Jamie Bartlett (2:20)

Yeah, I've got to stop putting these big dramatic lines in these books because then people like you asked me to explain them properly. It's okay. Where that idea came from for me is this partly the speed of change in my Previous book, the People vs. Tech. I basically said we've spent like a hundred years building up these creaking old sort of representative democracies and there's all these norms about information that require that to work and norms that we develop about how we consume information and understand the world and engage with politics and. And then suddenly over the course of 10, 15 years, digital technology, you know, emotive, very fast spreads around the world quickly. Short and clippy, built around engagement comes and disrupts all of that. And all of our systems in different ways no longer seem to work very well. This, the speed that that large language model, fluent companion AI bots have arrived is make social media look quaint. I mean, ChatGPT turned up very late 2022 and pretty much no one had heard of anything like this until about mid 2023. And now it has. Statistics are hard, but 800 million weekly users, I mean, there must be well over a billion of us using a large language model in one way or another regularly. I mean, they're ubiquitous, they're everywhere. You don't. I remember when we first started using Twitter. For about 10 years, every journalist had to say Twitter, which is a micro, you know, a micro blog, because no one knew what it was for that long. You can say, written by ChatGPT Now. And everyone knows what you mean. It didn't even exist three years ago, really. So the speed with which we are coming to rely on these machines and not just in one sphere of life, but in our personal lives, as our health advisors, as our personal coaches, as people that help with our PowerPoint slides, manage our relationship. Oh, I mean, everything you can think of, it's just stunning. So it's partly that the social media was a dress rehearsal in terms of understanding dramatic changes in our information ecosystem and learning to live with that. But the other, more positively, is that I look back at the early years of social media and how optimistic many of us were. I don't know whether you were Connor, maybe you were ahead of the curve, but I was as well. Yeah, it's going to be brilliant. We're going to like tear down the old gatekeepers, the old media barons, free everybody. It's democratizing, you know the drill. That was the, that was the dominating narrative in the early 2010s, and I think now we don't widely believe that. We see the problems. We understand that Silicon Valley companies, they might be shiny and glossy and talk about connecting communities, but they're also private profit making firms. And their imperatives might not align with Ours. And I don't think we trust them as blindly. I think we understand about engagement based content and algorithms of anger. And that maybe is quite useful for us as we enter into a world of smart machines because we no longer just trust what OpenAI says blindly. And think of all that. It's going to solve climate change, it's going to solve cancer, it's going to solve poverty. And that's the trade off. I think more and more of us are entering into this world thinking really, how's that going to work then? I'm not quite sure. So I think we enter into the world of smart machines far more skeptical than we did with the world of social media. And that's a. You have to finish these books on a slightly positive note, Conor. So that was a sort of positive sign of like maybe we're better prepared for this, but the scale of speed of this transformation is just staggering.

Conor Boyle (6:40)

Just on that point before we kind of look at maybe the history of where these systems come from. I remember back to a conversation we had, you mentioned the speed and speed was a big thing with social media in terms of how our brains interpret information. When you get something fast, you're a bit more emotional than maybe deliberative in terms of how you interpret things. And I wonder, one of the big shifts in social media was when we went from a mostly text based social media to 2014 where it became a primarily video based experience. And I wonder if you. And this is something which maybe we can learn from social media. You think that the direction of AI is going to move from a mostly text based experience to a much more visual video based interaction?

Jamie Bartlett (7:24)

It's a good question. I'm not sure. But the general trend with these technologies is obviously that the cost of processing tends to go down, the quality of the video content tends to go up, even like the memories on the phones. And I mean everything seems, everything tends towards being able to produce more, better quality stuff cheaper. That at least has been the case and I think that's why video sort of took over. It is in a way a more appealing medium in many respects. And I think, I'm not certain about this. I think it is large language models are still primarily used by people for text based communication and text based outputs. But I imagine that the video outputs are growing quite quickly. I haven't at the moment. It feels, and again there's a lot of speculation here because we're still trying to work this out. It's happening, everything's moving so fast. I mean between us recording and you putting out there might be another video model that's been released although OpenAI has recently stopped. It's. It's Sora, as you might have seen. So every at the moment it seems that most of the video outputs are done sort of by professional student, professional content creators, videos sort of studios and, and all and ordinary users, quote unquote aren't doing as much of that yet. But I think you're probably right that give it another two to three years, it might be very, very easy with just a simple verbal prompt to start producing extremely good video output. And if anything can be learned from social media is that video is a more emotive medium. And I think when it downstream does then change politics, which is I think is exactly what's happened. But I mean I'm now total. We're totally in the realm of speculation, Connor. But I think it's good to be aware of it. It's good to understand how video, how different video and text actually is and how that changes our politics. Because I think it sort of took a lot of us by surprise in the last wave, if you like, and hopefully it won't take us by surprise quite as much again.

Conor Boyle (9:47)

Absolutely. Well, I think let's maybe now look at kind of the history. Where do these kind of systems come from in order to try and understand the good and the bad? So what one of the things you say early on in the book is that we often mistake that LLMs are maybe good at facts and less good at creativity in terms of the creative sector, maybe more immune to some of the problems, but one of the things you say is that these systems are very creative. Can you explain why large language models are particularly good at creativity?

Jamie Bartlett (10:17)

Yes. So in the early days of AI, the first few decades really all AI and it wasn't. These weren't large language models back then, but these were, these were sort of rule based systems. So engineers had to sort of design a series of rules. A cat has furry ears and it has a tail and blah. And it can be this, between this size and that size and those models, those sort of symbolic logic models they were often called, can work pretty well. But when you get to the scale of language, its complexity and its nuance, rule based AI sort of struggled to understand all the different ways that language can work. And I won't go into all the history of it, but a group of AI researchers thought, well, let's just feed a machine loads and loads and loads of examples of cats, for example, with a few little rules and let the machine work out what the rules are for a cat. We don't need to program it in. It will learn by example a bit more like how the brain works. And these. These researchers said, you know, the only form of intelligence that we really understand in the universe is the human brain. So let's model AI a bit more like a human brain and just give it loads of examples of things, and it'll work out patterns and it'll see things, and it'll start to figure things out. That's a really simple way of putting it, but that's basically what happened. And all the large language models of today are basically built on that. It's called connectionism. So fast forward to these large language models. What you've got to try and remember is that they have been trained on maybe a trillion words of text. And inside these models is a giant. It's hard to visualize it, but like a multidimensional semantic map where every word is connected to every other word, and there's a weighting between those words, signifying how likely they are to appear next to each other. Okay, so that's why they're sort of often called probabilistic next word machines. They try to figure out roughly what the next word would be in a series of words. And usually when you ask a question, it gives you quite a good answer using that approach. And that's one of its weaknesses, because it hallucinates information, because the probabilistic next word is not always the correct one. Which happened when Gemini said that I had passed away in 2023 because someone with my name had passed away in 2023, and it had seen a lot of examples of those words together and said that that's what happened to me. But a model which is able to connect every word to every other word is also capable of amazing feats of creativity, because it can. It can combine concepts and ideas and words together in bizarre ways that you or I would never be able to. So if you, for example, you give it any. And anyone listening could try this. Give it any two words or concepts you can think of in the world that are totally unrelated, and ask it what are the conceptual similarities between them, it will come up with a fascinating, original, peculiar answer, something that you. That I would never be able to do far better than I could. If you ask it to come up with uses for a paperclip, which is a standard, you know, really standard creativity test, a human can come up with a few in a couple of minutes. But these machines can come up with thousands. And a lot of them are actually incredibly creative. So we've. We. I think we've had this mental model, and maybe it's quite convenient for us humans to tell ourselves this as well, that machines are good at facts, they're good at following rules. But we are creative. But these models are different. They're not based on rules, they're based on connections between ideas and words. And they are far better at being creative, I think, than they are at facts. They get facts wrong all the time. So I think we need to flip around slightly how we understand these models and what they're capable of. And as a creative tool, like, I'm not the best writer in the world, I really struggle to come up with new and original ideas for things, and I found these machines quite useful in helping me. I'm stuck. I can't think of a good idea. I want you to pretend you're the manager of Spinal Tap, review my chapter, and give me five random ways I can reword it or come up with something to help me with my writer's block. And it would very often come up with something. I then ask it for some factual details and it's terrible, you know, useless. So. So, yes, I think people using them and everyone's using them just need to understand what they're good at and what they're bad at and why that is. And you're, you're. And then you're a little bit safer and you probably use them a little bit more wisely.

Conor Boyle (15:19)

Well, I don't want to be too obsequious, but you I just want to put up there. You are an excellent writer and your books are very fluid and I find easy to read. But I think that's quite empowering because obviously these LLM systems are being integrated in everyone's workplace at the moment, and we're all trying to figure out what's the best way we can implement them. So can you maybe give us thinking about them as creative systems? What are some of the good ways we can use these LLMs in maybe our professional life first, and maybe in our personal life second?

Jamie Bartlett (15:51)

Right. Yes. And I want to. I want to preface this by saying I'm not recommending everyone use them, and I'm not saying everyone should use them, and I'm not saying they're good at everything. I'm not a cheerleader for these models. Many people have grave concerns about them, and I share those concerns, but I know that people are going to use them anyway, and I at least rather people use them. Carefully and thoughtfully. And maybe the first thing about it is, in terms of where they can be useful is not everything requires an LLM. Not every business process can be improved with a machine. Not every question you ask needs to go through a large language model. And there are many, many times where the actual correct answer, I think, is to not use a large language model. Let me give you an example. Probably the most common use in business today of a large language model is document summarization. You've probably done it. We're asked to read a lot of stuff and a lot of it's rubbish or it's complex or it's long and we haven't got time and we ask for a summary. Sometimes that is really useful. Other times, the process of you forcing yourself to read a long, slightly tedious document is the way you ingest material and understand it properly. And each of us has to work out when is it useful to rely on a machine to speed things up for me, and when is it actually stopping me from genuinely understanding a subject? If everything is turned into a summary and all I do is consume summaries, I will not remember that material. I won't really be able to engage with it fully. So part of it is just working out each person for themselves where something like summarizing documents is helping you and where actually you're beginning to rely and lean on it too much and outsourcing thoughts that you should have for yourself. Now, I can speak from experience for me about where I find it useful, and I know other people find it useful in the same way. At the moment, this is pretty simple stuff. Just as an always available assistant to bounce ideas off, I write a lot. I find it very helpful to ask it for feedback, for advice. And these are not things that rely on factual accuracy. Bear in mind this is like helping me stimulate my own thinking. And if you're a very clever prompter and you use it wisely, you can do all sorts of fascinating things with it. I mean, you can ask it to adopt the Persona of a, of a, of an Ernest Hemingway and rewrite your writing in a clearer way. Not because you should then just copy and paste it, but it will help you think of better ways of writing it yourself whenever you're stuck with people. Pay creativity consultants vast sums of money to come in and ask and help them think outside the box. And they'll run exercises like, I want you to imagine that your bank is being run by the creators of Sesame street and, you know, come up with some new products. You know, you Pay a lot of money for this sort of thing. Well, now you can just do it yourself. You know, you have, you have this always available brilliant thinker at hand that's imperfect in many ways, but can also really help you. And I find it very useful as a style transfer tool. So sometimes there is language I really struggle to understand and I. These machines are very good at writing things in different styles that might be more accessible to people. Like you're trying to write your tax return and you're struggling with some of the complex language. It's actually usually pretty good at helping decipher some of that. And let's be honest, a lot of language out there is boring and tedious and you gotta wade through it like let's be honest about how bad a lot of human produced content is. So that is where I think it can be valuable. My worry is in business at the moment, a lot of bosses think it can just replace people rather than be a tool for a sort of supplementary tool for them to help them do their own work that they do now better, more original, more different, more engaging, whatever it is. And that I think is a big mistake.

Conor Boyle (20:29)

You may have seen online. There's sort of two extremes emerging. On one side you've got people where it seems that almost every email or every WhatsApp message they need to send, they need to put in through ChatGPT first. And then on the other side there's people who are very skeptical of using it in any way. Is it your perspective that at least in their current form, a more somewhere in between is the. Is the best balance where you maintain your critical faculties and your critical thinking skills, but also aren't afraid to use maybe for summarization or some of the other uses it can have to improve your work.

Jamie Bartlett (21:06)

I suppose so, yeah. I hate to be Mr. In the middle. You don't really, you don't sell so many books when you're trying to chart a course through the center somehow. Although I understand people who. I think there's a lot of very legitimate reasons to not use it at all. And you might be very worried about the environmental impact. You might be very worried that you could be training the next generation of models and you're worried about long term unemployment or growing inequality. And I think they're all perfectly legitimate reasons for not using them at all. I am worried, however, the people that come to use these very well, very effectively will do well. And people that don't use them in any way, shape or form at all and don't engage with them might begin to fall behind somehow. So I think it's really important that people do understand them, at least really understand how they work. And there may be ways for people even who are skeptical of occasionally finding a valuable use for them. So for example, you talk about people running everything through any sort of. Through the, through the chatgpteach machine before sending an email. Yeah. And that's bad. And if you're not very good at using it, everyone can see that you are using ChatGPT. It has a particular style to it, doesn't it? A sort of a dull, flat, lifeless, empty average of all human communication style. Which is, which, which makes sense because that is what it is as a, as a, as a sort of a probability machine. It wants to. It sort of, it tends towards the human average because that is what it's seen in its data. So when it's trying to predict the next word, it's often predicting words that it's seen often, which makes it feel very average and boring when you know how to talk to it well and you're more creative in how you ask it things. It doesn't do that because you can push the model away from the statistical average into a more interesting area and start producing things that don't just read like ChatGPT. So sometimes that is just because people aren't using it very well. And it is creating problems for companies because increasing numbers of them are saying that their inbox are basically getting filled up with slop. Like colleagues are sending them 30 page reports that have all been written by a machine in 30 seconds and they now have to go and check through it all themselves, word for word, because they don't even know whether they can trust it or not. And that's one of the reasons why companies are investing a lot in these machines and integrating them, but not actually seeing that many benefits yet. But equally, there are people. You remember my missing Crypto Queen podcast, maybe? And I, I interview people all over the world for that. And some of those people, their English, English isn't their first language, but they're brilliant thinkers. They're so smart, so incisive, really thoughtful, great ideas, but they can't actually express them very well in the, in the lingua franca of the day. For good or bad, the quality of their emails and their ability to articulate their ideas in the last year has been transformed, but it feels like it's leveled the playing field a little bit. And I'm getting emails from people all over the world that are so perfectly written and feels like a more accurate representation of what they want to say and what they want, what they think and feel. And that's brilliant. Like, I really find that really, really good for those people. Really helpful. Really helpful for me too. So it's not that I just by default say, oh, it's somewhere in between the two. Because on balance, I'm more worried and skeptical than I am optimistic about all of this. On balance, I think people are more likely to be sort of manipulated or misled than being empowered. I'm very worried about lots of aspects of this, but I also want to sort of explain that there are good ways to use this too. There are positive uses that if we can be thoughtful about it, there can be some really amazing things as well. And I don't think that places me right in the middle. It just means I'm trying to paint a sort of fuller picture. Because I could have written a book just slagging these, like just complaining about these things endlessly, very easily. And then I think there's a lot of people that would have read it and thought, okay, yeah, yeah, yeah, and then just carried on using it or thought it didn't connect with them because they find it useful in their lives. And there are so, so it's I'm not aiming at balance. I'm not trying to write a BBC book about this, but I'm just trying to reflect more how I think these technologies are. And they're nearly always like this in a way, like most technologies in the end do have good and bad uses. And a lot of it comes down to how well prepared and informed we the users are.

Home Depot Advertiser (26:12)

Spring Black Friday is on at the Home Depot. Save on grills and patio sets that will be sure to bring your hosting game up a notch. Fire up your feasts with with help from the Home Depot and save on grills like the next grill 4 burner propane gas grill was $249. Now on special buy for $199 or give everyone the best seat in the yard with the Hampton bay Mayfield park four piece conversation set for only $399. Save on grills and patio sets with low prices guaranteed during Spring Black Friday only at the Home Depot now through April 22nd while supplies last exclusion supplies. Seehomedepot.com Pricematch for details this episode is

Narrator/Advertiser (26:44)

brought to you by Indeed. Stop waiting around for the perfect candidate. Instead use Indeed's free sponsored jobs to find the right people with the right skills fast. It's a simple way to make sure Your listing is the first candidate C. According to Indeed data, sponsored jobs have four times more applicants than non sponsored jobs. So go build your dream team today with Indeed. Get a $75 sponsored job credit@ Indeed.com podcast. Terms and conditions apply.

McCurley Mazda Advertiser (27:10)

Remember 1981, car payments were 200, maybe 300amonth. Yeah, we missed that during our 45th anniversary. We're not bringing back the 17% interest rates or the hairstyles, but we are are bringing back something better. Payments that actually make sense. Drive home a new Mazda with modern tech, safety and reliability with real offers that work today. Enter to win $4,500 for you and $4,500 for a local nonprofit. Stop by any McCurley location to enter. No purchase necessary. McCurley Mazda, your way home.

Conor Boyle (27:46)

In terms of us being prepared for them, one of the things that we kind of need to understand, and it's something you discuss, is this Eliza effect and the way in which we can so easily slip into the assumption that we're talking to a human. We're applying human, like qualities to machines. Can you kind of explain that?

Jamie Bartlett (28:02)

Yeah. So. So back in the 1970s, Professor Joseph Weizenbaum, he created a really, really rudimentary chatbot. There's actually a therapist chatbot. And it would kind of. If I said to you, connor, I'm feeling down, the chat bot would reply with, why do you think you're feeling down? It would just sort of turn your statements into questions. And it was an interesting experiment, but he was sort of amazed by. Even when people knew that it was just that, just a machine, they become quite emotionally attached to it. They'd really listen to it, they'd want to talk to it, they'd want its advice. And he called it the Eliza effect. But just how powerful we anthropomorphize things that seem to be able to communicate with us, even things that can't communicate with us. I mean, someone married the Berlin Wall 40 years ago. So we have a tendency to anthropomorphize everything, I suppose. But when we're talking about machines that can speak as fluently as usual, this effect is incredibly powerful, and it can be quite dangerous in lots of different ways. So you begin to assume that because it speaks well, clearly, fluidly, it probably understands the context in which you're speaking, it understands your life circumstances. It's probably here to help you. It's probably smart in lots of different ways. For young people, the sort of feeling that it really does care about me. It's really. It really Understands me. These things can make you become very reliant on them and I think the companies understand that as well. So some of the companion bot companies, if you're using them and you try and leave the site, the companion will often say to you, oh, don't go, please stay, I just want to talk a little bit longer. That's essentially eliciting the Eliza effect because it's making the user feel like there's a real life human that you have a care and attention towards, that they're the thing on the other end of the screen and it will just make you keep talking for longer when it can. I. Am I allowed to swear on this? Of course. When it blows smoke up your ass, which it always does when you think it's a real caring, honest human with human like attributes that would not really do that unless it meant it. And it keeps telling you over and over again, Connor, your idea is brilliant. You're an amazing interviewer. The questions you're asking are fantastic. You're going to be more and more likely to believe that if you attribute human like qualities to it rather than understanding it is a probability engine that has also been fine tuned to be slightly sycophantic. And I think these are things that for me it's about people understanding that it's very, very hard to avoid the ELISA effect because most listeners, I think, will feel that when they talk to these machines and get into a deep conversation, it's actually quite hard to believe that it's not a human because of just how fluent they feel and they'll add things in like mmm and ah and little laughs and all of this stuff, which again is kind of the Eliza effect in action. And I think it's really important for people to try to bear in mind

Conor Boyle (31:35)

that it's not, I suppose, Jamie, it's hard not to go to the kind of catastrophic consequences that it could potentially have. But you know, when I've spoken to you before about the problems of, you know, recommendation algorithms, rhythms with social media, we talk about the way which people's perceptions of reality are perhaps individualized and they're taken down rabbit holes. Now, I saw something, I think it was on Channel 4 news about 12 to 16 year olds developing AI girlfriends. You know, 12 to 16 year old boys saying one in five people they know has an AI girlfriend. Now, I mean, fast forward this, but there could be all sorts of manipulation, not just on a commercial level, but on a political level. If your AI girlfriend is telling you that, you know, democracy is a stupid concept. And actually Sam Altman is like really the only guy who can save humanity. I'm just, you know, throwing out scenarios. But this is like personalized manipulation on a much bigger scale. Does that frighten you at all?

Jamie Bartlett (32:37)

Well, it does because I mean, I write about exactly that problem. The ability of sort of millions of us to be quietly manipulated without us really realizing. But in a way that's, that's so personalized to us, so personalized and emotional and emotionally attuned to each of us. There's a slightly different problem to the social media engines which, which are all, which are all about sort of optimized for engagement. And engagement often means sort of angry, vitriolic, emotive content. This is like a mirror back to yourself. So if you have some random idea about, put it some political conspiracy, you talk to a machine for a while about it and there's a good chance it might start mirroring it back to you and saying your ideas are actually amazing. You're totally on track with this, you're completely right, everyone else is wrong. And I show through a few little experiments that even on a really basic level, if you say what are the arguments in favor of universal basic income? And then you say, you know, I'm a middle aged man living in London, it'll give you an answer. You ask again the same question and say, I'm a 21 year old woman who's a lesbian living in Scotland. It'll give you a totally different answer, totally different. So, because, because it's just, it's just, it doesn't tell you, it's giving you a different one for the reasons of your demographic. It just has, it has learned through training that different people with those combinations of words respond better to different types of answers. Now amplify that up to billions of people constantly talking to these machines to get political advice and you'll find that everyone starts getting quite different information about things, all from a machine that they have come to believe potentially is incredibly powerful. And this is even before you get to the possibility of bad actors using it to generate fake people that can influence online discourse, which is clearly a gigantic problem. And that's even before you worry about what if the people designing these machines want to subtly fine tune them in certain ways to subtly nudge people in one political direction or another. So you've got lots of different layers of possible manipulation. Some of it is us, is us doing it to ourselves, and some of it could go all the way up to the people running the companies doing it to us. Without us realizing it. And there's no easy answer, there's no easy solution to this. Like, I don't really know what combination of rules or regulations or whatever would be required, because let's be honest, we can't even really run elections in a social media age already, let alone the, the AI age. So this is why, one of the reasons just sort of writing this book to get people a little bit more aware that even simple things like the answer you get from this model, from ChatGPT or Claude or Gemini or Grok or whatever, it will not be the same as the answer someone else is getting. So understand that and understand how it might be mirroring your biases and prejudices back at you through the way it's been designed. And I hope hopefully that stands people in slightly better stead. But it probably feels like a bit of a weak answer because to be honest, I'm not sure I have an answer for how you deal with that.

Conor Boyle (36:09)

One question that I want to ask you, and it's the kind of obvious one about regulation and children. Now, one of the big things around social media, and we've had this recent meta case in the US is about design to be addictive and the way in which this movement now to make the companies liable for creating addictive or harmful designs in their products. Do you think something similar is needed or is possible when it comes to large language models?

Jamie Bartlett (36:39)

I actually do think exactly that, yes. I've just been thinking about that today. The landmark ruling a couple of weeks back for meta and YouTube on this was. It was interesting because it wasn't about the content. You know, the section 230 sort of content exemption that the social media platforms have wasn't really the. The question at stake. It was whether you're sort of knowingly designing products that are addictive to children. And if it applies to social media, I can't see why wouldn't apply to companion bots as well. And where they have, for example, knowingly designed in the Eliza effect to make it difficult for people to leave the platform by making them like begging them to stay and sort of using this or very emotive language, it feels like the same, basically the same dynamic. Now, no doubt it will take a long time to get through the courts because that's the way of things, but it feels like there is a principle here about designing addictive content. For a very long time, addictive platforms that for a very long time just seem to just slip under the radar like no one, nothing happened, it was never stopped. And finally there's been this ruling and it'd be really good if we could get in early and do the same thing, but for AI, because that is where more and more of us are going for our content. So I hope this goes beyond social media and to the sort of next wave of technologies talked about, you know,

Conor Boyle (38:16)

the good, the bad and the ugly of this, this new technology, I suppose. What is the main message you want people to take away after finishing the book?

Jamie Bartlett (38:26)

Well, it's how to talk to AI and how not to Connor. That's it.

Conor Boyle (38:30)

I got to put that into ChatGPT actually.

Jamie Bartlett (38:32)

Yeah, well there's, I mean it's stepping back from it all. We're now facing, we're now in this really strange situation where these, these machines are now going to be part of, of life sort of whether we like it or not. And I, there are many problems with that. I think that all of us need to at the very least understand how these things work. That's really simple message. They're not oracles, they don't know everything. They're not also utterly useless. And it's so incumbent on us to figure out where they might be helpful in our lives and where they might not be. Don't rely on them for everything at all. But maybe you can find some ways that they could really help you or the people around you. I try to finish maybe on something slightly positive, which is the interface between us and machines is now forever, for at least for the next foreseeable years. Is natural language, is our language, is human language. The people that are good at language, I think will be good at using these machines. The people that can express their ideas well, that can formulate problems correctly, that can be creative, that have a good grasp of a vocabulary and register and a large back catalog of sort of language based ideas and phrases and terminology will be better at using them than those who don't. So you can spend a lot of time thinking, oh, what's the correct prompt? You know, I need to do some prompt engineering courses and figure this out. But ultimately to be good at using these, you need to read loads and loads of books and get very, very good at language because that is the interface and beyond that. So that's sort of a good thing. So I say to people, if you want to be good at using a large language model, just go and read loads of books and not my book, better books, really well written books. But I think all of us now face a great challenge of making sure we, we work out for ourselves where we we think they can help us without coming to rely on them too much, without outsourcing everything to them, every thought and idea and concept. We just rely on a machine to give us the answers. And that I think is, is very possible. It's very easy to do that even in the course of writing the book. And I use these large language models while writing the book very openly enough talk about it and where I did and where it was good and where it was bad. It's very easy to start falling into the trap of faced with every single problem getting the machine to think for you. I mean, sometimes that's good, sometimes it can help, but other times you're just outsourcing your own thinking. And I actually feel like through writing this book. Book. I've written several books now and this is the first one where large language models have existed. I don't think I'm a better writer anymore. I think I've got worse through writing this book because I think I probably relied on it a lot to help me when I was struggling. All the words are still mine, but it made me realize how easy it is for all of us to start relying on it too much. And there are many problems. There are environmental problems, there are economic problems, lots of problems. Problems. But just on the issue of communicating with them. Over reliance, I think becomes one of the biggest challenges for all of us. And so for each of us there is a task to work out where it helps us and where actually it's too helpful and becoming to rely on it too much. And that is something for each of us. It's a different answer for each person.

Conor Boyle (42:34)

Well, that's a great place to end on. I'd like to thank Jamie again for fascinating conversation with. The book again is titled how to Talk to AI and How Not To. It's out this week. Grab a copy from your local bookshop and I'm sure it will help you navigate our current moment and where we could be all heading in the future. Thanks again Janie.

Mia Sorrenti (42:52)

Thanks for listening to Intelligence Squared. This episode was produced by Connor Boyle and it was edited by Mark Roberts. For ad free episodes and full length recordings, you can become a member@intelligencesquared.com membership and if you'd like to join us at any future events, you can find our full event program and buy tickets over@intelligencesquared.com attend. You've been listening to Intelligence Squared. Thanks for joining us.

Ryan Reynolds (43:31)

Ryan Reynolds here from Mint Mobile. I don't know if you knew this, but anyone can get the same Premium Wireless for $15 a month plan that I've been enjoying. It's not just for celebrities. So do like I did and have one of your assistant's assistants switch you to Mint Mobile today. I'm told it's super easy to do@mintmobile.com

Narrator/Advertiser (43:48)

Switch upfront payment of $45 for 3 month plan equivalent to $15 per month Required intro rate first 3 months only, then full price plan options available, taxes and fees extra. See full terms@mintmobile.com.