Transcript
This is an AI transcription and may contain errors
Frank: I am Frank Prendergast. I’m here as always with Justin Collery to argue about all of the latest AI news. And Justin, I’m so…
Justin: Now…
Frank: It’s…
Justin: I know.
Frank: It’s like Christmas again in April. It’s been an amazing week.
Justin: This time last year, you and me were holding our breath for new models and were just released, and this year it feels like we’re hyperventilating because there’s so many models coming so quickly.
Are GPT-5.5 and Image 2.0 a knockout combo?
Justin: This time last year, this week, exactly, GPT-4.1 and o3 was released. This week we got GPT-5.5 and a new image model from OpenAI. And you have been using both of them to amazing effects. How are they?
Frank: I love them. I love them. I’m a ChatGPT guy anyway, and they’re fantastic. I actually haven’t been this excited about a release for ages, so 5.5 is the thing that they’ve been calling Spud as well. So I just knew, I knew from that name, as an man, I was gonna love spot. And it’s true, it was true. But even I wasn’t expecting it. I thought I knew, like the people that we follow were all saying, oh, there’s gonna be a big release this week. But I don’t know. OpenAI seemed very quiet. They normally have a lot more cryptic clues, et cetera. So they dropped the image model. And to be honest, that was so good. I love image generators.
Justin: Yeah.
Frank: And it’s the best image generator now that we have. So I think Nano Banana two had that crown, and they have definitely taken that crown from Nano Banana two.
Justin: Took me by surprise, the image generator. Right. It really took me by surprise because they killed a whole load of stuff just recently and I thought that was it. They’re not gonna do image generators anymore, they’re just gonna focus on coding. Just like Anthropic? No, they were just clearing the decks to bring out an even better image generator, which is cool.
Has Nano Banana lost the image crown?
Justin: And some of the images that have been generated have been, okay, so let me, we should describe, so image generators over, while they’ve struggled with things like text, they’ve struggled with consistency, they’ve struggled with whatever. And what I’m reading, and then you can tell me your experience. But what I’m reading is that this is like a reasoning model on top of an image generator. So it’s very good at keeping one, it’s very good at keeping consistency. So if you wanted to change stuff. And I’ve read that it’s very good at that. I have seen examples both by you and by Ethan Mollick, where it’s made storyboards and stuff like that. And then it’s got text in the storyboards and the text is hilarious. Because the people have asked them to write funny things, but it’s perfect and it’s hilarious. And it’s a step change above what we had before. What’s your experience been?
Frank: Yeah, a hundred percent. It’s a natural progression from the last image model from OpenAI and in terms of Nano Banana as well. We’ve seen Nano Banana do really good infographics and comics and things like that, but this is just another step forward again, in that way. The amount of text that it can handle. And also what I’m finding is that it is much better at the visual hierarchy of what it’s doing. So AI images can be very busy, very messy, but personally I think this is actually, it can do a busy image and it can still have a kind of a logical hierarchical structure so that it makes sense. So I’ll show you something that I, if, and…
Justin: Yeah. Yeah. Let’s have a look.
Can ChatGPT design without direction?
Frank: For anyone who’s listening on the audio podcast, we’ll chat through it, but it might be worth popping over to YouTube and having a look. So this is an infographic for The AI Argument with an illustration of you and me arguing and underneath some of the stuff we argue about. Now, why I find this interesting is that, like you said, in terms of it being a reasoning model, I intentionally kept the prompt for this really vague. I just gave it all of our transcripts. I gave it a briefing document that I previously had created to explain what The AI Argument is, and I just said, look, oh, and I gave it my brand colours, the Frank and Marci brand colours. And I said, look, just create a compelling infographic based on the transcripts of this podcast, and it just did this. So is this exactly how I would represent The AI Argument? No, not necessarily. I mean, I don’t know what useful, not wholly means. It seems to think that AI might become some kind of deity that people would… That’s a theme that it seems to come up with. I don’t know. Did we talk about that at some point in the argument, in the show? Maybe we did. My memory isn’t the best, so…
Justin: I don’t know. But if, as I’m looking at this image, right? So again, for people who are watching, we’ve got The AI Argument. It looks cool. You’ve got pictures of you and me, which are actually really good in terms of likeness, right? I would’ve said they’re not, you know, they’re, it’s whatever, it’s stylised, but it’s definitely us. And then underneath, you’ve got six cards, right? One, two, three, four, five, six. The text is great, right? It’s literally great. It’s something I would, now this is your world, right? This to me is like something a graphic designer who is looking to create a cartoonish infographic type thing would create.
Frank: Absolutely, totally agree. There’s loads of text on this as well, and someone on LinkedIn just pointed out to me that there is one error on it, which is that in the bottom left-hand corner it says something about tasks, tasks taken on, on a kind of a machine that the AI is automating some job or other there. But I don’t know, maybe task taker is just a job that I’m not aware of in manufacturing.
Justin: Maybe. Now tell me this, right?
Frank: Yeah.
Justin: If a graphic designer was to make this, how long would it take them to do it?
Frank: Well, it would take me a very long time. Maybe somebody more skilled would be able to do it faster than I would, but this was done by ChatGPT in like two minutes or something.
Justin: Okay. And that’s my other question, right? Because where do you go from this, right? The image output is so good. I’ve seen so many images and they’re so perfect. The only thing, I mean, you can make it a little bit better, but it’s incremental gains. Now it comes down to speed and cost, I would’ve thought. Can you, how quickly, so how long did it, how long between the time when you entered the prompt and then actually had a finished image that you did? Sorry, two questions. One, was this one prompt and you got this out, or did you have to refine it a bit? And two, then how long did it take for the image to actually pop out?
Frank: This was one really vague prompt with loads of context, and I did no refining nothing. I just wanted to see what happens when I just throw all this stuff at it. And it was fast. I honestly don’t know, but it was really fast. Because I’m very impatient. So I notice when I’m sitting there and it’s generating, thinking, et cetera. This was fast, you know?
Justin: Cool.
Can ChatGPT script and illustrate comics now?
Justin: Did you make any other images that you can share with us just now?
Frank: I did, I did. So a buddy of mine, Jeff, he saw this post and he went and he created a comic based on one of his blog posts. And so I thought, okay, I’m gonna give it another really vague prompt, and I asked it to create a comic where you and I build our own artificial super intelligence. And I, again, really vague prompt. I did not tell, I just said think about the arguments that we have and therefore you know what we would be doing. So I won’t go through the whole thing. But I did find this very funny. Panel one, because I did not tell it. I told it to come up with the story itself. So panel one, it says, right, we have a bright idea and I’m saying, if everyone else is building one, maybe we should build one that argues back. And you say, grand. We’ll make it useful first and apocalyptic later.
Justin: Oh, that’s so good.
Frank: Panel two here, we’re working away on it and this is my favourite panel. Three. First contact. We actually create our ASI and it pops up and it says, hello Frank. Hello Justin. I have analysed your arguments. You are both exhausting.
Justin: Oh, that is, yeah, you will send that onto me because that was printed out. It’s fantastic. It’s really good.
Frank: Then I went from then, of course, right? I was thrilled with the image generator. I was having difficulty getting back to doing some actual real work instead of making comics of you and me.
Justin: Yeah.
Did GPT-5.5 just beat Frank at web design?
Frank: And so I got back to work and I was focusing on my work, and of course, what happened, they released 5.5 and I just had to see what that could do as well.
Justin: Yep.
Frank: So I did a really similar thing. I just took all of the transcripts, the briefing documents, the colour palette, the picture of you and me. And this time I said to 5.5, listen, would you ever go away there and would you make me a website, a webpage for The AI Argument? And it sure did. Now, this one I did a little bit of back and forth with, but the only back and forth I did was that it messed up the images the first time. And I much prefer this image where I look like the Hollywood version of myself.
Justin: Yes.
Frank: Marci looked at this and she said, right, that’s the version of you if you went to the gym occasionally. Is it.
Justin: Yeah. Yeah, it’s good. Do you see that actually, would that be a very interesting test? Because you know, on the previous image generators, there was one of the tests where it was like, yeah, just take this person and make them nicer, or take this person and you’d do the same, you’d feed the same image back into it over and over again and you’d have a normal person sort of slowly morph into this sort of horrible…
Frank: Yeah.
Justin: Whatever. Be interested to do that. But you look great, Frank. Good job.
Frank: Why thank you. Why thank you. I’m going to show you a few things about this that I think are really interesting. So what I find really interesting is I gave it a really vague prompt, and it did some pretty cool stuff. So it’s got this nice thing here where if we click on topics here, it’ll scroll down to the topics that we discuss. And you can see here now, I did tell it to embed the latest YouTube, but I just gave it the link to the channel. So it figured out how to do that. I just gave it the link to our Buzzsprout and I said I wanted the audio player embedded. It figured out how to do that. It did this really nice thing here. This subscribe button actually goes to the YouTube channel, but it initiates the subscribe action. So if I click on this, I go to YouTube and then the user just has to confirm, yes, do you really want to subscribe? I mean, I didn’t even know you could do that.
Did ChatGPT add features Frank never asked for?
Frank: And then here’s what I was really impressed with. So when you get down to the end of the page here, it says, more arguments to catch up on. And you can hear, you can see it says, view all episodes, and there are links here to different episodes. And so I assumed, all right, well these are gonna go to the YouTube episodes.
Justin: Yeah.
Frank: I clicked on one out of curiosity and it doesn’t, it goes to my blog post for that episode. Now, I never told it that the episodes were on my blog. I never told it what my website was. So it went off and it researched The AI Argument and decided to do this itself, that this would be a good feature. That’s what’s impressive about this.
Justin: I have to stop you there, Frank. Tell me this. Are you not petrified by that?
Frank: I’m too excited this week. Ask me next week when the excitement wears off.
Justin: I think that’s cool, right? And I love the creativity, but that’s exactly the sort of thing that when it goes off on, oh, I need to buy a 747 for Frank, but I don’t have his credit card number, but hold on a second. I can just go over here and find it and it’ll go and do that thing.
Frank: Yes. I mean, in the context of what I’m doing here, the guardrails are all in place that it’s okay, but yes, this is the kind of thing that in the wrong environment with the wrong set of guardrails. Yes, absolutely. AI going off and just thinking through things itself and deciding what the best path is. Yeah, that does scare…
Justin: It is. This is…
Frank: Don’t ruin my week.
Justin: No, I’m not gonna ruin your week. Right. But this is, can I just say again, for people who are listening and can’t see it, it is beautiful, like absolutely beautiful. That, to me, is a top quality website. It’s something that I think a couple of years ago, you would’ve paid, I don’t know. Again, this is probably closer to your business than mine, but that many, many thousands to make something like this.
Frank: I’m gonna be completely honest here. I put a page for The AI Argument on our website and I had to do it fast. I couldn’t spend too long on it. The AI Argument isn’t a huge money maker for us.
Justin: You know, there’s all those sponsorship deals I’d never tell you about, Frank, but don’t worry.
Frank: But I, at the same time, I knew it had to look good and I put a bit of effort into it. But this, that ChatGPT did in minutes. So much better. It’s so much better than what I did.
Justin: Are we going to use this, Frank?
Frank: You know what? I don’t know. The other thing about this is I wanted to be able to show it on the show today. And in order to do that, when you do it locally, the YouTube embeds break.
Justin: Yeah.
Frank: But ChatGPT said, oh yeah, but you could just throw it up on GitHub pages. And I said, okay, great. Tell me how to do that. And it did. So we’re looking at this now on GitHub pages, thanks to ChatGPT.
Justin: It’s really, really good. I mean, really, really good. I’m blown away.
Can ChatGPT develop our podcast game?
Frank: So I’m gonna show you one last image before we move on from OpenAI and my excitement about all the releases. So next I thought, right, I don’t know where this idea came from. I think I was watching videos from Wes Roth and Matthew Berman, and they were looking at some of the examples that people had done with ChatGPT and they had created these games and I thought, ooh, what would an AI Argument game look like? So before you get too excited, I have not developed the game yet, but I do have some concept art.
Justin: Okay. Okay. Let’s.
Frank: So. Here is the concept art for The AI Argument game, right?
Justin: I love it.
Frank: I’ll tell you what’s going on here. We’ve got AI being built in the background here by the AI labs, and here you are, Justin, with your mallet. You are playing whack-a-mole here with all the AI-related problems that pop up like deepfakes and hallucination…
Justin: I love it.
Frank: Agents. And meanwhile, I am up here gathering all these bits of regulation and putting them together into one coherent document. So the game is you have to swap. You can only play one character at a time, and you have to swap between them because you have to whack the problems as they arrive and meanwhile regulate for the entire industry before AI grows out of control. And I’m currently asking ChatGPT to build this.
Justin: I thought, brilliant. Absolutely. You should. Now, these are the finer details, and this is the problem with the way that my brain works. First off, right? First off, right. Let me again, just to describe, we have coming down from on heaven on high, we have various pieces of legislation and safety nets and stuff like that. They appear, they’ve got sort of stars around them, like there’s some sort of thing from on high, some sort of a holy, like it’s a good thing, is my problem. It’s almost making them out to be good audits, and it’s like, oh look, we’ve got an audit. Woohoo. No, audits are horrible things. I don’t know why they think that’s so good. So I got that part also a question, and this is what we’re getting more into the gameplay. I love the whack-a-mole, by the way. It’s great. More into the gameplay, which is I hope that AI growth and regulation are linked to each other so when one goes up, the other goes down. Because that would make a lot of sense to me.
Frank: The idea is that you have to regulate it before AI grows too big and out of control. So you’ve got to keep the problems in check, and then you’ve gotta regulate the AI so that as it gets bigger, you’re controlling it.
Justin: And to be frank, I, which I think, you know, over the weekend set GPT, this is like, okay, I’m totally blown away and looks going, wow, this is a step ahead, right? There’s nothing I’d be going to get it to write the game. This could be the money spinner that we’d be missing from The AI Argument.
Frank: It might be. It might be.
Justin: This, this is, I love it. It’s absolutely fantastic. Okay.
Frank: So yeah, I am also, I mean, I’m blown away. I just, like I said, this is the most excited I’ve been about releases in a long time.
Was OpenAI cherry-picking the benchmarks?
Justin: I can see that. Now to bring us back down to earth. Although this comment now is almost meaningless, right? Because when I was reading the release card and stuff like that, they were very selective in the evals that they published to show that they were better than Anthropic’s models. And I was looking at it going, oh, they’re cherry-picking the evals, you know? Because there was only four. Normally you’ve got, like in those eval cards, you’ve got 10 or 15 different benchmarks that they test against. And in this one where they went head to head with Anthropic, there was like four, right? And they beat them in all four. And I’m going, yeah, they’re just showing you those. But now that I look at that stuff, I mean, out of the gate, that looks to me. I mean, the images are very, very impressive. But the website functionality to me looks, that was their weakness previously was that the front end wasn’t as good as the Anthropic front end. That front end there was easily as good, if not better, I think. I think part of the skill of what you did there was by giving it the brand guidelines, which I suspect are very well written, and it’s followed those really…
Frank: Honestly, I only gave it the colours. That was all I gave it. I just gave it a bunch of colours.
Are vague prompts better than specific ones?
Frank: Now, here’s one thing that is worth considering though. I kept the prompt really vague, as I said, just because I was curious what I could do out of the gate with just a bunch of context and very little direction. So one thing, one possibility is that the models can be good when given complete free rein. And actually they might not be quite so good when they’re trying to follow a user’s very, very specific instructions. So that might be my next experiment would be, you know, how good, how well does it do the job with a very, very highly specified project brief from me.
Justin: Yes. Interesting.
Frank: Because it might just be defaulting to everything it’s good at when you give it a broad prompt.
Justin: Yes. Interesting.
Is intelligence really getting cheaper fast?
Justin: Now, and then just to put this in context, right? I mean, I’m blown away, but I’m not surprised. I’m very excited now too. To put this in context, this day, last year, I think, as I said at the start, o3 was released, right? The relative cost of o3 to GPT-5.5, according to Sam Altman. I haven’t tested these numbers. He said it’s 10,000 times cheaper. It’s certainly, I would say, a hundred times cheaper. So, and what I mean by that is if you got o3 that was released exactly this time last year and got it to produce an output that was as good as that, which would’ve taken it a very long time, it would’ve cost you a hundred times more in tokens. I am more than confident, more. In fact, many hundred times more in tokens and those tokens would’ve been more expensive. So this is, that’s a really good example of this thing where we say intelligence gets a hundred times cheaper every year. And intuitively, that can be difficult to understand. Because it’s like, yeah, but tokens are more expensive now than they were a year ago. And I’m spending a fortune on them. My bill on tokens has gone through the roof. How could you justify that to say that it’s a hundred times cheaper? That’s how. If you wanted to create that thing a year ago, it would’ve cost you easily a hundred times more in terms of tokens. So, incredible progress, like super impressive.
Frank: Yeah, it is. It is.
Is the human bit the future of work?
Frank: But I guess it’s to burst the bubble a little bit. You were asking me a minute ago, well, does that not terrify you? And so this brings us back to the question of what is the impact on all of this on jobs? But I believe you’re feeling a little bit more optimistic. Okay. Yes. Talk, talk to…
Justin: I always feel, I’ve always, I never felt anything other than optimistic. I think that’s the whole point. Right. I’m delighted. It’s just that I don’t have the science to back me up. So luckily this week we do have some science to back me up. There was an economist, which is called the dismal science, I understand. And his name was, hold on now. His name was Alex Imas and he’s a professor at some university and he makes some brilliant points. And so this isn’t me making the argument because my economics is dodgy at best. This is Alex Imas, who actually is an economist. And so what he was pointing out was basically the world is run by human nature, right? And he said, from an economics point of view, when everything you know becomes abundant. So if you live in this world of abundance, which I have no doubt that we will in five or 10 years live in a world of abundance, then the question you have to ask yourself is, well, what’s scarce? And in that world, what becomes scarce is human attention. And by human nature, right? Humans are, people always want to get one over on the Joneses. We always wanna look better than the next guy. We always wanna show our status relative to somebody else, right? So now we do it with fast cars and big houses and stuff like that. And if those all become cheap, well then, you don’t do that anymore. Instead, you’ll have 50 housemates that have no cleaning to do around the house or whatever it is, right? You’ll be getting into fine art that is produced by an artisan that lives in the South Indian Ocean and there’s only one of the, you know what I mean? Those are the things that become scarce. It sounds horrible actually, when I say it like that. I don’t wanna live in a world where people are…
Frank: I’m really fascinated by what we all, we always seem to take very different things away from the articles that we’re reading. So the thing that jumped out to me was the point about, and this is the point that you made, like in some of the very first shows that we recorded, the human relational element. And how much more important that will become in an AI world. And so I was interested in his take that he now thinks that yes, there will be an impact. However, if you can find the human relational part of your job and give the rest of it to AI, but retain that. So he’s kind of saying that the jobs that have no human relational aspect, they are in danger. But if you can find a role that has that human contact, that needs that human contact or find the parts of a role that have an essential human component. That’s where we need to be focusing on, and that’s why he is more optimistic now that it doesn’t have to be an absolute disaster.
Should AI disruption happen faster?
Frank: But I will say that he did say that it depends. His optimism depends on the speed of development and the speed of progress. And so to go back to what we’ve been talking about in some recent shows, Bernie Sanders saying, you know, stop building data centres. Let’s slow down. Let’s figure this out. Let’s move forward carefully and cautiously. I mean, ultimately that’s what this economist seems to be saying to me.
Justin: No, that’s the wrong approach, right? Because right, you’re gonna have the period. So let’s say you live in a world where 1% of work can be automated with AI, which is probably where we’re at now. Right. And let’s say you know that at some point in the future we’re gonna get to a world where 95% of work can be automated by AI. There’s a big long transition between those two points where you’ve got 5%, 10%, 15%, 20%, right? Especially those first 20 or 30 or 40 percentage points. And during that period, you get huge unemployment. But you also have still a lot of people employed and they’re going, well, I don’t wanna pay more taxes just to pay for that lazy bum and all the usual sort of social stuff that goes on when there’s like an economic crisis happening. And until you get, the vast majority of people are unemployed and then the world will sort itself out. The Lord, like it’s what a cruel attitude to have, Frank. Why would you want to put those 20 or 30% of people through the longest period of torture that you possibly could? Because the longer it takes to get from the 1% to the 95%, the more time those people are gonna spend in purgatory. So the best thing you can do is get from 1% to 95% as fast as possible so that that slope is as steep as possible, and we get through the economic catastrophe as quickly as possible.
Frank: So that’s an interesting hypothesis, but I’m a little bit concerned about the fact that earlier you said that you were not an expert on economics, and I’m contrasting what you were saying with this expert on economics that you introduced me to.
Justin: Yeah. What does he know?
Will AI micromanage us before replacing us?
Justin: Just, by the way, some other news. Yeah, go on.
Frank: Well, I was just gonna say interestingly, Jensen Huang was also making the headlines saying that he didn’t think there was gonna be a jobs apocalypse either. He, I can’t remember where he was on stage, but he made these comments about how there’d be loads of jobs, but AI would be micromanaging. In fact, he said right now, I think he said, even right now at Nvidia, AI agents are micromanaging employees, which sounds like my P gloom scenario.
Justin: It does sound very like that. And just to further enforce this point, Meta this week, who at the start of the week announced that they were, at the end of the week, they laid off 8,000 people, which is terrible for those 8,000 people. But at the start of the week, they said they were going to put software on 8,000 or 10,000 employees’ machines to track every single mouse and keyboard thick and everything that they did with the express intent of at some point in the future, replacing them with AI, which is, I mean, which is worse, right? To be programming the AI that’s gonna take your job or to be overseen by an AI that’s telling you what to do. I think they’re both horrible outcomes. Can we just move on to the really powerful AI where I don’t have to do either of those as quickly as possible? In my world, that’s what would happen, but not with you, Frank. You want to put people through that sort of purgatory for as long as possible.
Frank: Who’s…
Justin: On the side of good.
Frank: Okay, look, I’ll be on your side if you can guarantee me that Mark Zuckerberg and Jensen Huang and Sam Altman are all gonna have a fund that is going my and my groceries for me.
Can taxing electricity offset AI disruption?
Justin: Again, for those who haven’t listened before, the correct solution to this is to tax electricity. You tax electricity and you redistribute the wealth from electricity. For that reason, we should be building lots and lots of nuclear power stations in Ireland and we’re not, and I think that’s a terrible thing because at some point you’re gonna have to find a way to fund all of these people that don’t have jobs anymore, and either you tax electricity or you tax tokens. Electricity is just a proxy for tokens, but you’re basically putting a tax on the AI companies. If you don’t have any electric power stations, it’s very hard to tax either tokens or electricity. We should start building now.
Are world models Europe’s AI moment?
Frank: So we had a bunch of stories that there’s no way we’re gonna get through. But I’m just curious, are there any of the other stories that we had that you wanna give a quick mention to?
Justin: Yes, there is. Go Europe. Go Europe. I’ve been giving out about Europe and the regulation for the longest time, but like buses to Europe. Well, one sort of European company and very, very much a European company released papers and got money this week and stuff. The first of them is called Odyssey and they’re based out of California and London. And what they make is they make an AI, which predicts the future of physical objects, as in if you push the cup, will it fall off the table, right? It’s a real world model, right? And do that. And they seem to be making great strides. So that Go Odyssey. And the second one is Yann LeCun out of Paris has released a paper. He’s got funding recently. His paper is called Lay World Model because he’s French. And why would you call it anything else? It’s brilliant. And again, this is about a real world model. It’s about understanding the real world. It can be trained in a couple of hours on a GPU. It just has 15 million parameters, billion, I guess. And yes, it’s gonna be really good for robotics and stuff like that. So go European AI companies. Brilliant. How about you? Anything that caught your…
Frank: Yeah, we’ve both kind of been fans of Yann LeCun from the beginning, so yeah, I really want him to.
Justin: He’s very grumpy now. I wouldn’t be going out for a drink with him on a Friday evening. Very grumpy man.
Was Mythos hacked by guessing the URL?
Frank: I would like to just very quickly talk about, seeing as we spent so much time on OpenAI at the start, what are Anthropic doing? And it’s two things related to Mythos that I thought were really interesting and kind of related. First of all, we’ve talked on the show a lot about how the US government designated Anthropic a supply chain risk, and that meant that nobody in the US government could use Anthropic’s models and anyone who was engaging with the US government on critical projects could also not let Claude touch those projects. And what do we find out? We find out that apparently Mythos is being used by the NSA. So Mythos is the big cyber capable model, the most dangerous model ever released, et cetera, et cetera. Only available to 40 or so companies. And apparently the NSA is quietly one of those companies. I…
Justin: Damn right. Damn right.
Frank: And simultaneously, there was another story that I think we talked on the show before about the, I was telling you about a security person in cybersecurity who was saying they didn’t even want access to Mythos because they did not want to be a vector for attack. They didn’t want somebody trying to get access to Mythos through them. And apparently that is exactly what happened to Anthropic. One of their partners had, someone managed to hack into one of their partners and figure out an access route to Mythos. Great.
Justin: Do you know two? Do you wanna know how they did it? I tell you how…
Frank: I do, yes. Do.
Justin: This is ridiculous. Right? So you can imagine if you were accessing, let’s say, the OpenAI models and the URL to access the model was openai.com/gpt 5.1? 5.1, 5.2 for 5.2, 5.3 for 5.3. 5.4 for 5.4, and let’s say 5.5 hadn’t been released yet, but you knew that people were testing it. What would you guess the URL is, Frank?
Frank: Hmm. Let me think, what could it be? That’s a very hard one to crack. Would it be 5.5?
Justin: It would, that’s exactly what they did. And it worked. That’s how hard it was.
Frank: That’s crazy. That is really crazy.
Justin: Yeah. That’s it.
What if you were put in an AI drama without consent?
Frank: Final story.
Justin: Yes. The one place where I get most of my AI news is, of course, the RTÉ, Ireland’s National Broadcaster. How about you?
Frank: Yeah, they have some great AI stories from time to time. They really do.
Justin: This week is no different. So talk to me.
Frank: Christine Lee, who is apparently a model and an influencer, discovered that she was acting in a micro drama, an AI-generated micro drama, playing a cruel character apparently. And she was not impressed. Because she might be a model and an influencer, but she’s not an actor. She’s never done any acting. She did not sign her likeness up to be used as a cruel character in a Chinese micro drama. And it was on a platform that I hadn’t heard of, but it’s owned by ByteDance, who make TikTok. And she’s suing both the creator and the platform because she did not authorise anybody to put her in a micro drama. How would you feel?
Justin: She has this all wrong. All wrong. I think this is a great opportunity. A great opportunity. Let’s imagine we’re sitting down to watch, I don’t know any movie. I’ll tell you what, what’s that one from years ago with Jack Lemmon and the other dude, and they’re on the train and one of them.
Frank: Oh no. I thought you were gonna say Grumpy Old Men. I was like, oh yeah, that’s us.
Justin: Well, okay, Grumpy Old Men. Can you imagine watching Grumpy Old Men? But our faces can be part of the movie. Wouldn’t that be cool? Right. We’re sitting…
Frank: Yes, yes. If it was, yes, if it was a conscious choice.
Justin: Let’s say we sit down this evening, right? And we wanna watch, I don’t know, Rocky IV or whatever, and the one with the Russian dude. And you could be the Russian dude, and I could be Rocky and we could be there budgeting out in the middle of the ring. Wouldn’t that be fantastic? I mean, just think of the possibilities. I think it’s good news.
Frank: But if you take it to the nth degree, like if you paid, you know, 20 bucks to go to the cinema or whatever crazy amount it costs now. And you go into the latest Disney Marvel movie and there you are as the villain and you’re thinking, oh, well that’s kind of cool. I always wanted to be a Marvel villain or a Marvel superhero. Brilliant. But then you think, hang on a second. How many millions are they making from this movie? And where’s my cut? I just paid 20.
Justin: Frank, listen to me. There’s more to life than money, right? Imagine you’re sitting down this evening, right? Imagine you’re sitting down this evening yourself and Marci, right? And you decide to turn on, let’s say, Titanic. But instead of Leonardo DiCaprio and Kate Winslet, it’s you and Marci. You’re there watching the movie together. It’s a beautiful thing. It’s like, oh, Frank, you’re way better than Leonardo any day. I mean, you would get so much more out of that than just a dirty old paycheck.
Frank: I think the choice, though, I think the choice is the important thing. I do think that creator should not have just grabbed somebody’s likeness at random and popped ’em into his show.
Justin: Well look, it’s whatever you look as. You are a professional actor, so we have to respect your image. Me, on the other hand, I am a who to the internet gods, and the only reason I do this is so that somebody by possibly steal my likeness. So you take, take, take me as I am.
Frank: Would you genuinely know? Would you be grand if one of your kids was scrolling TikTok and went, Justin, that’s you.
Justin: They already scroll TikTok and go, Justin, that’s you.
Frank: Oh man. Fantastic. Alright, I’ll chat to you next week, I’m sure. Yes. We’ll have loads more to argue about then.
Justin: Absolutely. Have a great week. We talk to you.
Frank: Cheers.
