Transcript
This is an AI transcription and may contain errors
Will Claude get Justin’s travel costs reimbursed?
Frank: Welcome. I’m Frank Prendergast. I’m here, as always, with Justin Collery to argue about all of the latest AI news and updates. Justin, you’re back from your secret location.
Justin: Well, only just. I only just made it, so I was at a secret location, a beautiful island in the South Indian Ocean, and I ended up actually seeing a number of beautiful islands in the South Indian Ocean because I was due to fly back through Dubai. I was literally sitting in the airport looking at Twitter, and then I could see, oh my God, we’re not going anywhere fast.
Justin: So I did what every red-blooded Irishman does when their flight’s being delayed or cancelled. I went to the bar and had…
Frank: I assume there’s probably nobody listening who was like me last week and completely disconnected from the news, but just in case there is, bombs were dropped on Dubai, or missiles were fired on Dubai. Most of them, I believe, were prevented from landing, but not all, I don’t think. I have a friend out there at the moment. Thankfully he’s fine. And he said it’s quiet there at the moment. But just for anyone who might’ve been completely disconnected from the news, basically your flight path was taking you through a potential war zone.
Justin: To a war zone, yeah. So here’s what I did. Right after I had a couple of beers and the flight was finally cancelled, we certainly weren’t going anywhere. We got taken to a hotel. It was total bedlam. So I quickly ran to the front desk, got my internet sorted out, and I fired up Claude. And I said to Claude, look, I explained exactly what was going on. I had been talking to Claude anyway through my entire holiday. And so I said, right, our flight home, it’s been cancelled. Here’s the reason why, you can go check it out. I need you to review all the clauses in my travel insurance, all the clauses in my Emirates flight contract. What should I do? So Claude goes off and does all the checking for me, and it comes back and says, Justin, you need to act really quick.
Justin: It says that previously, when this happened, Emirates will give you two options. They will offer to either A, tell you to stay in the hotel that we provided for you until we can fly you out, or B, you can just get a refund for your flight. But Claude says, you have a third option in the contract. They are part of this particular whatever agreement, international agreement, that says they have to fly you home by any carrier, not just by Emirates, and you can book it yourself if they don’t make reasonable accommodation. So the first thing you’ve got to do is send an email to them asking to be rerouted on any available flight.
Justin: So it gave me the email. In fact, it gave me two emails. It gave me the kind email and it also gave me a more sort of legalistic one. And it said, look, given that you’re under stress, maybe go with the tougher email. And I go, no, no, no need. So I got the kind email, sent it off to the Emirates people, and I got an automatic response which says, it’s Saturday, we’re not at the office till Monday, so we’ll respond to you on Monday.
Justin: So I gave that to Claude and Claude said, that’s even better. They haven’t made reasonable efforts to fly you home. Go book yourself a flight right away. So I did. So I ended up going to, I can’t tell you where I was initially because it was a secret location, but I ended up then having to get a flight to Réunion. Réunion is amazing, beautiful place. Cost a fortune, cost an absolute fortune. We’re talking many thousands to get home. And managed to get back up to Dublin. And I was on Claude again today, right? And I was sort of saying, look, now we’ve got to go through the whole process of getting the money back. Can you just double-check? The hallucinations and whatever. Can you just double-check to make sure that what I’ve got is right? And it actually drafted the whole thing for me, drafted a big huge document, said, yep, you’re totally covered. Here’s the clauses that you need to quote, and it’s actually stronger because you’re flying to Ireland instead of the UK.
Justin: And it made that sort of mistake. It thought I was going to the UK. I said, no, in fact you’re flying to Ireland, even better. Again, this is the contract, here’s the clauses, and you just send this document off and they’ll send you the money back. But they’re gonna argue about it first. But I’m home. That’s all that…
Frank: So you’re home safe. That’s the most important thing, but you haven’t been reimbursed yet.
Justin: No, but I will be reimbursed. Right. And let me just put this in context, because I could have been stuck. I mean, I could still be on that secret island if I’d hung around. I don’t think I was going anywhere any time soon. Right? I think I’d…
Frank: You’d left it up to the airline to… Yeah. If, like most people, you’re probably waiting for the airline to tell you what was going on, et cetera. Yeah.
Justin: Yeah, yeah. So my hundred-dollar Claude subscription this week has actually paid for itself 20 or 30 times over. It’s the value of thousands.
Frank: As long as you get reimbursed. And I… Stay tuned. Stay tuned, listeners and viewers. Come back to us next week because…
Justin: The cheque is in the post.
Frank: Fantastic use of Claude if you get reimbursed.
Justin: Yes.
Frank: If you do not get reimbursed, it will be a warning to everyone.
Justin: Well, I still got home, right? So that’s something, right?
Frank: Very true. That is very…
Justin: But look, I’m counting it as the cheque is in the post. It’s as good as cashed. Claude, it’s saved me a couple of grand. Well done for a…
Frank: I love your faith and trust in AI.
Justin: So look, loads going on, yeah.
Frank: Yeah, no. Yeah.
Did Dario Amodei defy the Pentagon?
Frank: So we’ve got loads of personal drama with AI going on, and loads of drama going on in the AI sphere at the moment. Over the last two episodes, we’ve been talking about the whole debacle between Anthropic and the Pentagon.
Frank: And there’s been more.
Justin: Let me just review the whole thing in case anybody hasn’t caught up, and then you can tell me what the updates are. So the Pentagon are using Claude right now, in fact, to bomb Iran. The very thing that stopped me from getting home. They’re actually using Claude to help them identify sites and stuff, but in the ongoing contract negotiations, I don’t know exactly how this came up, but Anthropic said, look, these are our two red lines. You can’t use it for mass surveillance of American people, and you can’t use it for autonomous weapons. Importantly as well, the important distinction here is that they didn’t say they never wanted their models to be used for autonomous weapons. It’s just that the current models, basically Dario was agreeing with you, the current models can make mistakes. They’re not yet reliable enough, so you shouldn’t be using them.
Justin: There was a big fight and Pete Hegseth, whatever his name is, the guy who’s the secretary of war, whatever, he said that he was going to designate Anthropic a risk to national security, which meant that they will be banned from all government contracts. And he, in fact, had a huge wide-ranging thing. He was gonna ban any company that did any business with the American government from also doing business with Anthropic. He was basically trying to kill Anthropic because they put these two red lines. That’s where we stood. What’s happened?
Frank: Yeah. And just to go back to the threat of the supply chain risk, because they threatened both… They said that they would designate Anthropic as supply chain risk, which would basically cause all kinds of red flags for people. But definitely, if you had contracts touching military contracts, you could not use Anthropic. But there was another one, what was it? The Defense Production Act. They also threatened them with the Defense Production Act, which would’ve meant forcing Anthropic to make a version of Claude that would do what the Pentagon wanted. Those two, threatening those two things simultaneously seems bizarre. Like, we’re gonna threaten you with supply chain risk and nobody should touch you, but we’re also gonna force you to give us your software. I don’t know. Very…
Justin: I don’t think it’s that bizarre, right? Sorry to get into the fight before we actually get into the fight, but…
Frank: No, great. Yeah. This is why we call it The Argument.
Justin: I don’t think so. People are going, oh my God, this is ridiculous. How could the American government do this? And they’re saying it’s gonna cause problems now with investment in AI because there’s always this sort of government sledge that’s hanging over you. I totally get it from the army’s point of view. It’s like, oh, we’ve bought your system, you’re sort of telling us now we can’t use the system in certain ways, therefore you’re a supply chain risk to us. You are a risk to us doing what is our business, which is blowing things up. So from that viewpoint, I’m going, yeah, they are a supply chain risk in a certain way, right, from that particular… I may not agree with it, but I can certainly understand their point of view.
Frank: But to threaten, say, for them to say, you know, nobody else can touch your product, even if it’s not for autonomous weapons or for internal surveillance, to say that other companies can’t use Anthropic, but then to also say, but we are gonna force you to give us Claude to use, those two things seem kind of at odds.
Frank: Anyway, they gave a deadline to Dario Amodei, and they said, you have to remove these red lines by the deadline. And Dario Amodei said, we absolutely can’t. This is too dangerous. These two very, very specific red lines are too dangerous. We cannot back down from this.
Did OpenAI just stab Anthropic in the back?
Frank: OpenAI, among other companies, during that week had signed a letter saying that they were in support of Anthropic, they were in support of these red lines, safety was very important. And then what did OpenAI do? They swooped in and they took the contract with the Pentagon.
Justin: Of course they did. They need the money.
Frank: And what was interesting was a lot of people were saying, hang on a second, OpenAI now signed a contract that has the same red lines that Anthropic were including,
Justin: They didn’t though.
Frank: No. They didn’t. Dario Amodei had said in an interview that the Pentagon had come back with this language that kind of included the red lines, but it was couched in legalese. That meant that the red lines were just kind of mentioned and were not binding in any meaningful way, shape or form, and we all have to assume that that is the contract OpenAI signed, because within a couple of days there was uproar. Apparently people were leaving ChatGPT, they were signing onto Anthropic, and Sam Altman tweeted that he maintains that he was trying to do this to de-escalate the situation,
Justin: He is such a good guy. He’s so…
Frank: So good. But he did, what did he say? He did admit that it looked sloppy and opportunistic. And I read that and I thought, yeah, if…
Justin: Oh, sorry. Did I take that 200 million quid? So sorry. I didn’t mean… Did I just take 200 million quid out of my pocket? Oh my goodness.
Frank: And worse, he’s now trying to renegotiate the contract with the Pentagon to make those red lines firmer.
Justin: Yeah, I’d say he’s negotiating really hard. I’d say he’s putting everything into that negotiation. He’s gonna sock it to the man.
Frank: Yeah.
Frank: I was reading a newsletter about this the other day, and unfortunately I can’t remember whose newsletter it was. But they said that there’s gonna be a movie made about this for sure. And they said that they’re already canvassing for Matthew Rhys to play Dario Amodei. And I happened to be watching The Beast in Me at the moment. Myself and Marcie are big Matthew Rhys fans, and as soon as he said that, I was like, brilliant. Brilliant. Absolutely. I support this. Matthew Rhys to play Dario Amodei. Fantastic.
Why did Dario Amodei have to issue an apology?
Justin: But the story has moved on now because Dario has come out today, or yesterday I believe, and has apologised. So there was a memo internally and they said that 80% of it was, was it safety theatre, right? And that the guardrails weren’t real. But it was quite a damning…
Frank: Oh yeah. He said OpenAI had lied through their teeth. This isn’t an exact quote, but basically he said they’d lied through their teeth. Said that OpenAI had given the administration, or Trump specifically, I’m not sure which, dictator-like praise. That was one that really made me go, whoa. He also said that the real reason that they didn’t get the contract was that OpenAI, or people within OpenAI like Greg Brockman, had been putting millions into Trump’s, what are they called? PACs. Donating millions, anyway. It was a pretty fierce internal memo.
Justin: Yeah, fierce stuff. So he’s come out then yesterday and he’s apologised completely, and actually it felt like a heartfelt apology. And he said, look, last week was one of the most turbulent weeks in Anthropic’s entire history, and he shouldn’t have wrote the memo and it was wrong for him to write the memo. And he is not proud of it, and he’s very sorry. So he’s done a complete 180 and he’s rolling back from the… And I think he genuinely is trying to de-escalate the situation.
Justin: Also of note this week and last week was that Anthropic services were up and down like the proverbials because of the load on their servers. There was an enormous shift of people from the OpenAI servers over to the Anthropic servers. And I think that’s interesting because OpenAI’s business model is mainly predicated, actually it’s a dangerous combination as I say it, it’s mainly predicated on consumers and now the military. Consumers and the military together, eh, that’s gonna be fine. So it’s mainly predicated on getting lots and lots of consumers. Whereas Anthropic’s business model is based on getting lots of businesses, right? They’re more of an enterprise company. And so they’re getting, I think at the moment, a lot of business which they probably don’t really want, right? They’re getting a lot of people using the free version and a lot of people using the $20 version, which is not what they’re looking for. They’re looking for the people who are paying the $100-plus and $200.
Justin: So it’s interesting to see that dynamic. I think what might happen, though, is that over time, the usage limits, you actually pointed this out to me yesterday, the usage limits on Anthropic are much lower than they are on ChatGPT, and I think this is gonna work out okay for Anthropic. So what’s gonna happen is you’re gonna get a whole load of users, they’re gonna switch over to Anthropic. The ones that are at the free or the $20 plan are gonna use their limits really quickly, and they’re gonna go, oh, I wasn’t finished doing that. And they’re gonna leave and they’re gonna go back to OpenAI. But the people with the money are gonna pay the $100 to get the plan where they never really run out of limits. And so Anthropic will win. Right. I still think Anthropic are gonna do okay out of this.
Justin: So long as the… I’ve read, there’s a guy that I follow on Twitter and he’s a legal expert, and he says that this designation is completely scarless. He said it’ll never hold up in court. And even the narrower version, which seems to have come out now, which is, you know, you can’t use any Anthropic products if you’re working with us on defence projects. The original one was you just can’t use them at all, right? But now they’ve come out with the narrower designation. He says it won’t stand up for a day in court, which I hope is true, and hopefully this goes away.
Frank: I hope so too.
Is Anthropic now toxic for Pentagon partners?
Frank: The one thing that I was wondering about to myself about all this is that, I think we talked about on the show before, Dario Amodei said that they had never received any official documentation about the threats of the supply chain risk or the Defense Production Act, that they were the same as everyone else reading it in tweets.
Frank: It struck me that, with the administration, maybe they never intended to follow through officially. Maybe, you know… But what is the damage that could be done by making these threats and by being anti-Anthropic? Because that will still instil in people a concern about using Anthropic if they are dealing with the administration on an official level.
Justin: So they did announce yesterday, they did actually get official notification from the Department of War that they were having this designation levelled against them. I still think it has to go to Congress, is what I’m reading. And what’s the damage that it can do? And I think this is probably what they’re trying to do in the Department of War. It’s just reputational risk. So if you’re a big company, right, and this is like a sword of the… I can’t say Damascus or whatever, hanging over you, it just puts a bit of extra risk.
Justin: Now, what I’m also reading on Twitter is that while the Department of War may wish to put additional risk on Anthropic just to damage them commercially, actually the effect is to put risk across the entire economy. Because if you’re any company now and you want to do a deal with the Department of War, it makes you less likely to do it. And if you’re any company that wants to invest in an AI company that wants to do a deal with the Department of War, well then you know that at any point in the future, the Department of War could do this to you. And so it adds risk to that investment. So it’s probably not a good idea to do this. But look, people are angry, so they’ll do silly…
Frank: And it’s kind of bizarre to think about the amount of damage that can be done without having to go through the legal process and just doing it via social media, which, now that I say that out loud, actually brings us nicely onto our next story in a way, because it is the Citrini Research essay, I guess you would call it, that caused a huge stir.
Did Citrini Research just spook the markets?
Justin: Yes. A brilliant essay.
Frank: Yeah. And sorry, the reason I’m linking it to social media is that it had a massive impact on the markets and it’s essentially a blog post. And what a bizarre world we live in, that tweets and blog posts are having potentially massive repercussions within the markets.
Justin: So let’s just outline what this blog post said, right? So this company actually is an investment company or a stock company, right? They’re a financial-type company and they explicitly called out that this is not a prediction. This is just one of many possible futures. And so what they do is they say, look, it’s two years from now, and then they start to go through the news, and then they sort of build up a story of how we got here. And the story that they’re painting is that they have what they called ghost GDP. So what’s happened is companies have laid off loads of staff because AI has taken their jobs. Then as those companies have laid off their staff, the companies who buy, let’s say, SaaS software have been squeezing the margins on the SaaS companies, the Salesforces and all these other big companies that make lots of money these days. And also because they’ve reduced their headcount, they just have less seats to buy anyway. And so what’s happened is this has collapsed your and my spending, right? It’s collapsed consumer spending, even though companies are making vast amounts of money now because they have these employees that just sit in data centres. And I mean, I read it and you can sort of fill in the gaps. I’ve just sort of given a very generic classification there. But I read it and, I mean, they say… They didn’t put a probability on it, but it seems quite real to me. I think it’s a reasonable prediction and I think, in fact, that we will start to see it. Their timelines were, they said the layoffs would start to happen in 2026. I don’t think that’s gonna happen. I think you’re gonna see an increase in employment in 2026, and in 2027 you’ll see what they had said would happen in 2026. What was your read on it?
Frank: Well, first of all, it’s really interesting that this essay caused such a stir and you had shares in places, I think in SaaS companies because of what you just outlined, shares in SaaS companies took a dip. Shares in credit card companies took a dip because, just because of one part of the essay that talked about how these people will have these agentic systems that are looking for the best deal for them. As they’re looking for the best deal, they will realise that credit card charges are costing them a lot of money and they will start using cryptocurrency to avoid credit card charges and credit cards will basically become almost obsolete. And so anything that was mentioned like that, SaaS companies, credit cards, anything that was mentioned, the stock prices for those types of companies took a hit, which is crazy. That’s mad.
Frank: But it’s a scary read, but also it’s kind of like, you know, we’ve been talking about this since we started the show, so it’s kind of bizarre that it made such an impact.
Justin: And good.
Frank: Obviously, yeah. I was just gonna say it’s not good that it affected the market that way, but it is good that people are paying attention now and at least having the conversation.
Why do AI forecasts now read more like sci-fi?
Frank: I think it’s written almost like… Some people have said that it’s kind of, oh, it’s science fiction because it’s written narratively, which is, again, just as a side note, I don’t know if this is happening in any other area, but within AI we’re seeing a lot of these essays that are, rather than these kind of boring, dry forecasting reports, they’re written narratively of like, okay, it’s two years into the future and here’s what we’ve just been through. And it’s almost written the way you’d write a little novel.
Justin: I think that’s because the data doesn’t exist yet. So what they’re doing is they’re projecting what might happen forward and they’re telling it in a way that’s easily digestible. And then as the data comes in over time you can sort of track the real data and have your dry report against what was the narrative that they’d built up initially. Now, a lot of…
Frank: Yeah, but I’m sure these kind of forecasting reports would’ve been written previously, but they just would’ve been really dry. Kind of like, here’s a projection, here’s a table of figures…
Justin: So let me just go through the arguments against, right? So a lot of people came out… What was good about it was that it generated an enormous amount of debate and talk about it. One of the big things, so it said a lot of stuff that I think was true, and I’d say one of the things that I think is important, one of the lines in it spoke to the fact that governments never react until the crisis has occurred. Right. And they would tell it narratively like people were losing confidence that the government could actually deal with this, the difficult teenage years, as we’ve talked about it before.
Will AI boost jobs before it kills them?
Justin: Now there’s a couple of things, right? A number of people have come out and they’ve said, you know, one of the things with all the white-collar workers, that those jobs will be overtaken by AI. And a lot of people are coming out and saying, well, that’s not true, right? If something becomes cheaper, then you tend to use more of it. And so, in particular, there was that Block company, they laid off half their staff this week because they say, look, we can just do the same amount of work with 4,000 people instead of 8,000 people. But other CEOs are coming out and going, well, if I spend a lot of money on developers right now and they’re so productive, but now with AI they’re gonna be 10 times productive, well, there’s just more stuff I can do. So I’m actually gonna hire more developers in order to 100x my output, because now they’re developing so much, they’re creating so much value. Why wouldn’t I have more of those people that are creating that amount of value, right? If they’re now creating 10 times more value today than they did last week, well then I want more of those guys, not less of them, because they’re creating so much value for me. So that’s one side of it.
Justin: The other side of it, I think they said a thing which I think is totally true, which is that at some point in the purchasing cycle, companies who buy software from SaaS companies are gonna turn around and they’re gonna say, look, we could just write your SaaS software ourselves. Why wouldn’t we do that? Right. That’s, I would say, already happening, right? I would guarantee that’s already happening. And it makes a good point about purchasing cycles tend to happen in conjunction with yearly cycles. So it’s not gonna happen, right? You’re not gonna see that in the numbers this year because those decisions were made last year before coding really changed, which was around last November or December.
Justin: Next September or October, when the purchasing decisions are being made for the following year, absolutely those companies are gonna get squeezed, right? People are gonna be saying, you can just write your software, why are we paying you all this money? We don’t need to do that any more. That is going to lead to an increase in employment, not a decrease in employment, because if you’re a company and you’re paying millions to a SaaS company and you turn around to that SaaS company and say, well, sure, I could just write that software myself, would you still need somebody to write the software? You still need somebody to be sitting in front of Claude Code asking it to do the particular thing that it needs to do. So you’re gonna see, I think, an uptick in employment as developers are hired, yes, to replace those SaaS companies. So you’ll see an uptick in employment. And then I think you probably will see a downtick in employment afterwards, but it’ll be a year later than what they talk about in the story. So I think it’ll be probably 2028 when the employment starts to go down. You won’t see it happening until 2028.
Frank: So yeah, you’re saying their timelines are out, but you’re also kind of illustrating another view on something that they kind of point out, which is that some of the downturn will be masked for a certain amount of time, which only exacerbates the problem of governments not taking action in time. And by the time, as the refrain that I keep singing on this show, by the time we see the actual effects and we need to take action, it’s too late. It’s not too late, but it means that a huge swathe of people will be negatively affected for much longer than they might need to be.
Should we plan for AI’s worst-case scenario?
Frank: And I think that’s the important thing with this essay, that its detractors, its people criticising it are saying it’s science fiction. It requires too many things to all happen at once, and this wouldn’t happen that way and this… But the authors themselves, as you said at the top, they never said this is what’s going to happen. They’re… It’s not a kind of a doomerism view of this is the inevitable outcome. They said themselves, they called it a left-tail probability, which basically means it’s not that probable, which is why it’s on the left tail. It’s a less probable negative outcome. However, the negative outcome is so severe that we should be planning for its eventuality. And I think that’s the main thing, is that this is a stress-testing scenario that we need to have some kind of mitigation for if this happens, and…
Justin: But hold on, Frank, you’re an actor, right? This is only act one in a three-part play, right? So yes, this is act one, is the difficult teenage years, right? Governments haven’t really sorted it out yet, people are getting laid off, the economy is transitioning to some new thing. Chapter two then is the age of abundance. It’s the path you have to go through to get to the age of abundance. And they didn’t write that chapter. They only wrote up to the bit where the bad bit happens. So it’s not doomerism. I still, I mean, I agree with it. I read it, I go, yeah, I think that’s pretty likely. You just haven’t written the bit afterwards where the government figures out they need to tax electricity, because all these companies are making loads of money. There’s loads of good ideas. You can tax electricity, you can create money that expires so you have to spend it so you have money velocity going around. We’ll figure it out. It’s just we need to be careful of that bit in the middle.
Frank: Yeah, but in the meantime, if you don’t have a plan, if you don’t think about it in advance, you leave these huge swathes of people exposed for far too long with no real support. So that’s my concern. And also the life of abundance afterwards is not a given either. We have to plan, we have to start planning for that to make sure that happens.
Justin: But it’s in our gift.
Frank: It’s another possible outcome.
Justin: Okay. Well I think I’m okay with that echo. I think it’ll be fine. But I do think that there, look, we’re seeing it already. Even the stuff that’s going on between the Department of War, Anthropic, right? This is big senior hurling, they say, right? There are ructions, there are things happening, right? And it’s big stakes. We’ve always got through it in the end, right? And we’ll find a way through this time as well. You just gotta go with it, roll with it. I got back from the other side of the world, for God’s sake, without losing my head. So it’s fine.
Frank: But we’ve always got through it in the end isn’t always a huge comfort when we’re dealing with something that we’ve never dealt with before. But anyway, let’s talk about some cool tools and releases.
Why did OpenAI ship 5.3 and 5.4 this week?
Frank: So first of all, OpenAI announced 5.3. I’m just trying to remember, we’re back to, what on earth are these names? 5.3 Instant, I think it was called. They released that early this week. And when they released it, they said, oh, by the way, 5.4 is closer than you think. And I thought, what a weird thing to do, to release 5.3 Instant. They haven’t even released 5.3 Thinking yet, and they’re already teasing 5.4. But of course they were, because we got 5.4 yesterday. So now we have 5.3 Instant and 5.4 Thinking and Pro, which is very confusing. I don’t know what they’re up to, but the models seem pretty good, so I’m happy.
Justin: Well, the vibes on 5.4, and I don’t use any OpenAI products any more. I’m completely an Anthropic boy. But the vibes on 5.4 are amazing. There was a mathematician who’s a proper mathematician, serious hardcore mathematician, said he had his Move 37 moment with 5.4. He had a maths problem that he had been curating for the last 10 years. He gave it to 5.4. It solved it in 10 minutes. So the vibes on 5.4 look really good. Couldn’t come at a better time for OpenAI, given everything that’s happened in the last…
Frank: True. That’s a good point.
Can AI finally generate real vector graphics?
Frank: There’s another cool new tool. Let’s see if I can share my screen, because this is a tool that I have been waiting for for a long time. This is an AI image generator, but it doesn’t generate pixel images. You know, ChatGPT, Midjourney, all the other image generators, they generate these pixel images. This generates vectors. So Justin, do you know the difference between a pixel image and a vector image?
Justin: I’ll do my best to explain it. You can tell me if I’m right or wrong, right? So if you create an image, usually with AI, it’s a bitmap, right? Which means it’s lots of little dots. And so if you zoom in very closely, a bit like looking at an old CRT screen or an old TV, you’ll see all the colours and everything, right? It loses definition. Whereas a vector image is actually lines that are drawn. So if you were to look at it, it’s actually like maths equations, so it’s lines and circles and stuff like that. It means you can zoom right in and it never loses any quality.
Frank: That’s it, exactly. That was it exactly. You can see I’m zooming in here and it’s all shapes. So this is scalable, whether I’m looking at it this size or whether I print it on a billboard, it’s going to be the same quality. And up until now we’ve had some AI vector generators, but what they’ve been doing is generating a simple pixel image and then converting it to a vector. And so the vectors are pretty awful. What this is doing is this crowd realised, well, hang on, AI is good at coding. Vector images are essentially just code. Let’s just have it coded from scratch. And when it’s free at the moment, you get, I think, 20 uses or something. So anyone can hop on and give it a go. It’s not brilliant yet, but like the rate that these models tend to improve, I think this is well worth watching.
Frank: I gave it a prompt to create a logo for us, Justin, and it came up with some pretty kind of iffy ideas. Hang on. Where are they? There, you know, it came up with some bad ones. That’s pretty… But look at this, this is genius. Let me just bring this one up. Where is it? Here we go. Look at this. Two paper clips. Paper clips, Justin, so…
Justin: I object. I…
Frank: For anyone who doesn’t get the significance, I’m sure you do if you’ve listened to any of our shows, because we are always talking about this idea that an AI would get out of control if you gave it a job to create paper clips. If you don’t carefully monitor AI, it might decide it had to make an infinite amount of paper clips and it would end up destroying humans in order to use their molecules to create paper clips. And all I’d said was, make me a logo for The AI Argument, and it came up with this idea. I love it. Absolutely brilliant.
Justin: What’s the one here, Frank? You showed us the one that was your face originally. What’s the one just one to the right of it there? This is you before the Ozempic, is it?
Frank: This, yeah, this is its interpretation of me. I have another tool that I wanna share with you, but I have to admit, I feel dodgy about sharing this.
Justin: As you’re doing that, I’m just gonna say that I’m gonna refer to you for the next segment of the show as Giorgio. And those of you who know Giorgio from Daft Punk, he gives an intro where he talks about who he is. I’m Giorgio, grew up in Germany. I wanted to find the sound of the sixties, the sixties, the seventies and the eighties, and I thought, aha, I know what it is. I sync the Moog modular to the click, and away I went. And you’ve generated the sound of Frank and Justin’s debate to the fifties, sixties, seventies and eighties.
Should this AI music tool even be legal?
Frank: This is Suno, Suno, I dunno how to pronounce it, Sonauto. And this is… I can’t, I’m actually surprised that this is still operating. I came across this just after our show last week, and I was sure it would either be locked down or, if not removed, that the guardrails would’ve been improved on it, because you can generate music in the style of anyone using this tool.
Frank: Now, brilliant and fascinating, but ethically very, very dubious. And I would imagine it probably runs afoul of at least laws in certain jurisdictions, if not laws everywhere. I know I was asking ChatGPT about it. It said, well, you know, style is not protected. So it’s in a very grey area, but definitely ethically dubious.
Justin: Let’s have a listen to just see how… Hank Williams, the first one is. Let’s have a listen and we’ll…
Frank: So this is Hank Williams singing about our podcast.
Justin: Okay, let’s go. That sounds very Hank. I love it.
Justin: We may have a new lead-in, but let’s go. That was the fifties, so…
Frank: So that’s the fifties. Let’s… Well, I skipped ahead a little bit. I went to the eighties, so one of my favourite bands.
Justin: I love it. Actually, this is quite good. That’s actually very good. You should release that.
Frank: I hopped forward a couple of more decades, and we get this.
Justin: I’m loving it. This is actually really good. This is fantastic. Fantastic.
Frank: I have to admit, I feel like I should not be using this tool. I should not be supporting it in any way, shape or form. But I also felt like I just had to try it out. I had to try it out. And I feel, I’m trying to be careful in my use of it. I wouldn’t put the songs out there apart from on the podcast. I wouldn’t download the whole songs and put them out there because I would not feel good about that.
Justin: Good, good.
Justin: Send it on me though. Send it on me. I’m not sure I have the same…
Frank: I have to assume this will get locked down. We saw when Sora launched, it launched with no guardrails and people were doing Martin Luther King doing travel…
Justin: Well, yeah. Yeah. That was wrong. So it was, but…
Frank: And then recently we had Seedance and we had that Irish director, was it Ru Robinson is his name? He created a clip of Brad Pitt fighting Tom Cruise. This seems to be the playbook now, is like launch with no guardrails, no respect for IP, grab as much attention as you can and then maybe lock things down when you’re forced to.
Justin: Well, are they? But is it free? Are they charging for it?
Frank: It is currently free, yeah.
Justin: That might be the loophole because if they’re not making any money from it, there’s no commercial gain to them. Maybe that’s their out.
Frank: Interesting. I’ll have to ask ChatGPT about that.
Justin: More on this next week, Frank. Another great week.
Frank: Fantastic stuff. I will chat to you next week.
Justin: Talk to you next week. Take it easy.
