: governments can jump up and down and kick and, scream as they always do. and pass legislation that, severely restricts the, developments in AI. It's not gonna change much, because it's out in the open. It's out in the wild. People will do with it what they want.

David Clinton: The market will do with it what it wants. some have argued, as you probably read, OpenAI's requests for government regulation might just be, a subtle attempt at, at locking in the market. since it's on top, once the market is regulated, no one else will be able to enter the market once they're, on top, they'll stay on top.

David Clinton: So some have interpreted it that way, as I'm sure you probably read. but I don't think it's practical. Some things can't be regulated.

Miko Pawlikowski: it wouldn't be the first time. I remember a certain Zuckerberg, pleading a very similar case?

Miko Pawlikowski: Fai [00:57:00] enough. I wanted to do one more thing before I let you off the hook. I found your chapter, I think it was chapter 11 where you presumably got some friends,who happened to be experts in, in AI and you asked them a series of interesting questions like, "what were the problems", "what were the biggest surprises", what do you actually use the most, and all of that, what were the biggest unexpected items that you sell them?

Miko Pawlikowski: you know, you obviously went with some presumptions of what you were going to see there. I'm curious what you got that you didn't sign up for.

David Clinton: That's an interesting question. And sadly, I can't remember the details well enough. I appreciate it. I do. The answers were great. And I was, it added a lot to the book, I think. But sadly, I can't remember them in detail. I didn't, it was, I have to say that I was expecting [00:58:00] some more embarrassing, step on a rake incidents than I got.

David Clinton: Cause I've had plenty of step on a rake incidents myself. I'll tell you an example. I'll come out in the open and tell you one of my,I guess you can call it a failure, but it was an experiment. Early on when possibly before GPT-3 came out, II started with OpenAI.

David Clinton: I believe it was still GPT-2. I'm not sure they called it that, but I don't remember now, but it was before GPT-3. right at the very beginning, I I experimented to see how it would, generate written content, technology content. So II conversed with it back and forth and got from it, the content for five books.

David Clinton: One was on, low code. another was on, a couple of security topics. I don't even remember. And I assembled all the content andI edited it to, to give it some coherence and give it a narrative and a flow and I published them. And I,I just want to see what would happen. Can I effectively [00:59:00] in two hours, I went from zero to up on Amazon Kindle.

David Clinton: I wanted to see if there was a business model. so first of all, I'm not the only one who had that thought, right? So Amazon is now flooded with literally millions of such books and most of them are trash, and most of them you can tell pretty quickly are trash and people avoid them and the market is correcting for that.

David Clinton: Amazon asks you now, whenever you upload a new book to Kindle, they ask you, was this, or. was all or most of this book generated by, an AI. Sothey want to know. They're not penalizing you yet, but they want to know, and it's a fair question. these four or five books, I did publish them.

David Clinton: I'm not embarrassed by them, because I think the content is useful and helpful. I think it's genuinely helpful content. I read through it all, and I don't think there are any mistakes there. I don't think there's anything misleading there. But it's cheap. It's not something I'm particularly proud of. But I did it, it was an experiment, and I, I think the whole thing took me two [01:00:00] weeks, all five books.

Miko Pawlikowski: Tempting and interesting. So they ask whether 50% or more of the book was generated.

David Clinton: Something like that, it was a substantial amount of the book generated by AI, maybe the word they use is substantial, I don't remember.

Miko Pawlikowski: they're facing this problem, they have to figure out a way to deal with it. Their platform is being flooded with this garbage.

Miko Pawlikowski: Yeah. Although even with, legitimate books, I would expect that very close to 100% has some percentage that was at least rephrased?

David Clinton: You have to be crazy not to. it can make you so much more efficient and more accurate, as long as you use it carefully, and as long as you don't abuse it. The sum of my book,"the, obsolete guide to generative AI", 10% maybe? 20%? I don't know. is, is AI generated? I

Miko Pawlikowski: a good, path to do a newer revisions. You just, regenerate the book with a newer AI.

David Clinton: that's possible. actually, Manning and I were talking about that,a lot of their authors are not English as a first language. There are [01:01:00] foreign language speakers, primarily great technologists, good authors, but English is not their first language.

David Clinton: So they have a lot of problems sometimes getting a manuscript from, the author's computer to a publishable state in English, and they'll invest a lot of money in that sometimes. So they wondered, and we were playing around with the idea of feeding chapter by chapter. Into, GPT and then getting it to revise the book, to rewrite the chapter, I don't know if they ever used it.

David Clinton: I gave them a couple of, proof of concept experiments. I'm not sure they ever actually applied it or not, but it's an interesting idea, certainly.

Miko Pawlikowski: at this rate, we're just going to end up with, everything touched or rephrased or rewritten by AI,Going back to what you said about SEO, there are so many people really banking on spitting out so much content, generating entire websites just to capture some of the traffic with AI, which sadly means thatthe era of someone having a [01:02:00] website about something that they're interested in andthat being, found is just gone by now, I think.

David Clinton: how that's going to fall out. I don't know what's going to happen To the accessibility of the internet, that I don't want to predict.

Miko Pawlikowski: yeah, it's weird, when I speak to different people, I feel very different. for example, I was speaking to a friend and, about how people rewrite their text with ChatGPT and technically they wrote it, but not really. yeah, it's generated by AI. Almost exactly what you were describing about getting a chapter and getting it to be more English first language.

Miko Pawlikowski: and, the argument was,would you also tell me to not use autocorrect or like a dictionary

Miko Pawlikowski: that's cheating, which kind of shut me up. And I've used that argument since on the other side, and that made me feel, "okay, that's, that's probably a silly argument. But then on the other hand.

Miko Pawlikowski: I [01:03:00] wonder if people who sell cars and they were like, Oh, no, we like our horses and what's going to happen to all these horses. And no, this is cheating. It runs on petrol. It doesn't even have to stop and drink water. Whether I'm one of those people, sometimes when I speak to my friends,it's weird times we're in, isn't it?

David Clinton: Yeah. thinking back,10 years to how workday went and the tools I'd use and my first book I wrote, 35 years ago.was, probably before you were born, possibly. I wrote it as I mentioned somewhere, I wrote it on paper with a pen.

David Clinton: And, you write differently. When I wrote, on a typewriter, you write differently, cause there's no correct, now forget about autocorrect, there's no correct at all. See you're stuck with the version you have, okay, you could use that,what was it called? you probably don't remember.

David Clinton: They used to have this. Correcto type, I think it was called a little tool. You could add a little white ink over a mistake in a typewriter. it was kludgy [01:04:00] and you just had to get it right the first time. I've seen it round the time that word processes became popular 25 years ago,?

David Clinton: some editors, one editor in particular said that she can always tell a manuscript that was written on a computer. Because there'll be far more stupid mistakes. On a typewriter, when you're nervous and you can't correct it,you think about it very carefully before you type.

David Clinton: On a computer, you, nah, just go back and get it later. And you usually don't. that's from the typewriter to the, word processor stage. But from the word processor stage to, AI, the changes are going to be, are going to be fundamental and remarkable. Heh

Miko Pawlikowski: who, wants to go and jump the book, once again, it's called "the complete obsolete guide to generative AI", it's available at manning.com and it's got a lot of humor. And if you don't mind, I'm just going to quote you on what you just described. The book says. Let me give you some context: I'm a lot older than you might think. I wrote my first book on sheets of paper using a pen. [01:05:00] It may have been a very old technology, but it was solar-powered. Translation: it was only useful when I opened the window or turned the lights on. That's the kind of humor that you're going to get in droves, when you read through, David's book, David, thank you so much.

Miko Pawlikowski: This was absolute pleasure. Hopefully, next time we speak, we are both not replaced by AIs, or maybe we are, and it's actually going to be a better conversation.

David Clinton: Or maybe we are already. How do you know that these aren't, avatars, and deep fakes? But, I think you're real, and I'm fairly confident I'm real. and I, as a real person, I enjoyed this, this conversation, and I'mlooking forward to seeing where it goes.

Miko Pawlikowski: and so did I, at least in this version of the simulation. Thank you. See you next time.

David Clinton: Take care. Thanks a lot. Thanks.