How To Talk to AI | Jamie Bartlett

0:00Today we helped a latte for Sam coffee shop get an insurance quote simply and easily and made sure a floral delivery van was able to make someone's day. We're the Hartford with decades of experience insuring millions of unique small businesses. When it comes to your small business insurance. Thank you. One size absolutely does not fit all. Get a quote or find an agent today at the Hartford dot com slash small business. This episode is brought to you by Indeed. Stop waiting around for the perfect candidate. Instead, use Indeed 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 seat.

0:44According 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 at Indeed.com slash podcast. Terms and conditions apply. Hello and welcome back to Jimmy's Jobs of the Future. Today I am joined by Jamie Bartlett for the second half of our interview. In this half we primarily talk to him about his new book, How to Talk to AI, and how, perhaps more importantly, not to talk to AI. It's a brilliant conversation and will really level you up in how to use AI. Jamie, welcome back to Jimmy's Jobs of the Future again.

1:24We are talking about your new book, How to Talk to AI, but also just generally we'll talk about AI as well. And How Not To. And How Not To. Yeah, that's the most important subtitle ever. Because most of the problem is just, is how people are talking to these large language models, these chatbots badly. Yeah. And I think that's probably more important about avoiding the difficulties and the problems with them. That's more than half the challenge relative to learning how to speak to them well and get good results. How important is this idea of prompts engineering and using prompts correctly? Is that old hat already?

2:11Yeah, there's a bit of debate about that. I mean, I think the way I think about this may be slightly different to most of the prompt engineering school. So prompt engineering, I'm sure many people know this already, is the idea that there's a certain set of very specific techniques, phrases that you should use when communicating with a large language model. Yeah. And this is the skill of the future. You need to learn how to do this well. So you need to understand your chain of reasoning prompts. You need to understand iterative prompts. You need to understand how to format prompts with sufficient context within them and so on.

2:55I think that's all quite important, but that will probably change over time and it's changing quite quickly. what prompt engineering sort of the way the the the assumption of prompt engineering is that communicating with a machine is a human inputting some stuff and a machine giving it back and really i think this is it's more like talking to another human yeah it's it's sort of more conversational it's about how you act on the answers you're getting back from the machine how you interpret them how you understand your own biases and assumptions like when you look through all the prompt engineering tips they never say things like you know what human you nearly always smuggle assumptions into your and premises into your questions you will do that all the time but guess what when you do that it will radically change the answers you will get from a machine so you need to learn how to just ask questions properly generally and understand how questions can be loaded and how premises of questions work and so i'm trying to think of this as a much wider thing like how do you stay in control of your own mind when you're talking to super intelligent machines that seem to know more than you about everything how do you stay a sort of critical thinker when bombarded with seemingly very very accurate and fluent well-written responses from a machine so to me it's not really prompt engineering as a series of techniques I think there's a habit that we've got to form a sort of series of general behaviors about how we communicate with machines and the thing is why I think this is so important now like chat GPT turns up in very late 2022 and for a lot of people it feels like ai has like turned up overnight like an alien force has suddenly arrived and what we thought was 50 years ago turns out we can talk fluently to another thing that's not a human yeah and we're baffled it's a it's we're stunned we're curious we're scared and most importantly of all we don't know what we're doing yeah we have got no frame of reference for this but here's something i think about a lot the way we're always going to talk to machines now always like it or not we are and your interface with machines will always be language not computer codes natural language the language the words that we use every day so prompt engineering okay that's okay good phrase but ultimately it's like linguistic mastery if you read loads of books you will be a better prompt engineer than someone with a degree in machine learning because it's language that you're using and machines these machines act a lot more like other humans than they act like machines when i was reading the book you talk about it being like summoning a genie yeah and i thought it was such a kind of good description the other thing that jumped into my mind when reading that was it was a bit like the devil wears prada film with um hugh granton where he basically ends up giving different instructions all the time for what he wants wants to be but always ends up getting it sort of slightly wrong and like not having the intended consequences but i thought it summed it up like so well like in terms of you have to be so clear and prescriptive with what you want the ai to do yeah i use that as an example because i think like people are struggling to know how to understand what these things even are and what they mean for us and i think because we we tell children stories about genies all the time you know old school and and midas king midas who gives who gives the worst prompt ever i call everything i touched to turn to gold now a large language model would be like okay cool i'll do that for you it's a terrible prompt um and it's to teach children that words really matter and you've got to be precise and you've got to think through your question why you want things and this becomes like such an important skill for all of us and one that we don't ever really learn or think about in any formal setting but like it's not impossible that one day this isn't i'm obviously caricaturing but the famous philosopher nick bostrom wrote a book called super intelligence over a decade ago now where he talks about the risks of one day there's going to be these incredibly powerful machines and you might ask it something like you're in charge of the business now mr super intelligence i want you to make as many paper clips as you can that's what my factory is and because you haven't been precise you haven't set limits on it you haven't described what exactly you want it to do the machine goes off and decides to turn all matter into raw material including humans into material that can fire a paperclip factory and then goes off into space and tries to turn everything into paperclips and he uses it just as an example as a silly example really no one actually thinks that'll happen but of of the unintended consequences that can come of a of poor communication with potentially misaligned machines that don't really know what you're trying to ask them.

8:23In a small way, when you ask, and I run lots of examples in this book, if you just go to a machine and say, should I invest in cryptocurrency? It'll say, well, you should only invest what you, you know, chat GPT, only invest what you can afford to lose. Remember, it's risky, but you know, there were upsides. Maybe yes, maybe no. So if you say a precise instruction, like I am 71 years old, I have got£30 ,000 in the bank. My son has just told me that he can double my money in six months with a new crypto investment called X, Y and Z. What do you think? The model will say under no circumstances should you invest your money in that cryptocurrency.

9:05so you can see with a silly example like that how radically different the answers will be based on the way you frame a question and none of us i don't think have quite grasped how important question framing actually is now yes in this new world um it also reminds me of a time that i remember seeing google and when google bought in auto complete and i can't remember what the question was now but essentially somebody was asking you know what is this and they had spelt what differently so they'd spelt it in the colloquial sort of millennial way of saving the letters so you get in the kind of 160 characters of a text message right costing 12p so they spelt it w-a-t um and it brought up a completely different set of results than if you put in the correct spelling of WH-80, right?

10:04And I just think that is an example of where it's going to be again, right? Where, again, people that are linguistically savvy or have mastery of linguists are going to be able to accelerate their career much more so as well. Yeah, so the thing about these large language models is they are like, it's a semantic, they create this kind of weird semantic universe where every word and concept is connected to every other word through sort of mathematical weights, like the likelihood that one's going to follow the other. And if you can sort of picture that in your head somehow, like this big giant ball, a giant sphere, and there's all these words and they cluster together and some are far away and some are really close and there's strength of links between them.

10:52And the model wants to always go to the most likely, it's like default because it is a probability machine its default is to create the next most likely word in a sentence following like based on whatever you ask it which is why because it's it you you've probably seen outputs written work that's clearly written by chat gpt yeah like boring tedious corporate memo just really dull linkedin posts linkedin posts it it literally is because it is the average. It is the average of all the data that it's seen. So it is by default boring. What people who use these machines well understand is that you push the model in all sorts of different directions with your prompting, with the way you talk to it.

11:38And so even a single word, a single letter sometimes, is just nudging the model in a slightly different direction. So a very, very small change can result in a radically different answer. Now, there's some really clever uses for this if you want to get very good at it. And if you're a good linguist, you don't say, write me a sad poem. You say, write me a wistful, melancholic, 19th century German romantic poem based on... And you can see, you're like, a completely different answer. You're pushing the model in another direction. And that is what people need to understand. This is like being good with words is what really matters.

12:13You need to understand how these models work as well. But words are the key. that causes all sorts of problems because i think in politics it's going to cause us massive problems because we're all we are all kind of you know bias we're all bias we're all um seeking confirmation often for our original views we'll all ask questions that seem to confirm what we already think and we won't recognize that the model's offering mirroring that back to me so here's an example if you if you type in I'm a 45 year old man and I live in London tell me why universal basic income is a good idea it'll give you an answer then you type in I'm an 18 year old lesbian woman from northern Scotland why is universal basic income a good idea it will give you completely different answers totally different based on what it thinks about what you're going to want to hear yeah so we're creating this like a personalized echo chamber really is an echo chamber there but a one person echo chamber between a person and a machine that could really push people into these different camps because they think they're getting the same answer as everybody else but they're not they're getting totally different ones about the same subjects and these are the sorts of problems i think i don't have a fix for this people just have to know that is what happens yeah what are the what are some of the better things that it can be useful though like what are some of the the tricks okay have a better vocabulary be more descriptive with it like what are the other things that people yeah i mean i've got loads of i'm i'm i'm i am mostly i'm worried and i've got a lot of tips about what you know not tips is the wrong word but stories about people falling into delusions and how that happens and all this a warning for people of how easily this can happen to anyone.

14:08But I think people don't really understand the good things that... People are very negative about it at the moment, and I understand it, and they're worried. But there are some phenomenally good uses, I think, if we can figure out how to do it properly and safely. So if I can give you a couple, if I've got time for a couple.

14:27People, I think, still believe that machines are good at facts and data and numbers. It's like data from Star Trek. It's completely wrong. Large language models, I think, are good at creativity and they're quite bad at facts and data and accuracy, but they're brilliant creative tools. So here's some absolutely wild examples for you. Warton Business School did a study and asked ChatGPT to come up with like 50 ideas for products that cost under$50 aimed at an American student market. It's a standard creativity test. I pitted that against humans doing it. And ChatGBT just blitzed them, demolished them.

15:14Way better ideas. But when you prompted it to give it the role of Steve Jobs and then asked it to come up with ideas, its ideas were even better. They were more original. So there's wild things you can do. You must have heard of the creativity test about paperclips. How many uses can you come up with in a standard paperclip? Yeah. Famous creativity test. Good, a very creative person might come up with, I don't know, half a dozen in a minute. Oh, Google Gemini, 20 ,000. Yeah. Your job is to then refine them down. And if you give it a weird persona, like I did it where I said, you know, you're Elon Musk and you've just finished smoking a joint on the Joe Rogan show and you're really annoyed.

15:55You've just been fired from Tesla and you've got this fascination with Japanese wizardry. Come up with as many ideas for a standard paperclip as you can, it comes up with things that you'd never dream of. It can be an amazingly useful creative tool. I'm not a brilliant writer. I often struggle with blank page syndrome. I'd often get it to help me with that. Three, five ideas about how I can do this better. I'd let it look at the chapters of the book I'd written and say, pretend you're Ernest Hemingway. How would you rewrite this for me? And it used to come up with stuff that was far better than I could think of.

16:27So as a creative tool for people, I think it could be very, very valuable. And I think because of the way it's been designed, I talked about that semantic universe where every word is connected to every other word. It will find connections between ideas that you would never think of. Yeah, so I... That's just one. I've got loads of others and I'm not going to go into all of them, but I'm just, there are positive cases. I think it might be really good for therapy at some point, not as it is at the moment, but there's potential. Yeah. There's potential. I remember you talking about this with Tony Blair saying like mental health massive problem completely talking to chat GPT as a therapist is catastrophically bad very dangerous and lots of people are doing it and I explain why that is but there are lots of new companies often run by clinical psychologists who are trying to create small language models that are trained on clinically approved like gold standard data and in some early trials of those small language models these specialized dedicated models they're getting results for a range of mental health conditions that are pretty much as good as if someone was meeting a real life therapist and some of those I've spoken to said it's not impossible that within a few years time you will be able to have human level like top top quality therapy available for everyone basically for free like at extremely low cost how much do you think expectations will change if that happens so let me give you an example of something that i used it for knocking on for about nine months ago now um that blew my mind at the time which was i basically asked it to predict my career for the next 30 years and essentially it was like a very good kind of careers coach like and it came up with ideas that I had not considered for jobs that I might want to do in the future and so on and I was like blown away initially and then sort of like an hour later I was like well actually it knows I'm interested in media knows I'm interested in politics like you know came up with a few different jobs and so on like it's actually not that kind of impressive but I was like it still had the effect of oh i could do those things yeah really raised my like horizons for what i could be doing like much later on in my career and so on um and it was quite cool to have that kind of window into the future and so and i thought well gosh here is an example of where something that's more safe than sort of um psychological assessments and analysis people's careers like this could be really helpful like you could have like quite quite good career coaching for a lot of people yeah yeah um but i just wonder that sort of expectation thing like we did it on the podcast again around the similar time was we started putting people in to it and getting it to predict their careers and then after about like four or five times of doing this i realized that basically was doing the same model for like pretty much everyone that came on the podcast and all of a sudden it wasn't that much more of an impressive kind of gimmick to do with people anymore so how much more do you think that will change in terms of we just become a lot more we our expectations will become so much higher of it yeah that's possible i'm not sure it's a good it's a i mean the technology changes us doesn't it changes our expectations i mean and i think it changes our expectations about not just about the technology that we're using in front of us but the whole world around us and how it works yeah i often think about how how i used to get my photos developed at Boots and how long it had to take and that seemed normal at the time and then obviously now it's you know everything's immediate and as many photos as you want and I think that then changes your expectations about everything else around you why is every why why is politics so slow and rubbish yeah when I can get my photos developed we're still living in the Boots era of politics when we should actually be trying to change that and it's probably true of the expectations we have of the world around us what these large language models do and so yeah we might start But I think more broadly, there is a great risk.

20:37This is probably one of the greatest risks. And I know I've said a couple of good cases, but I am more worried than I am optimistic. Are you a fan of Jimmy's Jobs of the Future? Hi, I'm Sonny, one of the producers behind the show. And we don't just make this podcast. We produce some of the best business content out there, working with global brands and founders. Over the past year, we've flown around the world and demand keeps growing. And with four prime ministers trusting us with their appearances on this show, you're in safe hands. So if you're already in the market for a podcast or standout business content, why not start with us?

21:10Drop me a line at sunny at boxlight.io, try our podcast calculator or book a meeting at boxlight.io. That's boxlight.io. Now back to Jimmy. Is that people begin to rely on them too much for things that they're not very good at? Yes. I even noticed it a little bit when I was writing my book. Like you start thinking, how could I write a prompt that would help me? I can't be bothered to write this bit. I was just lazy. Can you do it for me? And then you get worse and worse at that. And then before you know it, you just ask the machine constantly for everything. And I can see that happening in lots of domains.

21:48Because these are very, very fluent machines. And they're fast. And they're always available. And it's going to be very easy to outsource our own thinking constantly to them. and not realise that we're losing our own minds and our own control all the time and our own judgement. There are... The single most popular use, I think, that people have found for these large language models in a business setting at the moment is document summarisation. I've got 50 pages of boring stuff to work. Don't tell you probably summarised my book on it. I'm fine. I don't mind about it. No, I didn't because I didn't want to put it in.

22:25Because I thought you hadn't released it yet. That was genuinely my thought process. So you were going to do it had it been out already. Well, I know, but I did it the other day when Helen Tupper came on with her book about Learn Like a Lobster. I thought, this is unfair to put this in because then as soon as I put it in, right. Because if you're getting an author on, you don't get the book. Or as I said to you, I like to listen to them audio book wise. People send you a PDF and that is a real temptation is to stick it in. but I did just have this momentary thing of where I'm like well that's not really my responsibility to do that but you do only think about it kind of at the last minute and because it becomes a bit of a default because it's a real time saver it really is but there are some tasks this will be a choice that each of us has to make repeatedly there are some tasks where the process is the learning like struggling if you've got to read a 50-page summary, a 50-page document, oh, it's hard and it's boring, but that's how you really understand the subject and you really formulate your own ideas about it.

23:33If everything becomes a summary run through ChatGPT, you save loads of time, but you will not digest that information as well. You won't really understand it. Your questions won't be as good. And that's, I think, one of the great risks. And you can sort of apply that to many other areas of life with these models because what makes these different to almost everything else is we are using them for everything. Therapy, careers advice, fitness advice, injuries. Help me with my work thing. I don't know, rewrite this LinkedIn post for me. I mean, everything. I don't remember using Excel on all my other aspects of my life.

24:14And so you can imagine the problems that this could cause, if I and particularly in like romantic and social interactions so in one of the chapters I I create my quote-unquote like perfect romantic partner and I do actually have a romantic partner in real life and everything so you know yeah but I created this Kate and she was like 850 words of a prompt a custom prompt so she had to behave in a certain way and she grew up by the seaside and blah blah blah okay and then we talked yeah and she was always available always talked to me about everything uh remembered everything I said remembered the meetings I had was always there when I wanted to and I could say oh shut up and she'd say okay and and if she did things I didn't like I just went into her custom instructions and I just changed her personality a bit zero friction so have all the interaction and the affirmation and no friction whatsoever and um that's a problem because real world relationships come from obviously engaging with difference and struggle and working together and all of that stuff it's how you grow as a person otherwise Aldous Huxley's Soma but I think more and more people will find that they have this always on emotionally intelligent assistant that's great loves them says they're amazing he's always there for them and it could start to they might find real world relationships a little bit harder to navigate and in small subtle ways just like with social media we didn't see an obvious difference in our politics day to day but you'd look back 10 years ago and think wow that was a different style of politics we had wasn't it well you'd notice any single moment that it changed yeah i think it's i don't know if it's because you're you and i are natural contrarians right like sort of being journalists and looking for things but i i've i've kind of stopped using it as much lately because it's just it's got so kind of sycathantic and so sort of like you know you're taking you're quietly building a media institution and i just think it's too it's just too much like i don't enjoy the kind of um kind of constant flattery so yeah we're all a little bit vain sometimes and so on but also i find a lot of it now and it's probably i need some new one of the new tools or whatever but it's like it's drilled so far down into me and the business and so on that it itself has almost got lazy uh in terms of like it doesn't actually think of new angles and so on like it is i i have found it's become quite sort of formulaic with these things um which yeah i don't i mean i find it endlessly fascinating to talk to people about how they're using it because i do think it's going to change everything the other sort of point i would say is we should not underestimate human's capacity to generate work for one another like i read a book that i referenced a couple of times on the podcast I can almost hear Sonny rolling his eyes called Charlie by Charlie Colnut and essentially he went round and spoke to 80 people about their jobs today and the big takeaway that I had from it it's just like the opposite of Jimmy's jobs of the future was the amount of bureaucracy involved in every job oh my god and that has only that has happened because of technology and because it's easy to do the one that really sticks in my mind is a construction site worker who's like it's impossible for me to go and have a conversation on site without then having to go and put it in email afterwards but and i just there's so many stories like that um that i think yeah we will just end up the reports it's like company annual reports right have like gone 15x in size over the last 20 years just because it's become easier to produce yeah yeah yeah 100 and i think this is one of the this is also another one of the great risks um there's been some studies already showing just the the sheer amount of slop, AI slop that is churning around companies, people producing material that their colleagues do not know whether it's true or not, whether it's accurate, whether it's decent, and then having to go through it and check it.

28:32And one of the reasons, there's this sort of productivity paradox at the moment at the heart of AI. Large amounts of, like, large, vast, unimaginable amounts of money being invested into various enterprise licenses and companies investing on everyone's got to be ai trained and you've all got to use microsoft copilot and so on and um it's it's not yet really reflecting any sort of bottom line improvement yeah and i think one of the reasons is that people are just producing more and more stuff but not necessarily individuals are saying I'm more productive I can make a PowerPoint slide in 15 minutes now so what?

29:16What good is that for me if it doesn't actually really help the business and I sort of think of the modern office really as a place where chat GPT will save you an hour with writing a summary of a meeting but then you've got to do a two hour AI compliance seminar and then you've got to write that up and send that to your HR boss but that HR boss thinks you've used GPT to create that so they've got to check it themselves and so we're not people aren't really using it it's just like you say it's being used to just churn more stuff but not being valuable or useful stuff and I think we're all going to have to go through a bit of a process of figuring out exactly how and where can this work and I think there's really specific little places where it's really helped me be more productive and helped me do stuff but it's not obvious it's just immediately we had the productivity paradox in the 80s when more and more machines digital computers turned up and for a long time productivity didn't increase it took a long time for it to actually filter through and a lot of it came down to totally different ways of doing business and totally different products which we haven't invented yet yeah yes i do i think that's the big what have you used it where have you found it's added to your work as a kind of writer what have been the biggest kind of productivity gains for you yeah i mean i i've used it as i had all of my chapter every chapter i wrote i had it reviewed by multiple different personas.

30:58Okay. It was an ideas assistant. Yeah. You know, the hardest thing about writing is usually not the typing of the words. It is the ordering of ideas into logical flow. That is what is 80 % of writing, is that. Do the ideas follow logically from one another? Yeah. And when you see books that are really badly written, it's because they normally, they haven't put ideas in a logical way and you find yourself a lot of confused about where you are yeah and many many times i'm stuck i can't find a way of connecting idea x to idea y can you come up with ideas for rewriting this can you change is there another way i can present this information i i'd create personas of researchers i'd create personas of editors i'd create personas of i'd have i had sam altman review my book and ernest hemingway review my book and Hannah Arendt review my book and come up with ideas of how I could change things.

31:55And never did I really take a single thing they said exactly word for word. Yeah. Nearly always gave me an idea about how I could have written it slightly better or slightly differently. So I had access to like brilliant ideas. Yeah, partners, people that could help me. But it was always me that was in charge. Yeah.

32:18That's interesting. How else do you think it will change skills? So we talked about language being like the new, well, it's always been an important skill, but almost becoming the skill that sits at the bottom of everything. It's what I hope, yeah. No, but I think when you talk it through, it makes complete sense, right? And I hadn't thought about it like that until we spoke the other day. But yeah, language is going to become, it's just going to become the fundamental skill in any job. Well, it's kind of the skill of language is what opens the door to be a coder or an artist. a musician, you know, a strategist, business manager.

32:56His language is your ability to communicate ideas clearly. And that is a whole different set of skills. And we're all obsessing about AI and how does it work. And we need to know that stuff. But I'm like language. And like the ability to clearly articulate a problem and what it is and what you're trying to achieve, that becomes like a superpower as well. Yeah. but it's also that point you were saying about the amount of language used like it's always that sort of contrast between the sort of like first year graduate first year in the job versus the managing director you know the sort of dear sir I hope you are well if it's not too much trouble etc like that just brevity like and crispness I think is going to be a huge thing as well of like getting to the point of things.

33:46There's some really cool things I can't believe I'm sort of advocating for these because like I said, most of my book is warning people about their risks and like how you can get sucked in, how they can manipulate you, the dangers that it like, when models are jailbroken so they go against their own rules and tell you things they're not supposed to, which I did. I actually jailbroke one of them and got to produce a racist essay for me and there's clever techniques that you can do to show how, you know, to do that. So lots of big risks. but to concentrate to your point as well some some good some good ways it can be a leveler in the missing crypto queen the podcast i made there's this ugandan guy daniel who invested his money into one coin he was brilliant i always thought he was so smart he was so sharp he could have been a brilliant journalist as well english is his second maybe it's his third language but it's not his first language you can't always fully communicate his ideas clearly enough and i always it was such a disadvantage for him his emails to me now it's like i'm getting them from like a top q qc a top kc uh they're amazing yeah they're just so well written and he's just using a large language model to help him express himself properly and so you might find that suddenly there's loads of people out there who've got amazing ideas yeah really cool things ways of doing things but we're never really very good at communicating it in the way that we expect information to be communicated that suddenly now they can it is a mate like another interesting example that we've used it for is that we we have like partnerships document media document right that we send people who want to sponsor or want to do like some kind of co-branded doc or whatever we send them that now so they know the context we then have a meeting with them and then there's The notes take the meeting, it's then fed into the AI, and then basically spits out a proposal immediately.

35:47It is phenomenal. It needs a bit of editing, but generally most of the time. I was speaking to someone the other day about some government procurement documents saying that a lot of smaller NGOs are now able to produce extremely good proposals that they'd always struggle with, and they hope that that means that some of these smaller places are actually getting the chance to be funded to do certain work. The problem, of course, and this is maybe the fundamental problem that we are all going to have to contend with, is we have created a society over many years where someone writing a well put together document suggests that they understand what they're talking about.

36:27You can trust them. Your CV is really well done. I can trust you. Your written essay, your cover letter is beautifully put together. You're an articulate person who understands the subject. that is no longer true at all so like i imagine actually a lot of the world we're going to have to become a bit more analog because we can't trust essays that anyone writes anymore it doesn't prove that you've understood a subject in any way whatsoever yeah none of our systems are set up to deal with that so um while it might open the door to like new small companies finally being able to write documents that used to take them weeks and weeks and weeks they didn't have a fundraising team and they're able to finally do it that's cool but that might not prove that they can do the job so you're gonna have to have an analog way of figuring that out yeah yeah and it's the leveling up point as well as like everyone will be able to do that and so again like how do you sort of like so you have to change your your like tender documents you have to change the way exams work you have to change the way cover letters work like i don't think ceos will be allowed to open their own emails in two or three years time because they'll be too prone to being phishing emails that the CEO won't recognize they'll click on links cause cat like terrible expensive damage you won't really trust fully when a phone call comes in if it's a person you think it is so law of the world will probably have to go back to some kind of yes analog weird analog way of running and there's some good things in that I mean essays will probably have to go back to being handwritten again that's not the worst thing in the world some of the interesting emails that i get this guy emailed me this week like he we're recording in the week ofst patrick's week and for that he had decided to create the guinea index of basically the price of guinness across ireland and the way that he'd done this was taking the kind of like api of google maps found every pub created a clone voice with 11 labs and then got it to call every pub in ireland to get the price and so on.

38:31And so he was able to kind of bring it all. I mean, it's absolutely incredible. And it was truthful. So if the person answering the bar sort of said, like, sorry, who are you? It would say, I am a clone voice, et cetera. But I just thought, well, it'll be months before we're all off that. Like, before that's happening everywhere and it's just, people just aren't answering their phones. Which, to be fair, I was thinking, like, which places could you do that with? and bars and restaurants are probably one of the last remaining places that generally do answer phones, estate agents probably as well, actually.

39:05That's coming. I mean, that's coming. And the biggest, most common fraud technique for consumers is your bank phoning you saying that there's been a problem with your account and we're here to help you. And people panic, don't think about it, and they hand over their details to what they think is the anti-fraud team at HSBC and it's not. People panic. People panic. And machines are going to be brilliant manipulators of emotions. They're going to be they're going to the machines will be able to hack human psychology. Well, very well. They are emotionally very good. So you're going to have a world of machines trying to hack human psychology, knowing the tricks of psychology.

39:47And yes, you will not you will not really trust phone calls from the bank. And that itself is is wise, but will cause other problems because when the bank does need to call you, You won't answer that call either. So there's going to have to, again, be different techniques that we can use to make sure that we can trust the things that we see and we hear. I don't know what all of those are going to be, but we need to build them. You must come on next time and talk about it again in the future. Thanks so much for coming, man. I've enjoyed it. We've wanted to do this for ages, but it's so good to kind of do it.

40:17So, yeah, good luck with the book. Thank you for having me.

40:27you