The AI Argument

Google’s ‘Good Enough’ AI, Anthropic Profits, and Monet Gets Roasted | EP101

• 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 | 36:31

Why did Google I/O feel so underwhelming? 

Frank and Justin unpack a massive amount of Google AI announcements: huge AI-powered changes to search that could change the web, the muted response to Gemini 3.5 Flash, Gemini 3.5 Pro being pushed back, Gemini Omni’s impressive video and world-model claims, Spark’s carefully staged agent rollout, and Demis Hassabis saying we may be in the “foothills of the singularity”. 

Plus: Anthropic reportedly moves into profit while OpenAI burns cash, Andrej Karpathy joins Anthropic to work on recursive self-improvement, OpenAI starts selling future compute contracts, Elon loses his OpenAI court case, and AI haters accidentally roast a real Monet painting.

Who wins the AI race from here: the focused labs like Anthropic and OpenAI, or the tech giants with ecosystems too big to fail?

00:00 Why is Google I/O so uncool?
01:06 Is Google search eating the web?
03:07 Is Gemini 3.5 Pro just not ready?
05:03 Is Omni the Nanobanana for AI video?
06:36 Are we in the foothills of the singularity?
09:23 Is Gemini 3.5 Flash fast but meh?
12:26 Is Spark Google’s safer OpenClaw?
15:16 What work should agents actually do?
17:00 Can Google win by being good enough?
21:08 Did Anthropic prove AI can pay?
24:46 Did Anthropic just land Karpathy?
26:57 Is OpenAI selling tomorrow’s tokens?
29:50 Why did Elon lose his court case?
31:49 Did AI haters accidentally roast Monet?

► 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/

Transcript

This is an AI transcription and may contain errors

Why is Google I/O so uncool?

Justin: Hello, good afternoon, good morning, good evening, and welcome to The AI Argument with myself, Justin Collery, and the ever-vigilant Frank Prendergast. Frank, good evening. It’s been another week of releases and new toys for people to play with, mainly from Google. How did Google get on this week?

Frank: Yeah, so they had their Google I/O developer conference and Google are weird. Google are just weird. To be fair, they announced loads of very cool things. We’ll cover some of the ones that caught my attention. They had all kinds of announcements about their changing search. They’re doing the biggest change to search in 25 years or something. But I’ll be honest, I don’t care. That’s kind of what pervades a lot of this. They announce loads of cool stuff, but they don’t seem to make it very cool sounding or cool seeming.

Is Google search eating the web?

Justin: Just on the changing search for evermore, how are Google planning to make money, do you think, if you don’t have to click on any websites anymore? Because the AI just summarises the website and gives you the content to your face without ever going there.

Frank: That’s a good question. I will refer you to my previous comment about I don’t care.

Justin: Tell me this. Another thing then that you might care more about or know more about certainly than me is, was there anything in the I/O conference about SEO in this new world where it’s not going to be optimised for search anymore, it’s going to be optimised for the AI summary, which you’re going to get and the view instead?

Frank: Well, I don’t know if they talked about it specifically at I/O, but yes, it’s being talked about everywhere outside of I/O. Not even just because of I/O, but AI has been prompting this conversation for a long time. And of course, now we have AI… I can’t even remember the stupid terms that are coming up for AI search optimisation. But yeah, you have all…

Justin: Call it E-E-I-E-I-A-I-O or…

Frank: Exactly. Exactly, yes. But Google did actually release… I say I don’t care, but some of this stuff does reach me, and Google did release a kind of, “Look, here’s what you need to be doing to be catering for artificial intelligence as well as traditional search.”

And a lot of people were like, “Yeah, this is what we’ve been saying. There’s no magic snake oil for AI. This is all the stuff you should have been doing for traditional search as well anyway.” And then, of course, you have the camp who are selling, I guess, probably selling AI search services saying, “Yeah, but it’s in Google’s interest just to say that, but actually there’s loads of secret sauce that we can sell you.”

But yeah, my main thing about Google I/O is just that I’m surprised. I just can’t believe how they just can’t make themselves cool.

Is Gemini 3.5 Pro just not ready?

Frank: The big thing that everyone was talking about, unfortunately about I/O, was that Gemini 3.5 Pro was not announced at the developer conference. So they announced the 3.5 family. They are releasing 3.5 Flash, which they are saying is super speedy and really good and blahdy, blahdy, blah. But 3.5 Pro, give us another month. It’ll be with you soon, we promise.

Justin: I’m going to knock a conspiracy theory on the head here straight off the bat. So there was some chatter, and it seemed somewhat reasonable to me, that maybe they hadn’t released 3.5 Pro because it was on a Mythos level, right? It was going to be as capable as Mythos. And yesterday morning was supposed to be the morning where the executive order from the US was signed and brought into force, where any model that was sort of Mythos-like capable size and so on would have to go through 90 days of testing by the US government.

90 days of testing, being made available to the government and what do they call them, banks with systemic interests and so on. But that executive order’s been cancelled. They’re not going to do it now.

Frank: Oh, wow. Okay. I missed that. That’s interesting.

Justin: So the upshot from that is the reason that 3.5 Pro whatever wasn’t released is not because of any potential executive order, it’s just not ready.

Frank: Something I read, I think it was Business Insider had an article, and their theory was that they wanted to get 3.5 Flash out there and into Antigravity, their coding assistant, and then use the feedback loop to improve 3.5 Pro specifically for coding. Again, I think this is just pure hypothesis on their part. I think Google just wasn’t able to make the deadline for the developer conference.

Justin: Yeah, a couple of stressed Google engineers.

Is Omni the Nanobanana for AI video?

Justin: So what did they release then during their I/O conference?

Frank: Well, they released Gemini Omni, which is a, what do you call that? Multimodal.

Justin: Yep. Anything in, anything out.

Frank: Yeah, so you can give it video, you can give it sound, you can give it images, you can give it all of those things together and ask it to do something with them. But there’s a lot of language around Gemini Omni that makes it sound kind of world model-y-ish.

I actually read some articles that said that Sundar Pichai had referred to it as a world model, but I couldn’t find a direct quote from him saying that. But the language is very world model-y, so I think journalists have taken that and run with it. But the big thing that people are latching onto about Omni is that it’s incredible at video.

So this is like what Nano Banana was for images, Omni now is for video. You can… If you remember when they released Nano Banana, they had a picture of a guy at a blackboard with a huge mathematical equation, loads of text, and all of it looked perfect. Well, they’ve released a video with someone writing on a blackboard.

It can do great writing, it can do great physics, apparently, which is where the world model-y stuff starts to come into play. And it can edit videos. You can do iterative editing, same as you can with images in Nano Banana. Looks super impressive and very, very cool.

Justin: Interesting.

Are we in the foothills of the singularity?

Justin: The world model thing is interesting because what I read, right, so they had this video where they had a marble, and the marble was running down a track. I don’t know if you saw that one. And you could see it worked really well, and it sort of fell off things and bumped into things, and you could see things that fell, fell over as a result of being bumped into.

And what they were saying was that the model was able to reason over what would happen next, which is kind of cool. One of the interesting things that came out as well is Demis Hassabis, the head of DeepMind, he was saying, look at world models, and by implication AGI and ASI, he figured was, I think he said, two years away, right?

He said that they were making great strides in that particular direction. It’s funny because world models, for me, aren’t just physics. World models sort of go far beyond just physics. It’s got a what are you going to do next, or if I do this thing, what is the consequence of the thing that I did? More than just physically, but in all sorts of different dimensions. So it’ll be interesting to see how that pans out. What else did they release, or what else caught your eye?

Frank: Just to go back to Demis Hassabis for one second, because the quote that everyone is latching onto was at the end of the main keynote. Demis Hassabis said that he believed that we would look back on this moment and see that we were in the foothills of the singularity. And I did think that was really interesting, especially from Demis Hassabis, because he tends to be very grounded.

He does believe we’re close to AGI, but his predictions always seem to be fairly conservative compared to other people’s. So I just think the singularity as well is such a loaded word because there’s the sci-fi definition of singularity, which is what I think of when people say it, which is where it comes a point where you cannot tell the difference between AI and humans, and in fact, machines and humans merge. That’s what I think of when I hear the singularity, so I was like, “Wait, what’s he saying?” But apparently his definition is more the traditional kind of labs definition of AGI.

Justin: Yes. Okay. And I’d be interested to… Because for me, the singularity is when things start to change faster than you can keep up with them.

Frank: Yes. I think that’s more or less his. Well, because his vision of AGI, is it… I get their definitions mixed up, all these guys, Dario Amodei, Demis Hassabis, Sam Altman. But yeah, isn’t… Because self-recursiveness is a big goal of Demis Hassabis, isn’t it? So that would make sense.

Justin: Yes, absolutely. So anything else?

Is Gemini 3.5 Flash fast but meh?

Justin: Because okay, I can tell you, right, that the reaction from people to the new model has been a bit meh. People are going, “Yeah, it’s not quite as good as the frontier models. It’s a bit more expensive than the frontier models.” And so there’s not a lot of love out there for it at the moment.

Frank: I think the 3.5 Flash, which is actually released and is actually in Gemini and it’s in Antigravity and it’s actually available, is meant to be very fast, so I think that’s undeniable. It’s cheaper than the other frontier models, but it’s more expensive than the previous versions of the Flash models.

And Google seemed to think that their internal benchmarkings, et cetera, were like, “Oh yeah, this is on a par. It’s only just barely below the other frontier models,” whatever. But I think the experience of people using it is not that it’s on a par with other frontier models.

I did think Sundar Pichai did say something interesting, which was he felt it was really fast and really capable, and he felt that there would be, in any organisation, and I’m paraphrasing here like crazy, but his kind of view was that there would be, I think a percentage, now I think he used a percentage of 80%, but even if it wasn’t 80%, there’d be a percentage of things that you could move to Flash, to Gemini 3.5 Flash, and it would make your processes much faster and much cheaper, and he had a cost estimate on it.

It’s worth looking it up. But that’s an interesting point in that it might not be brilliant across the board or it might not be able to do everything, but if you’re at an enterprise level, are there places where it can be applied so that it’s much faster and much cheaper?

Justin: Yes. I think that’s a really good point, right? Because I think we are entering into a period where companies will need to have, I was going to say three, but I think now four places to go to, right? So you are going to have to have a relationship with Google and a relationship with OpenAI and a relationship with Anthropic.

But what we’re seeing at the moment is the cost of tokens and the demand for tokens is increasing. So the cost of tokens isn’t going down. Sam Altman says it’s gone down by 1,000% for specific tasks, but in reality, they’re not really, right? Or it doesn’t feel that way, and agents tend to burn a lot more tokens, so your token cost is going up.

And so you have to have a relationship with all three because the models are good at slightly different things, and then probably an open source, sort of an on-premise or in the cloud open source model that you can turn to as well for other workloads. And then there’s going to have to be a really smart way to figure out, well, which model should I send it to? So, what did he say? 80% of workloads that you can move over to Flash. Which 80%? How do you know? It’s kind of a hard problem to solve.

Frank: Yeah.

Justin: You know?

Frank: Yeah, yeah. No, absolutely. Absolutely.

Is Spark Google’s safer OpenClaw?

Frank: I think the other interesting thing that they announced was a thing called Spark, which is an agent, basically, but it runs on their cloud. And when it launches, they’re being really, really cautious about the release of this. They are releasing everything to beta testers first, then I think it will be available to Ultra users who are the ones paying 100 or 200 a month.

They’re being really, really cautious with it. At first, it will only do agentic things within your Google ecosphere. So, if you’re in Google Workspace, it’ll work with your email, calendar, Sheets, all the rest of it. But the interesting thing is that they plan to bring MCP connections to it, which would mean that it would be able to communicate with third-party services.

So, in theory, probably a little bit oversimplified, but in theory, it means that people could have their own much more safe and much more secure version of OpenClaw, which caused a big stir earlier in the year.

Justin: Yeah. And do you think… Well, that’s an interesting thing, right? Because the thing that’s been sort of dawning to me for the last week or two is AI is really, really hard. And so OpenAI have looked to launch agents in their own cloud. Anthropic have looked to launch agents in their own cloud, and now Google are doing exactly the same thing.

And Anthropic so far have done the best job of it. OpenAI’s one has been very meh. I haven’t really been impressed with it at all. So it’ll be interesting to see. We haven’t seen it yet, right? So it’ll be interesting to see what they do. It should be easy, and it should just connect to everything, and you should be able to set it up and let it just run and repeat and do things automatically. So if they can crack that, well, that’ll be great, but we have to see it yet.

Frank: Yeah, and I think it’s cool. But at the same time, I already have access to Google Workspace Studio, which is not agentic, but it’s basic automations. But I think I was chatting with you about this before where I don’t really have that many things that I need to automate.

It’s good for really, really boring stuff if you’re a bigger company, like doing the first pass at responding to or classifying and responding to certain support enquiries that come in by email, for example. If you’re inundated with support queries, then you can start to triage and go, “Well, this should go to this person. This the AI can answer,” et cetera. But I don’t have operations like that that I need to automate. So automations and agentic solutions within Google’s ecosphere, I’m not sure what I would do with them.

What work should agents actually do?

Justin: Yeah, and I think as agents become… Here’s the thing that I also think we have to think about quite a bit. So let’s say if you run a small business or a medium business, at the end of every year you have to do your accounts, and accounts are somewhat deterministic and then there’s sort of judgement and so on, right?

It’s definitely something that you can imagine you could create an agent to do, and lots of people are trying to create agents to do it. But I think what it does is it increases the demand for the professional rather than replaces the professional. That’s at least where we are today.

So if you have a small business, right, I do. Even me, even me, would say it’s foolhardy to just hand over your accounts to an AI and let the AI do your accounts, right? However, I can certainly see that an accountant could turn an AI on your books. It could do loads of the sort of drudgery and the sort of work.

It could do a reconciliation between… It could look at your receipts, right? And you could use AI to look at the picture of a receipt and extract the important information, and then you could use that or maybe some deterministic code to do a reconciliation between that and your bank account, right?

After all that’s done, you can imagine 90% of it’s gone straight through and 10% of it needs somebody. That somebody’s still the accountant, and then when it’s all done and it needs to be reviewed to make sure there’s no sort of howlers in it, that somebody’s still the accountant. It’s not you as a small business.

So, I guess the other thing out of that too then is if you have the business, you have to kind of look at it through the same lens as a, well, where’s the bit that I create value? Because I should be automating and offloading everything else and just focusing on the bit where I create value. That’s, again, kind of hard to do, but that’s I think the world we live in.

Can Google win by being good enough?

Justin: The other thing is do you think that Google have lost focus? Because they seem to do an awful lot. I worry that the companies that aren’t focused are just not doing as well as they should do.

Frank: That’s a really good question. I don’t know that they ever had focus. I mean, if we’re talking, let’s say, within AI specifically, I don’t know if they ever had focus. They’ve always been an incredible research lab for AI, and then they’ve never fully pulled ahead.

Since OpenAI launched ChatGPT and forced everyone to start launching and everything, Google have had some cool announcements, and they’ve had moments where it’s like, “Oh, look, they’ve caught up, and they’re going to take the lead again,” but they’ve never actually taken the lead again.

And yeah, I don’t know if it matters for Google. I think it matters hugely for OpenAI and for Anthropic because they are AI labs. That’s it. Google is so much more. Google is such a behemoth that it might make sense for Google to be doing all this stuff and figuring out what works for their customers and then pushing forward with their huge customer base with what works for their…

Justin: Let me frame this a different way for you. AWS came out with Siri, I don’t know, maybe 10 years ago, right? Which was the first sort of consumer AI. They had a 10-year head start on everybody. They should have got there, won the race, had the best AI before anybody else, and they didn’t, right?

Microsoft own all of the corporate data landscape, right? Everybody uses Outlook, Office, Excel, Word, all of those things. They should have won the battle for the enterprise market already, and they tried with Copilot, and they used other people’s models, and they tried to write the harness, and they just didn’t do a great job at it, right? And so they didn’t.

Google, similarly, should own all of the personal and small, medium-sized business market. They have Gmail and Sheets and all those sort of things, and YouTube, and they have all the data. And they tried to write the model and the harness, and the model just doesn’t seem to have performed as well.

Whereas you have Anthropic and OpenAI, and all they do is write models and harnesses. That’s literally all they sell, and they’re streets ahead of everybody else. And I wonder, is it focus, right? It turns out that it’s really hard to write a really good model and a really good harness to work together, and the only companies who’ve really done it well so far, on the evidence we’ve seen, is OpenAI and Anthropic.

And the rest of the companies have tried, but they’ve failed because they have all these other competing priorities and considerations and things like that, except on focusing on doing that one thing really, really well, and it turns out it’s harder than we ever thought.

Frank: Yeah, but if you’re in Google’s ecosphere and Google’s AI is good enough, does it matter? Do they actually have to focus and be the best, or do they just have to deliver good enough to their clients?

Justin: I think that’s actually a brilliant question, and is the nub of what’s going to happen over the next two years. I don’t know the answer to that question. Do you know the answer to the…

Frank: No. No, I…

Justin: Because you don’t know until you see the other thing, right? And I think this is kind of to my point about focus. I’m sure those sort of conversations are going on inside Google, where it’s like, “We need to do this,” or, “We need to do that.” People are like, “No, no, it’s good enough the way it is. It automates 80% of the workflows. Let’s just… It’s good enough, let’s just ship it.”

And that will work for a certain period of time, but if the other models go ahead, pull ahead, maybe it’s not good enough. But it might be. I could be wrong.

Did Anthropic prove AI can pay?

Frank: Yeah, and let’s face it, the company that has focused the most and has actually… They’re the ones who prompted OpenAI to focus originally, is Anthropic. And so you had a story about Anthropic this week that I had missed.

Justin: Excellent news. My favourite AI company, Anthropic, have turned a profit, or will turn a profit in Q2, and not just a little profit. They’re going to turn a profit of half a billion quid, $500 million. Now, this is excellent news. Excellent.

Frank: That is impressive. And I didn’t get a chance to dig into this, but we all know, I don’t know the numbers to compare, but we know that OpenAI are not going to turn a profit.

Justin: I can give you those numbers too. Get this, right? And you will remember my little side quest bet that Sam Altman may or may not be the CEO of OpenAI by the end of the year. We’ll keep an eye on that. OpenAI, it was reported this week in The Information, for every $1 they take in, guess how much money they lose?

Frank: I couldn’t tell you.

Justin: $1.14. They make a profit margin of -114%. That’s how much money they lose.

Frank: I’ll be honest, that’s actually not as bad as I thought it would be. Obviously it’s not great, but I thought it would be worse.

Justin: Well, the numbers are enormous though, right? I don’t have off the top of my head, right? Because if you think, Anthropic’s run rate is going to outstrip OpenAI’s run rate, right, the amount of money they bring in, and it’s billions.

So when we say that for every dollar they take in, they lose $1.14, we’re not talking a couple of hundred grand here. They’re taking in maybe a billion, and they’re losing 1.14 billion instead, right? So it means they’re spending 2.14 billion for every customer that they’re trying to serve, right? It’s enormous.

Anyway, why is it good news that Anthropic are making a profit? Couple of reasons. First off, right, it shows that at the current price point, models are actually profitable. So the worry is that there’s a bubble in the AI market, and that bubble is going to burst.

What this shows me, the fact that they’re able to turn a profit and models are still improving and they’re being developed and so on, shows me that it’s not necessarily a bubble. Remember, they rent all their compute. They did a huge deal with xAI, right? And they can terminate that deal with 90 days’ notice.

So let’s say businesses decide, “We don’t need so many of these tokens. We’re going to cut back on our spend.” Well, so long as they’re still making… So long as the unit metrics work, which they appear to do because they made a profit, they can just go, “Cool, let’s not buy all that compute from Elon anymore. We’re just going to use less compute, and we’ll still make a profit.”

Frank: I wonder, along the lines of what you were saying there though, I wonder how much of their profitability is coming from about five companies in Silicon Valley burning through tokens.

Justin: Yeah, I don’t know. We don’t have those numbers, right? But I think that it’s a good signal for the AI market in general, right? Because I would be worried that there’s a bubble, but if a company can turn a profit, that means that it looks like it’s sustainable to me.

Did Anthropic just land Karpathy?

Justin: Other good news for my favourite AI company, Anthropic, this week, Andrej Karpathy has joined Anthropic. This is great news.

Frank: That’s big. That’s big news, isn’t it?

Justin: Yes, huge news. So Andrej Karpathy, for people who don’t know, was one of the co-founders of OpenAI. He was also the head of AI for Tesla. He went back to OpenAI, I believe, again, and then he’s taken a two-year hiatus just to do lectures and teach people and stuff. He’s brilliant.

Now he’s gone over to Anthropic and he’s going to head up a team in Anthropic that’s going to work on recursive self-improvement. Now, here’s an interesting thing in the AI companies, and this isn’t…

Frank: He’s going to try and beat Demis Hassabis to the singularity, is he?

Justin: Well, I don’t know. If you’re a techie and you’re a nerd into this, you can go onto GitHub and go onto Andrej Karpathy’s GitHub, and he has this thing called AutoResearcher, which…

Frank: Right. Yeah.

Justin: …thing which does automatic self-improvement at a sort of meta level, right, not down at the model level. And it’s really simple, right? You can literally run it on your machine, and it will just climb a hill of evals and get things better.

Now, what’s interesting, because he’s not the first to do this. You would imagine he’d go in there as the CEO or close to the CEO. He was one of the founders of OpenAI. No. Maybe he reports into the CEO. No. Maybe he reports into the guy who reports into the guy that reports to the CEO. No. He’s about four or five levels down, and he’s going to be stuck at the coalface working with other engineers on recursive self-improvement.

He is not the first to do that. People are saying that this is a signal. What he said himself was basically that this was one of the most impactful times to work in a lab for AI. And what people are saying is that signals that he thinks that this recursive self-improvement and AGI and the singularity and all those things that we talk about is imminent.

If he’s there and he’s saying, “I want to go in there now,” it means he doesn’t think there’s a year or two to hang around and then get involved in it. The time is today. So very exciting. Very exciting for him. Another interesting thing, just again, I’m bigging up my favourite AI company, which is Anthropic, as you may have noticed.

Is OpenAI selling tomorrow’s tokens?

Justin: OpenAI, who, may I just remind you, are making -114% on every customer that they sell anything to, this week also started selling forward contracts on compute. What does that mean?

Frank: Contracts on compute, right?

Justin: All right. So basically, I’m trying to think of a nice way to say this to you. Let’s say you were buying something from me, right? Whatever it happens to be. You’re buying widgets from me, right? And I turn around and say, “I’m going to sell you this widget for $100. And you can buy them as you need them, right? There’s your widget, $100.”

And let’s say I want to go on holidays or I want to do something, I just need the money now, right? I really could do with that cash now. What I would start to do is say, “Hey, Frank, I’ll tell you what, instead of selling you a widget for $100, if you commit to buy, let’s say, 100 widgets every year for the next two years, I’ll give them to you for $80. And actually, if you’d commit to buy them for three years, I’ll give them to you for $70. How about that?” Right? Sounds like a good deal.

Now, they’re marketing this as saying that there’s going to be an explosion in token requirements, in demand for tokens, and what this is doing is it’s allowing customers to guarantee their supply of tokens before that huge demand comes in, so that you know you’ve got access to your tokens.

No, it’s… No, it’s not. They need to get money in the door. That’s what’s going on.

Frank: Absolutely. I mean, I don’t know if I would be… The way you’ve described it, I don’t know if I would be buying forward… What did you call it? Forward…

Justin: I probably made something up, but yeah. Forward contracts.

Frank: I don’t think I would be buying that from OpenAI, who are losing 114% on every customer that they take on. I would be a little bit concerned that that guarantee would be meaningless if there was no OpenAI.

Justin: I know, and I think to your thing about, does most of the revenue come from five hyper users in Southern California somewhere? Obviously, if you’re a startup and you’re using an awful lot of tokens, maybe it is attractive, right? And you can say, “Hey, look, we’ve just reduced our costs.” I don’t know.

But it says to me that maybe there’s… Sorry, let me put it this way. I do think that there’s a big push among each of the companies to bring money in the door, and the fact that Anthropic can turn a profit now puts them in a very strong position relative to the other AI labs. That, I think, is true.

Frank: Yeah, I think that is just undeniable. I do think, as I said, there’s a difference between OpenAI and Anthropic and then Google, just because Google are just a whole other beast. But yeah, I think that’s really, really interesting in terms of the OpenAI and Anthropic relationship.

Justin: Yep.

Why did Elon lose his court case?

Justin: Before we move on, we like to keep an eye on what’s Elon up to.

Frank: Oh yeah, what is he up to?

Justin: So Elon’s losing, unfortunately, this week. So the court case between Elon and OpenAI, if you remember, Elon was suing OpenAI for a bazillion quid, who knows how much it was, because they tried to turn the charity into a for-profit. Thrown out of court. Elon lost on a technicality. I’m gutted for him, actually, if I’m honest. I was dying…

Frank: What was the technicality?

Justin: The technicality was statute of limitations. So it turns out he was claiming that this was a fraud, and it turns out that if you become aware of a fraud and you want to sue somebody, you have to do it within, I think it was, three years.

And so he was making the argument in court that he only found out about it recently. The court documents and he was asked about it on the stand indicate that actually he knew that there was a plan to turn it into a for-profit back in 2019, which is more than three years ago.

So the jury sat down, and before they actually looked at the merits of the case, the first thing they had to decide was whether the case should have happened at all. And they basically decided that Elon didn’t tell the truth, and that actually he knew about it all along.

Frank: Wow. You would think that that would come to light before the court case happened. I’m glad it didn’t, because we got loads of interesting gossipy nuggets from their communications from discovery and all that, so I’m kind of glad that the court case happened. But wow, you would think that the statute of limitations would’ve been figured out before it ever got to court.

Justin: And I’m… You know, he’s going to appeal it, of course, but I’m gutted because I really want them to discuss all the other stuff that happened, because there was a lot of murky stuff going on in there, and I would’ve liked somebody to have an eye over it. So anyway, there it is. That’s what Elon’s up to.

Frank: Mad, mad.

Did AI haters accidentally roast Monet?

Frank: Well, one other thing that caught my eye this week, and I just thought this was hilarious. Now, I’m a huge fan of AI image generators.

Justin: How come I thought you hated them? You want them all to be labelled as AI.

Frank: Well, yeah. I do. But I love playing with them.

Justin: So…

Frank: I love creating with them. But…

Justin: You’re just not allowed to share them. Okay.

Frank: No, I’m not going to create something that is purporting to be something real and put it out into the world as something real without labelling it as AI-generated. Except sometimes when I generate images of you doing things that you shouldn’t be doing and put them online.

Justin: What? Okay. Anyway, so…

Frank: So this Twitter account tweeted and said, “I just generated an image in the style of a Monet painting using AI. Please describe in as much detail as possible what makes this inferior to a real Monet painting.” Now, I don’t know if you know, Justin, but there’s a lot of hatred out there for AI, especially, as we’ve discussed on the show, in the creative fields.

So a lot of AI haters jumped into the comments and one person said, “The fact that…” I just realised we don’t have a policy on cursing on this show, so I’ll do the polite version. The person said, “The fact that it looks like SH1T and is SH1T slop. Doesn’t look anywhere like a Monet. Looks exactly like somebody trying to replicate the style and achieving 20% of it. Not as vibrant as Monet’s typical choice of colours. Looks dull.”

And it went on from there. People ripped it apart. The only problem was it was actually a photograph of a Monet painting.

Justin: It wasn’t money.

Frank: It was not AI-generated.

Justin: Oh my…

Frank: Now, there were some people, of course, who twigged it. There was an art historian who said, “Hang on a second, that is a Monet.” But there was a lot of AI haters who were just like, “That’s awful. The colours aren’t right. The composition isn’t right. That doesn’t look like water. Monet would never have painted that.” And they were all horribly and terribly wrong. A reverse image search would’ve told you it was a Monet. But…

Justin: Which is powered by AI, ironically.

Frank: But it raises… I’m sorry, I do find it hilarious. But it raises lots of interesting questions as well. And I think the big one is that it says more about perception than anything else.

Justin: It’s more about haters be hating.

Frank: It does say…

Justin: People being A-H-O-A-D-S’s.

Frank: And I saw some people saying, “Oh, well, this is invalidated by the fact that he deceived people in the tweet.” But there have been similar scientific studies showing that a lot of the time people love AI-generated output until you label it as AI, then they absolutely hate it. So…

Justin: Listen to me. It’s Friday evening, actually. I’ll tell you an interesting story related to this, right? They did a study years ago where they got people to taste wine, and some of the wine was expensive and some of the wine was cheap, and they told people… They give them the expensive one and say that’s cheap, and they give them the cheap one and say that’s expensive, and they’d monitor their brains.

And just by telling people it was expensive, they would enjoy the wine better, right? Not just in themselves, but actually when you monitor their brain, their brain lights up in the way that they would enjoy it better. The sooner that people get replaced by AI, the better. People by and large, people by and large are idiots.

Frank: Well, on that note, Justin, great chat as always, and I’m sure next week we’ll have loads more to argue about.

Justin: Comment about people being replaced by AI does not apply to the listeners of The AI Argument. Little…

Frank: Chat to you next week.

Justin: Talk to you.