The AI Argument

Vibe Coding Risks, HTML vs Markdown, and Camera AirPods | EP100

• Frank Prendergast and Justin Collery

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 41:17

🎮✨ PLAY THE 100TH EPISODE GAME ✨🎮
Frank vibe-coded a tiny arcade game for episode 100.

🧠 Balance AI growth
📜 Collect regulation documents
🔨 Whack problem bots
🚨 Stop chaos spiralling out of control

👉 Play here: https://www.theaiargument.com

—

Can vibe coding turn anyone into a software developer? Or is it quietly creating a new kind of AI brain fog? 

For episode 100, Frank and Justin test the promise and problems of AI coding, from Frank building an AI Argument video game with OpenAI Codex to the bigger question of whether AI agents make us more productive or just more overloaded.

Plus: HTML vs Markdown for AI workflows, Claude handling airline refunds, camera-powered AirPods as Apple’s possible AI breakthrough, and doomer and e/acc predictions for the next 100 episodes.

Would you ship a vibe-coded app without understanding the code?

00:11 Did we survive 100 AI arguments?
04:10 Did Frank just vibe code a video game?
08:33 Can you vibe code without coding skills?
14:06 
Is HTML better than Markdown for AI?
18:47 Is vibe coding bad for your brain?
21:22 Would Claude hallucinate less than Justin?
23:39 Should you vacuum while vibe coding?
27:36 Are camera AirPods Apple’s AI breakthrough?
32:29 Can doomers and e/accs be friends?
34:12 Are we heading for ASI or P-gloom?

#VibeCoding #OpenAICodex #ClaudeCode #AIProductivity #AppleAirPods

► SUBSCRIBE 
Don't forget to subscribe to our channel for more arguments

► LINKS TO CONTENT WE DISCUSSED 

► CONNECT WITH US
For more in-depth discussions, connect Justin and Frank on LinkedIn.
Justin: https://www.linkedin.com/in/justincollery/
Frank: https://www.linkedin.com/in/frankprendergast/

This is an AI transcription and may contain errors

Did we survive 100 AI arguments?

Frank: I'm Frank Prendergast. I'm here, as always, with Justin Collery, where normally we look at all of the latest AI news through the lens of a techno-optimist, that would be Justin, and an AI doomer, that would be me. But we might do things slightly differently this week. I want to start with a little bit of trivia, Justin.

Justin: Go for it.

Frank: Did you—

Justin: This is—can we explain that this is a very special show? That's why we're doing—

Frank: I'll get to it. I'll get—

Justin: Right. Okay. Okay, go.

Frank: So did you know that there are over three million podcasts, but actually only just over 400,000 of them are active? And there's research from Podmatch, and it says, on average, it takes two years and 100 episodes released before podcasters see results. 93.84% quit before reaching this point.

Justin: Wow. Can we—

Frank: 100 episodes before you start seeing results. Justin, fame and fortune awaits us.

Justin: R—

Frank: It is just around the corner.

Justin: I would like to compliment our listeners and our watchers, because they are the people that got in on the ground floor, in before this thing took off, before it went global. Those were the people in from the start. They are the people that know who the real AI Argument is before we sell out to whatever the big commercial interest is that we're—

Frank: Absolutely, yes. So here we are. We are at 100 episodes as we speak, unless, of course, I somewhere along the way got the numbers wrong.

Justin: To big it up, how many views across all our platforms, Frank, have we got over the 100 episodes?

Frank: I actually have no idea.

Justin: 530,000.

Frank: 530,000 views. Now—

Justin: Views.

Frank: Now, are you including social clips and everything in that?

Justin: I am including people who are sitting on the toilet and thought about us. I'm including everything. Including absolutely everything in that. But when you count up all the numbers, add it all up, 530,000 views, 10,000 likes, I don't know how many thousand comments. It's gone far and wide, way beyond where we would've imagined it might have gone when we started.

Frank: I think we deserve a pat on the back at the very, very least. I absolutely do.

Justin: We have brought the AI Argument to more people than live in the whole of Cork.

Frank: Amazing. Amazing.

Justin: Not quite as many as live in Dublin, but—

Frank: It is actually kind of fascinating, though. I don't know how many people listening might have their own podcast, et cetera, but for us, I would say most people actually engage with the podcast itself on YouTube. And I have been finding the audio side really, really slow, and I've been kind of a bit demoralised about how many listens that we have every week. But actually, I looked into it. It turns out we're well into the top 50% of podcasts, even on the audio side.

Justin: There you go. So the glass is more than half full, Frank.

Frank: The glass is more than half full. Absolutely.

Justin: I think this is an important historical record, and I think 20 or 30 years from now, when they do a Reeling in the Years of AI being rolled out, they'll have little segments of the AI Argument where Frank has predicted that AI is going to wipe us all out, and it didn't. How many times did you predict that? Justin predicted that Sam Altman was going to be out of a job by the end of 2026. Spoiler alert, he was still there in 2046.

Frank: He now rules the world.

Justin: And so they'll go through, and they'll have these funny little clips of us talking rubbish, and that'd be nice. It'd be cool.

Frank: It will. It'll be great.

Did Frank just vibe code a video game?

Frank: So a couple of weeks ago, OpenAI released their image generator 2.0, and when I was experimenting with it, one of the things I did was I imagined what a game for the AI Argument might look like, and I showed you some concept art for this game, and you said, "You know what? You should go away and you should vibe code that game."

Justin: Yeah.

Frank: Well, I did. So I would love to show it to you.

Justin: Let's go.

Frank: So let's—

Justin: So if you're listening to this at the moment, stop listening, jump onto YouTube and watch it on YouTube or LinkedIn, or wherever you get it, because we're going to do something on the screen now, and you probably want to see it on the screen.

Frank: Yeah, I think that's a good idea. We'll talk through it, but I think it really has to be seen. So if you go to theaiargument.com.

Justin: Yeah, there we go. Nice. Woo-hoo.

Frank: So again, this—

Justin: Look at that.

Frank: All thanks to OpenAI and Codex. The landing page was vibe coded. Now, I did go out to ChatGPT and manually create some of the images, and I did take some screenshots, but by and large, yeah, mostly down to Codex vibe coding this page for me.

Justin: Okay, so I'm going to go click on play the game, and load the game. Do I need to know what keys I should press beforehand, or is it going to tell me?

Frank: Oh, we've no sound. No, we've no sound. Let's see. You might need to reshare your screen and click on the sound button. The joys of—we actually, for anyone who listens on the audio or watches on YouTube, we actually go live on LinkedIn. We actually record this live, and this is the joys of working with a live podcast.

Justin: I can hear the music. It's sort of nice eight-bit music.

Frank: I'm still not hearing anything here.

Justin: All right. And there's nothing that I can do, I think.

Frank: Yeah, that's all right. So, yeah, we've got an eight-bit version of our... It's actually the same music that we use at the start and the end of the show, but an eight-bit version. Good idea. I'm surprised you went for read the instructions. That is not what I would have expected.

Justin: Okay. A or D controls the... Okay, and W, R, S, and space is action. Okay, let's go.

Frank: So I don't know if you—did you—what was your first computer?

Justin: Oh, it was an Elanex PC 333. My friends—

Frank: Of course it was. Of course it was. You had a complete nerd computer.

Justin: How do I go to—oh, there you go. There's me. Whack. Whack.

Frank: Okay, so now you've swapped to play yourself and you're playing Whac-A-Mole with the AI chaos agents.

Justin: Oh.

Frank: So we've got a hallucination bot, we've got a deepfake bot—

Justin: Okay.

Frank: And—

Justin: Good. I nearly... You see how well I'm doing? Look at this. You see, there's—

Frank: You are doing pretty well. Yeah, although you're not doing very much regulation. Oh, here we go. You've swapped to me, and here I am gathering documents, audits, and the like to put together regulation. That AI that's growing in the background is looking very angry, though. It's probably because the chaos bots are about to take over.

Justin: Total chaos, even though I was collecting all of the audits and everything. That's so sad.

Frank: So as you... This was actually a game mechanic that was suggested by you when I showed you the artwork. As you gather regulation, it slows down the AI growth, and as you're whacking the chaos bots, it speeds up AI growth.

Justin: Okay. I can tell you that was great craic. That was brilliant. Go to theaiargument.com, click on play game, and play it. It's fantastic. Tremendous fun, so it is. What's your top score?

Frank: I don't know, but I have gotten very good at winning it. I will say as well, anyone who's playing it should play it on a laptop or a desktop. You can play it on a mobile device, but I only just added that functionality, or I should say Codex only added that functionality, and it is very, very difficult. I have not managed to win using a smartphone yet, but I am going to adjust those mechanics.

Justin: Oh.

Can you vibe code without coding skills?

Frank: It's been an absolutely fascinating process for me because I have always intended to vibe code something, but I never had anything that I had any interest in vibe coding. So it was when this came up, I thought, "This is a good test now to see what this is like."

Justin: All right, Frank, couple of questions for you, for everybody's benefit out there. So how much coding have you done before you wrote this game? That game, by the way, was fully, for me, a totally professional game. Perfect, right? I remember years ago testing a game for Guinness, in fact.

Frank: Oh, yeah.

Justin: It wasn't even as good as that, right? So it was great, but it wasn't as good as that. So how much coding have you done before now?

Frank: That's a funny question because I could not code my way out of a paper bag these days. But it is true that, first of all, I have spent a lot of time managing development projects, managing web app development, things like that. So I've spent a lot of time around developers. And my first computer was a Spectrum 48K, and I did... There's a funny rattling, by the way. I'm getting a funny rattling. I don't know if it's at... Oh. So for anyone listening, that was a selection of Justin's pens, which he clearly has a lot of pens.

Justin: Signing all those regulations that I love.

Frank: But yeah, my first computer was a Spectrum 48K. So here's... I'd like to tell our origin story about how this podcast came about in a minute, but here's my origin story. I absolutely hated school. I absolutely hated it with a passion. But weirdly, it was just an interest thing. It was just if I was interested in something, no problem. Because when I look back on it, I was like, "Wait, I hated school. What was I doing going to a summer camp to learn how to code in Basic when I was in primary school?"

Justin: So you're a nerd.

Frank: Total nerd. Total nerd. So my Basic programming is not going to get you very far these days in trying to code a game. I would have a little bit of PHP, a tiny bit of JavaScript, but I could not code a game. I would not be able to code a game.

Justin: I'm going to ask you more questions, but just in the spirit of a bit of reminiscing about nerdism, I should tell you my origin story when it comes to this is one of the most exciting things that happened to me as a child. Well, you remember the Virgin Megastore at the end of the Quays in Dublin? You probably don't because you were in Cork, but the Virgin Megastore was at the end of the Quays, famous for selling condoms back in the '90s before it was allowed to sell condoms. And I remember with great excitement the day that DOS 6.22 was released, and I hopped on a bus from Lucan into Dublin because I found out that the Virgin Megastore had it. Walked in, bought myself a copy of DOS 6.22 and took it home and installed it, and it was one of the happiest days of my entire life.

Frank: Amazing. Amazing. I mean, I probably wasn't quite that nerdy because I would've been more excited, for example, about the pirate copy of Barbarian that was going around on cassette tape for the Spectrum 48K when I was in... Yeah.

Justin: I'm absolutely shocked you're...

Frank: I know. It's terrible. It's terrible.

Justin: Stealing other artists' work, I cannot believe it, Frank.

Frank: I do not condone it. I don't think I realised at the time how wrong it was, and if I could go back in time and tell my younger self, "Somebody put long and hard hours coding that game so that you could chop somebody's head off on screen and try to hide the game from your parents." But yeah.

Justin: Before we go into the origin story, because the game is great, right? But I just want to... Because there's people that say... Okay, there's a couple of things. So how much of the code do you understand? If there was something wrong with it, would you be able to tell me what was going on there?

Frank: No, not at all. Not a clue. I have no idea. I've actually, Justin, never inspected the code. And so I will tell you this, I am very, very glad that in order to vibe code a fun browser-based game, there is really no security aspect that I can think of that could go horribly wrong. I'm very, very glad of that. If there was any form of a database or any form of personal details being stored, I would not be putting this public.

Justin: That leads us beautifully on to, for me to announce, the Frank Prendergast bug bounty programme, where Frank is personally offering €1,000 for anybody who can hack the AI Argument game through the AI Argument website. Thank you, Frank. So okay. So you've got no real programming experience, right? You created a game and it's brilliant.

Coding Anything with AI

Justin: Do you think, would you be able to create an app that was maybe more of... Would you be confident enough to write more software as a result?

Frank: I would definitely be interested in keeping more of an eye out for apps that would help me in my work or productivity or what have you, and coding them, and either keeping them entirely for myself or ringing you up at midnight and being like, "Would you have a look at the security end of this for me?"

Justin: It's actually interesting what I did last weekend, by the way. I was having to plug out last weekend. Last weekend, I just needed to go away, but I wanted to do...

Is HTML better than Markdown for AI?

Justin: So Markdown is the language that people use to give, right, it's just text format and whatever. But on my Mac, you have to pay a fiver for a Markdown editor. So I just said to Claude Code, "Just write me a Markdown editor." Half an hour later, I had a Markdown editor. Worked perfect. No need to buy the apps from the App Store anymore.

Frank: Nice, nice. Yeah, yeah. This is a total aside, right? But on the Markdown thing, did you see there was a tweet recently from someone who works at Anthropic, and he said he's moved away from Markdown entirely, and he does everything via HTML now.

Justin: Listen, that is not an aside, by the way. So I was actually sitting down talking with one of my children yesterday. I was impressing upon them how exciting it is to be young at this period, right? I said, "Look, it's like a green field. Everything is there to be discovered, right? And anybody can discover anything." And I was thinking of that exact tweet. Was it Theodore or was it...

Frank: I've actually forgotten. Yeah.

Justin: The guy it was from, from Anthropic. But anyway, two years ago, I was involved in a project where we looked at medical documents, and I was explaining this to my kid, right? And so what we discovered was there was a young guy that was straight out of college. He was brilliant, and he was working with me. And we were trying to scan these medical documents and turn them into Markdown, right? But what we discovered is, in medical documents, you can have an arrow that points up, and sometimes it's green, and that means a good thing, and sometimes it's red, and it's a bad thing, right? So you might have an up arrow because your blood pressure's gone up, is red, that's bad, right? But an up arrow because your white blood cell count has gone up, that'll be a good thing, right? And so colours and symbols are important in medical documents. We did loads of tests, and we ended up, we settled on HTML.

Frank: Right.

Justin: While then, so we said, "Oh, yeah, look, actually HTML is better than Markdown." That was two years ago, and now people are starting to say, "Oh yeah, we should use HTML instead of Markdown." And my point was not to say that really that we were great what we did two years ago. My point was just that it's all virgin territory.

Frank: Yes.

Justin: Anybody could discover something that's groundbreaking and amazing. And did you see all the responses to that tweet? Everybody's going, "Oh my God, this is the end of Markdown. We're now in HTML."

Frank: Which I mean, I think you just made a really good point there about that's a brilliant use case for HTML, a very, very specific type of document. But I would not be rushing to move from Markdown to HTML because he even admitted, I mean, he highlighted it himself in the tweet, the additional amount of tokens that you're going to use and the speed at which it can develop HTML versus the speed that it will develop Markdown, there's a big trade-off there that I would be more looking for the use cases where it makes sense to use HTML rather than doing a complete—

Justin: I'll give you another, right, because he made a good point as to why you should move to HTML and not use Markdown, and that was to do with readability. So humans, when you just have a wall of text in front of you, right, it's quite tiring to read it. Whereas if you render it as HTML, right, it's easier. And I can actually give you loads of other good use cases because I do it myself at the moment. So instead of using Markdown, what I do is I use JSON format. So JSON format's a computer format. It's got tags and values and stuff like that. And then I have a very thin wrapper, and then on top of it, I've got an HTML viewer. All right? So everything looks beautiful, right? You can see it and you can read it easily, and it has all those controls that are really cool. So you can have a slider with the value from one to 10, or you can have traffic light symbols. Yellow, red, yellow, green to highlight if something is good or bad. You know what I mean? You've got all that richness which you don't get in pure text, but you don't have all the overhead of having massive amounts of HTML because it's in JSON format, and the models are just as good at writing JSON as anything else.

Frank: And that's another really good point. So I think, for me, what I would probably be doing, because the readability thing is a really important point, and that's probably the one area that I have used HTML, is actually instead of sending a client a big, long report or a big, long document, just sending them an HTML page that looks a lot more inviting, is easier to scan, et cetera. But what I would do is I would work with the AI in Markdown or in text and then develop a readable HTML at the end of it, rather than going back and forth with HTML the entire time like you might do with Markdown. But anyway, total nerd-out there for a bit.

Justin: What actually it brings us onto our first story. We're only 17 minutes in. It brings us onto our first story where, yet again, I'm right and you're wrong, and this is to do—

Frank: Hmm.

Is vibe coding bad for your brain?

Justin: This is to do with vibe coding, and vibe coding can be dangerous for your brain. Not just vibe coding, right? This was a study released by the Harvard Business Review, and what they discovered was that AI is leading to brain fog for a substantial number of people. Different types of people. 80% of people who use AI a lot, 80% of them reported brain fog. And what they found was that AI basically produces a whole lot of output, and as people use more and more AI tools, their brain becomes overloaded because there's just too much information being thrown at them. And they found the sweet spot was three. So if you had more than three AI tools going on at the same time, actually your productivity went down. You became less productive, and then eventually you just became burned out, right? They reported that they couldn't make decisions anymore, they had information overload, and an increased propensity to leave their organisations. Now, Frank, let me explain to you why this means that I'm right and you're wrong. Right? So what this says to me is that a lot of people, my good friend Frank Prendergast included here, say that it's very important that we have humans in the loop, sorry, and that humans should review all of the AI output. And I'm reading this study, and I'm actually taking exactly the opposite from it. And what I'm saying is that where you have a task that you are comfortable with to entirely hand it off to the AI, that's the one that you should identify. That's the one that you should hand off. And if you have a task where you're not entirely comfortable to hand it off, so you have to review it, then you should be a little bit more circumspect about getting AI involved at all. You should try and do that deterministically. So you want to remove the human in loop. You want to remove the tasks where you have to—now, there are different types of tasks, right? Some of them are creative and whatever. But in general, you don't want to have to review all of the output. You want to create agents where you trust the output to be correct so that you never have to review it. Otherwise, Frank, you'll end up with brain fog.

Frank: So I would not necessarily disagree with your overall hypothesis as you just presented it. However, where I suspect we differ is your guidelines for what are the things that you would entrust to such an agentic system versus mine. I suspect that I would entrust very little entirely to an LLM, whereas I suspect you might have a broader specification of what you would give it.

Justin: Guess what?

Would Claude hallucinate less than Justin?

Justin: Actually, a quick update on my Emirates debacle.

Frank: Oh, yes. Okay. So this was: your travel was totally disrupted. Claude told you what to do. The airline was doing nothing for you. You booked a whole new travel route home, and then Claude said it would help you to get refunded by the airline.

Justin: Yeah. So I set up a scheduled job on Saturday where it checks my emails and sees if it got any response from Emirates and so on, right? And I had got a response, and they asked me to fill in a form. So I said, "Claude, yeah, just go in and fill the form." And I actually looked at it myself, and I said, "Look, those receipts don't look particularly clear to me. Can you go just check my photos and get better receipts and whatever?" So we worked on that a little bit on Saturday, which is just annoying. I don't want to do this on Saturday, right? And so we sent off all the receipts last Saturday. So Emirates have come back to me, and I had claimed it was an extra three hundred euros, and they've come back to me and said that the receipts were not clear enough, and they're only going to give me 60 out of the 300. So to your point, right, I'm at this point, I'm like, "Just why am I even having to deal with this?" So when we're done here, I'm going to be getting onto Claude. The job is supposed to run on Saturday morning, so I'll just say, "Look, can we just go review the emails, please? And can you just go ahead and just deal with this? Just deal with it. You just respond whenever the res—" So I'm going to get Claude to argue for the 300 quid, not me. And I'm not going to re—I'm not going to—why should I review? Why should I review that? Just let Claude deal with it.

Frank: Well, are you not uncomfortable with the fact that Claude could make up details that are not—that do not reflect the actual truth of what happened on your journey?

Justin: Nope. No. Okay, I'm being facetious. I'd actually—I trust Claude wouldn't make it up, right? And I actually think for a man of my age, Claude will probably make up less stuff than I do. Claude is probably more reliable than I am, so I think that's... And they don't test—it's kind of an interesting social experiment because they do these tests where they get the AIs to sort of play games against each other and whatever and see what—and this is kind of like a little social experiment test that I'm going to do.

Should you vacuum while vibe coding?

Frank: To go back to the brain fog thing, right? Having had a vibe coding experience, I'm not at all surprised about the whole brain fog thing because one of the things that I found really interesting was figuring out... Codex is amazing and originally, when I originally just said, "Look, here's my idea for the game. Go and build the rough outline for this," it just went off and did it. So it does these incredible big jobs, but then even if you're doing something small, it's quite slow. If you're just making a small change, it's still quite slow at that. So trying to figure out what you do in the downtime is a really interesting illustration of how we're just going to have to figure out completely new workflows. Because I think the study was showing that this is what was happening, was that people were filling every second of the day because any break they had, it was like, "Okay, get that agent to do this." So coffee breaks were gone because when they had five minutes, instead of getting a coffee, they were getting their agent to do something else. Or previously, when they would've had a chat over the partition with their buddy, now they're getting their seventh agent to do something else.

Justin: Yeah, and I think you bring up a really good point, which is a gap in the way that the AI products are designed at the moment. If you think of building your game as a bit like building your house, right, Codex in this instance is like the mechanical digger. It'll get stuff done really quickly and you'll have the basics of your house built super quick and it'll be super strong and it'll work well. But when it comes to the finishing touches and the smaller things, they haven't invented the little trowel or whatever it is where you do the finer details. And actually, I do think for those, it's very frustrating sometimes to interact with these AI products when you just want to change a small detail. I just need to move that little thing there. I just want to change that small thing. That's quite frustrating at the moment.

Frank: Which also leads us to—so that article was about brain fry, but we've had several articles about cognitive decline from using AI, and that was something else I noticed that was different with using Codex to using ChatGPT, because Codex is going and doing stuff for you. So you get into this kind of role of, "Let's regenerate images for Frank's sprite. Let's do this, let's do that." But then what I discovered, of course, was, I don't know how to code a game, but I do know how to code a webpage. And when I got to doing the landing page, I would find myself typing things in like, "Yeah, great, but let's make that image a link to play the game." And then I would stop, and I would be looking at Codex, thinking this through and looking at all the files, and then I'd be like, "Oh, that's just an A tag. I could've gone in and typed that in two seconds. Why didn't I just go in and type it?" But you don't think of it, because that's not the mode you're in.

Justin: And you brought up another good point yesterday. You said you were, I don't know, cleaning the toilet or something when you were—

Frank: Oh, yeah. Yeah, yeah, yeah. So in terms of these workflows and stuff, I would... If I was working on a client job and then I did something on the game, that kind of task-switching was a heavy cognitive load. We know that from research that task-switching like that. However, the best time to vibe code, I can now categorically tell everyone, is when you're cleaning the house. Because you can be doing the vacuuming, keep half an eye on your agent, and when you see it pop up as done, you just turn the vacuum cleaner off for a second, go over, and it's not the same cognitive load of task-switching as it would be on a work task.

Are camera AirPods Apple’s AI breakthrough?

Justin: Yeah, see, I think you've hit on something there. And I wonder, is voice in fact the correct mode to interact with them? Is the device that we need a TV almost, right? Or a screen, so that you can be there doing stuff around your house, you're doing something else, it's doing, and then it pops up and goes, "Bring, I'm done with that thing." And you literally just talk to it and say, "That's great. Could you do this now or that?" Or, "Give me a look at that." "Okay, fine, go do that." So you're not really doing the full context switch, you're just giving it direction as you're doing something else.

Frank: That kind of leads us into the other story that we wanted to chat about today, which was, over the course of our 100 episodes, and I'm giving myself a pat on the back, we've talked quite a few times about Apple, and you've kind of felt that Apple have completely let themselves down and are becoming irrelevant, and I've been a little bit more like, "Yeah, I don't know. I think in the long run, this might work out quite well for them, considering their positioning in terms of having beautiful products that just work." But of course, now we have a new CEO at Apple, and he is announcing all kinds of stuff that is a lot more AI-related, and I was really surprised at the first announcement that I saw, and at first I was like, "That makes no sense. Does that make sense? That doesn't make any sense." And now I think, genius. So what he's announced is that they are very close to having a production-ready model of new AirPods—

Justin: Yeah.

Frank: That have cameras in them and are AI-powered.

Justin: What? Cameras in the AirPods?

Frank: Yeah.

Justin: Wait, do they point inwards or outwards? How does—

Frank: Yes, they point in at your brain.

Justin: They do sort of read in real time your brainwaves. I mean, I don't know. Really?

Frank: No, absolutely. They point out, so it's a bit like having another two eyes on your ears.

Justin: Are you serious?

Frank: For example, you just open the—

Justin: Are they like birds? Is it like a blackbird, where it can left and right, but it can't actually see straight in front? You have to turn your head like this in order to get the full view, no?

Frank: No, it's like another pair of eyes looking forward, so you can open the fridge and you can say, "Hey, Siri, what should I make for dinner?" That kind of thing.

Justin: That is cool. Are you serious?

Frank: Yeah, yeah, yeah. And they're—

Justin: Oh, wow.

Frank: They're also talking about having a Humane pin-style pendant, and I think there was something else as well, but it's the AirPods have got everyone's attention.

Justin: I think, okay, so the male disease, right, is, "Where's my car keys gone?" They're sort of running around the place, and they're right in front of you, right? This happens to every male on the planet, I think, at this point. And now you can just go, "Hey, Apple," or, "Hey, Siri, where's my keys?" "They're right in front of you, Justin." It's like, "Oh, yeah, you're right. There they are. That's great. Thank you." It's perfect. It's like a second set of eyes for your eyes.

Frank: If it wasn't for the privacy concerns, I think another great use would be, you're at a party and you have your AirPods in and somebody comes up to you and Siri just goes, "That's Jack. He was in college with you. He j—"

Justin: Brilliant, brilliant. Actually, I've got a brilliant idea, right? I could do with that, too. A bevy of teenage children, right? To have them in, they look at themselves in the mirror before that, and Siri should be able to kick in automatic, "You're not going out looking like that. Your father will kill you. Put on some clean clothes, comb your hair, go have a shower." Actually, I think this may be a genius product.

Frank: I actually genuinely do because we've talked before about the glasses thing, and I wear glasses, so I'd probably wear glasses and be up for having the display and all that. But if you don't wear glasses, you don't necessarily want to have this new thing that you put on. But I don't know, walking around town these days, I actually don't use AirPods, I have to admit, but walking around town these days it looks like AirPods are just part of people's anatomy now.

Justin: Can you imagine if they had emotion detection on them, right? So they're there, they're talking to you, and it's just whispering in your ear, "They're lying. They're telling you bullshit."

Frank: I think the EU AI Act might have something to say—

Justin: Oh, not again. Jesus. Actually, by the way—

Frank: Wait, come here. Oh, yeah.

Justin: Did I sign a waiver for you to use my likeness on the AI Argument game? I don't believe I did.

Frank: Well, hang on now, because I think I could legally defend that with the audio clip from a previous episode of you saying, "You should go and vibe code that."

Justin: Yeah, I take it. Well, no, I didn't, but I don't know now. Yes.

Can doomers and e/accs be friends?

Frank: In the interest of completeness, because I promised that I would tell our origin story earlier, I don't know if we've ever covered it on a show or not, but I think it's a good time, episode 100, to look back to the evening that we met over a game of poker in our mutual friend's place, Ronan. And Ronan had said to me, "There's this guy Justin coming. I think you'll get on really well. He's really into AI." And so we're playing poker. I was doing pretty well, actually. I was doing pretty well at the poker until you started talking about AI.

Justin: Then—

Frank: Yeah, and at first I thought maybe it was a tactic. I thought, "Oh, he's just trying to rile me up so that I'm off my game."

Justin: It went south.

Frank: And so, at the time, what is it, two years ago, there were people at the table who had not really been keeping up with AI at all and were completely freaked out by what we were saying. And I was arguing on the side, as I still do on the show, of slowing down and regulation. You were arguing on the side of, "No, we've got to go forward, forward as quickly as we can." And I still remember I said to you, "Have you heard of the effective accelerationists?" And you said, "Oh yeah, I'd consider myself an e/acc." And I said, "We can no longer be friends."

Justin: Well, here we are. Here—

Frank: I think it was about three or four days later, I got a message on LinkedIn from you, and it said, "I know you said we can't be friends, but how would you feel about starting a podcast?"

Are we heading for ASI or P-gloom?

Justin: Half a million views later and a hundred... And the world has not ended, Frank, so I mean, that's a good thing. Take the win.

Frank: Justin, I was thinking about this, right? And I had a feeling you might bring this up.

Justin: Yeah. The fact that the world hasn't ended. Yeah.

Frank: It made me think about George Orwell's 1984.

Justin: Mm-hmm.

Frank: And we do not live in George Orwell's world. We don't have, we don't all—the entire world isn't one dystopian, fascist-style dictatorship, totalitarianism, et cetera.

Justin: No, there's four or five of them.

Frank: Well, this is interesting. This is it, because we're not literally being watched constantly by one government entity, et cetera. However, I think most people would say that 1984 was a pretty prophetic book and that instead of being watched by a totalitarianist regime, we have opted to have the device in our pockets that tracks us all the time and reports back to Meta and Google. And so the dystopian future, I think, often unfolds more slowly than we expect and in ways that are directionally right, but not necessarily as obvious as we thought they would be. And I think that's what we're in the middle of now.

Justin: All right, so here we go. Let's do this, right? Let's mark the moment. Okay, so we're at episode 100. It's taken us just over two years to get here. Let's make some bold predictions about what we think the world will look like in two and a half years, or thereabouts, when we get to episode 200. And I will give you what I think the world will look like in two and a half years, and you can give me your dystopian view, right? So two and a half years from now, we're at 2026, 2027, 2028. We're Christmas 2028, there or thereabouts. At that point, we have what everybody considers to be ASI, right? It's smarter than any single human. America and China have rushed into the lead. Europe is now a laggard, and in fact, the greatest crisis for Europe at that moment in time is not that our privacy has been invaded, but rather that there's so much demand for AI compute and AI intelligence that the Americans and the Chinese are rationing it, and the Europeans don't have access to it, and they have no AI industry of their own to provide it. Having said that, a large number of diseases are in the course of being cured or have been cured. It's obvious to people that if you survive for another two or three years, you will live for as long as you wish to live for. That, in two and a half years, may be obvious to people. That things are being produced now faster than you could ever imagine, and that the price of producing those things is falling faster than you can ever imagine. There are, I would say, within two and a half years, factories that are in the process of being built, or if not already built, where there are no humans, but they just produce goods at the other end for prices that we could never imagine. And the alignment problem and all the bad things that we think would happen never happened. And the greatest issue that we have to deal with is: how do we spread the goodness that's come out of this? There's a lot of people that maybe have less to do than they did before. How do we make their lives meaningful? And when we start to think about that in two and a half years, we will just be getting into thinking about that, we may start to realise that actually work and jobs didn't give us meaning, they just gave us something to do, and that your work wasn't your worth at any point in the past, and people will start to feel better as a result. That's my techno-optimist future in two and a half years. How about you, Frank? What is your techno—what's your view?

Frank: My fear, and this is, I actually asked ChatGPT about things we got right and things we got wrong on the show over the last couple episodes, and it said, "Well, Justin makes these bold, specifically timed claims, and therefore is wrong more often. Whereas Frank hedges his bets, so it's kind of harder to tell."

Justin: Give me an opinion.

Frank: Yeah. No, I think my fear basically is that we move towards that P-scenario that I talked about before, where we don't actually get to ASI. Which, I have fears around that as well, that if we do get to ASI, it won't be aligned, et cetera. But I think my bigger fear right now is that we never get to ASI. We just keep incrementally getting better LLMs. And so yes, employment is impacted, but not to the level that we thought it would be, and as a result, we don't ever really deal with it. So we have massive wealth disparity even worse than it is now. We have large numbers of unemployment, but not catastrophic enough for people to go, "Oh, we have to restructure society." And that, yeah, we live in this kind of P-gloom world where a lot of jobs are humans kind of making sure that the AIs don't completely screw things up.

Justin: Okay, I'm going to ask you one more time. What is your prediction for what the world will look like—

Frank: That's the prediction. That employment—

Justin: We'll be in P-gloom at that point.

Frank: We will be in P-gloom in terms of wealth disparity getting much worse, unemployment levels much higher, but not at the level that forces people to take massive societal action. And a lot of jobs, yes, being kind of like you're now supporting an AI to do the job, so you are undervalued, you're being paid less because really all you're doing is supporting an AI.

Justin: Brilliant. Well, look, I invite everybody to join us on the exciting journey for the next 100 episodes over the two years. What we've just described is what they call in economics, the K curve. The K curve. So we're here, and my curve goes like this, and your curve goes like that, and sort of makes a K. And over the next 52 or 100 weeks or so, we're going to find out which one of us is right.

Frank: I look forward to looking back and seeing. I hope you're right. I hope you're right. I'll be glad to be wrong.

Justin: Well, let's see how it turns out. Frank, pleasure as always.

Frank: Chat to you soon.

Justin: Talk to you. Bye.