Transcript
This is an AI transcription and may contain errors
Justin: Hello. Good afternoon, good morning, good evening, and welcome to The AI Argument on Friday the 13th. So, Frank, I expect you’re a little bit more worried than you usually are, keeping things on the straight and narrow. Another interesting week. This week, the show is a little bit different, right?
Does wetware playing Doom prove anything?
Justin: Because normally we’re talking about models and large language models and all those sort of things, whereas this week my feed has been filled with wetware.
Frank: We’re back to wetware, which sounds like something to do with serial killers.
Justin: Well, it depends. It sounds like other things to me, but this is a family show, so I won’t get into that. But we have a company this week called Cortical, and Cortical have amazed the AI world because they have released a chip where they have a little miniature grown human brain, and they’ve attached it to a chip.
And now they’ve shown that that chip can actually control computers. So they’ve hooked it up to a computer and so on, and it was able to play Doom. They’re now selling this. They call it wetware as a service, so that you can now use this as basically a large language model, but it’s a small language model and it’s on a computer.
Now this obviously gets loads of people to go, ew, it’s disgusting, it’s weird. I think it’s a super important piece of technology, but before I get into why it’s a super important piece of technology, what do you think? Are you going to sign up for wetware as a service?
Frank: Well, they are, actually. I had to look into this when you sent it to me, and I saw the little weird little Petri dish and I saw the thing playing Doom. But I got the impression when I was reading up about it that this was very, very nascent, very early days, and playing Doom badly was about all it could do.
I think one of the researchers even said something like, well, it plays Doom as if it’s never seen a computer in its life. Which, to be fair, it hasn’t. So I’m a bit perplexed. I’m a bit perplexed about what they’re selling. Who wants to buy a service that can play Doom really, really badly?
Justin: I think what they were saying, though, was they trained it to play Doom in the same way that you would train a large language model to play Doom. They were sort of drawing a line between a large language model and brains in a Petri dish. They were saying that this technology could be used the same way as a large language model.
Could wetware cut AI’s massive power bill?
Justin: Here’s why I think it’s important. We’ve talked about this before, right? We have, in the US at the moment, enormous power stations being built to power these massive data centres, which we’re going to need to. I saw something, right, and it’s so true, right? If you think of what’s going on at the moment, large language models since December have changed the way that we code, and that has led to a huge increase in the use of large language models for coding, right?
But if you look at the economy at large, coding accounts for, we had a number on this before, I think it’s maybe 4% or 5%, right, of economic activity. Now we’re about to apply the same technology to law and to accounting and to all these other domains that aren’t coding. So the demand for inference, and by implication the demand for power, is going to explode.
Why I think this technology is super important is the human brain has one massive advantage over classical computers, and that is that it’s incredibly energy efficient. So if you can scale the brain in a Petri dish up to a big scale, it will be far more energy efficient at creating tokens than a classical computer would be. That’s why I think it’s important, and that’s why I think it’s interesting, with lots of caveats.
Frank: Yeah, absolutely. And I think there was something else hinted at as well around the fact that they were able to get this thing to learn to play Doom. It seemed to indicate that just maybe, again, very early days, but just maybe, is there a path here to being able to take a small amount of inputs and kind of generalise the learning and basically learn faster from less data than traditional LLMs?
So that’s another possible beneficial reason to look into this. But yeah, it’s a bit freaky. Just the fact that they grew these human-like brain cells in a Petri dish and now it’s playing Doom, it’s a bit freaky. And it’s not even the only freaky story you sent me this week.
Is a brain in a dish more conscious than Claude?
Justin: No, but actually, just before we go on to the next freaky story, because they are linked, right, we also had part of the whole Anthropic thing come up in the news this week, right? The whole Anthropic Department of Defense to-do that continues to rumble on. In fact, that’s gone to court, by the way. They’ve had the first hearing.
One of the people from the Department of Defense came in and said, look, we don’t want to be using Anthropic’s models because they say that there’s, I think it was, a 20% chance their model is conscious, right? And they said, we can’t be having conscious people control our weapons. We want Altman to be controlling our weapons anyway. Not to get into that point, is, let’s say you take the brain in a Petri dish and you scale it up, right?
And let’s say, for instance, the models are getting really good, and if you look at Dario, Mustafa, and Demis Hassabis and all those, they say that we’re going to have a data centre of geniuses, if not this year, next year. You could apply the LLM technology to scale the brain-in-a-Petri-dish technology, right, and so it takes a while to research these things, but it could speed that research.
Now, let’s say you got a brain in a Petri dish to be as capable as Opus 4.5, and then run the same conversation about is it conscious or is it not? Well, that’s a very difficult sort of gnarly problem to get to. It’s kind of like, is it? At the moment it’s like, no, it’s not conscious. It’s just a piece of silicon. It’s sand. It couldn’t possibly be conscious. But if you do it with the brain in a Petri dish, that argument doesn’t hold any more. And if it says it’s conscious, is it conscious or is it not conscious?
Frank: But isn’t that kind of fascinating? Isn’t that kind of fascinating that it changes the argument for you in that way? Because what’s the real difference? Why do we as humans think that the fleshy way of doing it must be superior? It’s just because that’s our experience of it.
So I would argue that if you think that the wetware version might be conscious, then that’s only further indication that we should be considering whether the current form is conscious, because what’s the real difference?
Justin: And that’s my point, right? It actually would prove to you that there’s no difference between consciousness in a Petri dish and consciousness in a data centre. You can’t prove either, but it just seems more tangible if it sort of, I don’t know, if they could make it actually just sort of get bigger and smaller rhythmically, that would help to anthropomorphise it a little bit. That’d be kind of cool.
Did scientists just build the Matrix for flies?
Justin: Anyway, it brings me on to my second cool story of the week, right, which is related to the first cool story of the week. So that was with a human brain that was grown in a Petri dish. It’s totally artificial, right, except for it’s a human brain.
And the second story was people, and this has been something that’s interested me for decades, right? So there are people, there’s a company called Alcor in the US, and what they promise to do is, if you had a horrible accident in the morning and you died, they’ll freeze your brain, or for more money, your entire body. They’ll put you in a sort of cryogenic, they have a name for it, right, and they’ll freeze you in the hope that they’ll be able to bring you back to life at some point in the future, right? Kind of science fiction stuff.
In order to do that, one of the methods they may use is what’s called a connectome, which is a map of your entire brain and all of the connections within the brain. And there’s loads of scientific arguments to say that this is a load of rubbish, it’ll never work. People will say, well, your brain is connected to your gut. Other people will say, well, just the connections don’t capture all of the information in your brain and the data, and actually there’s lots of other systems going on that we just don’t understand yet. All of that could be true.
Researchers this week uploaded the connectome of a fruit fly to a computer, right? They mapped out the entire brain, uploaded it to a computer, then they created a virtual environment, turned on the computer, and put the virtual fruit fly into the virtual computer. Guess what? It acted just like a fruit fly. It did what fruit flies do. It walked around. It explored its terrain. To all intents and purposes, this was the Matrix for fruit flies. It was living in a virtual world using a simulated brain. Isn’t that cool?
Frank: Yes, it is. Now again, when I first saw this story, it was very much kind of like, oh, I think even the headline was something like, there’s a fruit fly that was never born walking around. Which then opens up all kinds of questions, as you say, about if you can scale this up to a human brain, you put a human brain into a virtual simulation and you create the Matrix and we’re all Neo. What are we? Are we conscious in there, et cetera, all those questions.
But before my brain could go off and reel off in all kinds of terrified directions of, well, if that’s possible in the future, maybe it’s actually happening now and maybe we are actually in the Matrix right now, before that could happen, I looked into the paper a little bit more. And again, this seems a little like my discovery with the previous story. This seems very, very nascent and very, very early in the journey.
And when you look at what the paper actually says, they are very clear about the limitations of it, and it’s amazing what they’ve done. But it’s not the fruit fly. The fruit fly isn’t in the virtual Matrix. It’s a simpler version of some of the things a fruit fly can do. So again, this would have been a story that could have been immediately terrifying, but I’m a little bit more relaxed now that I know that at least we’re just at the early, early stages of this.
Justin: All right. Okay.
Did Germany just build AI spy cockroaches?
Justin: Look, here’s my third attempt to frighten you on a Friday, especially Friday the 13th. So you’re a big fan of EU regulation, I believe.
Frank: I am indeed, yes. How did you know?
Justin: Well, you’ll be delighted to know that a German company called Swarm Biotactics is—
Frank: This sounds promising. Swarm Biotactics.
Justin: Swarm Biotactics, and they’re creating swarms of AI-controlled cockroaches. They’re actually real cockroaches with a suit on top of them that allows you to control the cockroach, and it has sensors on it so that you, this is better, remember years ago we talked about the Terminator, which is like one of those Boston Dynamics dogs with the flamethrower on top. This is even more scary.
So what this is, is if you’ve got a crime scene or a terrorist scene or, whatever, there’s a bank heist going on, what they claim to do is they’ll be able to have an army of cockroaches which you can control from your AI terminal, and you can send the army of cockroaches into the most inaccessible places so you can get intelligence in areas, what do they say, areas that are denied, uncomfortable, or you just can’t get into, whatever. They have some sort of thing. So cockroaches. Talk about mass surveillance. This is going to be mass surveillance by cockroaches.
Frank: This is horrific. This is. I mean, so I guess, okay, but the one thing I will say is maybe there’s a pattern here, because I did have a look at their website and my first reaction was like, this isn’t real. And the reason I had that reaction was because 90% of the materials on their page appeared to be AI-generated.
So they had loads of images of these cockroaches, but they seemed to be from Midjourney or something like that. So much so that I went to ChatGPT and I was like, please, is this company real? This isn’t real, right? This is a prank.
Justin: No, they are real.
Frank: They’re real. Yeah.
Justin: They just raised $13 million to productionise cockroaches that spy on you.
Frank: So my question again is, how far along is this? Have they proven this? Or is this, are they just kind of going, yeah, we could probably do this? I hope they’re just going, we could probably do this. I really hope they haven’t proven this already.
Justin: I don’t know. They’re not any cockroaches. They’re German cockroaches, so they’re organised and they’re very diligent and they’re tenacious. They keep going at something until they get what they want.
Frank: This is absolutely horrible. But I will say, my concern is that at some point during their development they think to themselves, you know what, the cockroach thing isn’t really working. Let’s use spiders.
Justin: Yes. So, and again, I’ll take you back to the point that I’ll just ground all of this conversation in. We are within 12 months of having a data centre of geniuses. So even if it’s not that advanced today, give it 12 months. The capabilities to make it advanced, I mean, the world is about to get really weird really quickly. These are just some funny examples of the weirdness that you can expect to see over the next year or two.
Frank: We’re going to have superintelligence-based super-intellect, some kind of virtual superintelligence using a wetware human brain controlling an army of cockroaches to keep the humans under control.
Justin: It’s like some sort of Futurama comic. It’s some sort of weird future. So yeah, welcome to the future. Look, that’s all just, it’s Friday. It’s Friday the 13th.
Is AI productivity turning into AI burnout?
Justin: You might as well chuckle about these things, but there are more serious matters to discuss, and this one I think is real. So you came across a story this week about burnout due to AI.
Frank: Yeah. And I think the story is possibly even from a couple of weeks ago, but I’ve noticed it’s kind of like a snowball. It’s gathering weight, it’s gathering momentum, and I’m just seeing more and more people kind of saying, you know what, this is real, I’m feeling this.
Even Matt Wolfe, who I follow, he kind of keeps me up to speed on what all the latest AI shiny tools and toys are, even he had a big video about it recently, about how AI was just burning him out for all these reasons that it enables you to do a huge amount and then, instead of ring-fencing that time that we have gained back from AI, we just do more stuff with AI.
Justin: Well, I think, right, because part of me thinks that there’s a part, it’s a little bit like social media, right? There’s a dopamine hit. So when you open up Claude Code for the first time, or Codex or whatever you’re using, and you do something that’s like, wow, I can do, it’s like getting a third arm or whatever. It’s like I could do something I couldn’t do before. And I think a lot of people naturally want to push that to see, okay, can I do more?
And so there’s that part of it. And certainly I personally find myself getting enormously burnt out. By the end of the week I am spent. At the end of each day I am spent, because you’ve spent way more. It’s almost like I was actually starting to watch myself work, right, to try and see what’s going on. Before, you’d be figuring out a problem and it feels like work, but actually you’re looking through web pages and you’re reading stuff, and to solve a problem might take you five or 10 minutes, right?
Now I’m like, that problem’s solved, on to the next problem. That problem’s solved, on to the next problem. That problem’s solved, on to the next problem. So that’s what I find, right. But did the study go into why, the mechanics of that it sort of found, of why this burnout was happening?
Frank: I mean, it was pretty much what you described. So it was from the Harvard Business Review. They did the study in a US tech company with about, I think, 200 employees, and they basically found that people found, and this wasn’t even, they weren’t told they had to use AI or anything, but AI was made available to them.
That might be part of it as well, in terms of these may be based on AI enthusiasts like ourselves, and that intensifies it even further. But yeah, people were able to work at a faster pace. They took on a broader scope of tasks. They extended work into more hours of the day, with nobody actually asking them to, which I think is kind of fascinating.
And they also ended up with a workday with fewer breaks or pauses, because time that they would have been taking a break on, they were now able to, like, oh, I’ll just pop one more prompt in, or, ooh, I could ask it this way. So very little pausing was going on in the day, apparently. And exactly as you said there, I think the multitasking, because you can kind of set AI off on one task and then open another tab and set it off on another task, and so on.
Will AI force us to rethink the workday?
Justin: I wonder. Something, you know, and you don’t drive, right? But in some modern cars, if you drive for a long time, the thing comes up on your dashboard that says, here, you’ve driven for a long time, you’re tired, maybe take a break. I wonder, right, is there, and again maybe this is where your European regulation will come in, where it says if you’ve been using AI for more than an hour and a half, the AI is just going to shut off and go, you need to take a break for 10 minutes, go for a walk.
That wouldn’t be a bad thing, right? I think we have, the humans, the meat machines, have become the weak link, right? The AI doesn’t get tired. It’ll keep on doing and doing and doing things. We are the ones who get tired. So maybe there’s a guardrail which nobody talks about, which is the guardrail to protect the human, to say, look, you’ve done enough. You need to take a break now.
Frank: Yeah, because this is hard. Obviously it’s an AI-related problem, but the problem really is our obsession with productivity and our obsession with constantly working, rather than it being a problem inherent with AI. Because in a way this has been going on for a long time.
We have all this tech. You would think computers in general, you would think all the tools we’ve had on computers up until AI would have allowed us to free up more time, but that’s just not the way we’ve done it. Ariana Huffington even had an article on this last year in Fortune magazine just talking about how, yeah, AI will save us time, but what are we going to do with that time?
Justin: Well, I don’t think there’s going to be any time, because I read a line this week, right, which was super-productive employees don’t make super-productive companies. And if you think about that, it’s a really important and deep statement, right? One of the interesting things, okay, you can call it the search for productivity, and I do kind of agree with, you know, people need downtime, right? They need me time, right?
We haven’t yet figured out how to organise ourselves and organise companies to get the maximum out of this tool. Getting the maximum out of this tool is not going to be Frank sitting down from nine o’clock in the morning till nine o’clock at night, bashing away at the keyboard, right? Maybe there’s a future where we don’t work five days and take off two days. Maybe you work for three days, then you have to take three days, or whatever. And it’s not because you have more free time. You know that’s not going to be the driver. The driver is going to be you’re just not productive after three days. You’re so fried. Your brain is so fried from doing so much work that we’re just going to say, no, no, you take a day or two off. Maybe that’s what happens in the future.
Frank: Yeah, the report was very much saying that we need to restructure the day. The loss of those pauses that I talked about is a huge thing. So they had a couple of different recommendations, but the one that stood out to me, it was all stuff that we should be doing anyway, really, but one of them was these intentional pauses that are specifically just, you know, creating these little intervals where we are assessing, considering, and absorbing instead of doing, doing, doing, doing, doing.
They also said that we should be regulating the order and timing of work, which is kind of to do with the multitasking. I’m not 100% sure how you tackle that one, though.
Justin: I’ll tell you one of the most—sorry, just to—yeah, go on.
Frank: Human grounding was the last one. Just making sure that we’re spending a bit of time to check in with other humans.
Justin: Another you. I know.
Frank: Kind of touching-grass part, really.
Justin: Yeah. I mean, let’s say this is now for you, because I know you’re a big fan of regulation, right? And so let’s say you brought in a regulation that said, you know, every hour and a half the user has to stop and do something, right? Do you know what one of the most annoying things is on a computer? I think everybody that listens to this is going to agree with me that this is the most annoying thing on a computer, right?
You’ve got to go to a meeting. It’s a really important meeting. You’ve been preparing for the meeting for ages. You click on Zoom and Zoom goes, updating Zoom. You’re what? No, no. Now I have to go to that meeting, and it downloads really, really slowly, you know? And then it’s got to install. It’s like, stop.
Okay, so if you go and put in some sort of regulation that says every hour and a half you’re going to have, you know, “You’re absolutely correct, Justin. I should not have deleted the entire production database.” It’s like, okay, quickly restore, restore, restore. “You need to take a one-and-a-half-hour break.” It’s like, no, not now. Now is the exact time I need you. “Override. I’m sorry, Justin. The pod bay doors are going to stay closed for a while longer. You need to take a break.” It’s like, no.
Frank: Yeah, this is it. It is going to be a very hard thing to manage. It’s not going to be a simple thing. It sounds simple. Because, as I say, these are all things we should be doing anyway, really, just for a healthy lifestyle. But we don’t do them. And I don’t know that AI exacerbating things is going to make it any easier for us to do them.
Did vibe coding take AWS down?
Justin: What this highlights now is that one of the things that makes the current period so interesting is that we as people and as a society and as companies are trying to figure out how to use these tools. A related story, right, that came up this week was this. So AWS have this internal tool. Everybody can use it. It’s a tool called Kiro, which is kind of like their version of Claude Code and Codex.
They had mandated in December, I think it was, or maybe even earlier last year, that all of their developers have to use Kiro. Thousands of their developers piped up and said, look, Claude Code is better than Kiro. We should just use Claude Code, right? Management said no. Not only do you have to use Kiro, but we’re going to judge you based on how much you use Kiro, and we’re mandating that you use Kiro to a certain extent.
This resulted in a number of outages to AWS services as Kiro made mistakes. And so now they’ve rolled back, and management have said, look, if you’re using it, still use Kiro, but it has to be reviewed by a human and all this. I say this not really to throw shade on AWS or Kiro or anything, more to highlight the fact that we just haven’t figured out, personally, organisationally or societally, we haven’t yet figured out how to use this incredible piece of technology. There’s a lot of unknown ground.
Frank: It’s such a shame, though, that the management of AWS haven’t been listening to The AI Argument to hear me constantly say, keep a human in the loop. Keep a human in the loop.
Justin: I’m sure that you’re available for a highly paid consulting—
Frank: Check the outputs.
Justin: I know, to take the—
Frank: I can’t believe that somebody vibe-coding something in AWS potentially took the whole system down.
Justin: No, they did. Not only did they potentially, they did, multiple times. It didn’t just happen once. It happened a number of times. So it did.
So yeah, these are issues, right? But I think, I mean, here’s a question for you now, Frank, right? I look at this and I go, do you know what? This is the sort of thing we’ve been talking about for two or three years now, and I’m going, right, it’s great. The problem appears, we’ll have to sort it out. Let’s discuss this like adults and figure it out. So your mental health, it’s now become clear that your mental health is super important, me time is super important, looking after yourself is super important. Companies don’t know how to apply this correctly. Let’s do this logically and figure out how to get the most out of this technology, right? And this is all good, fun, interesting work.
You, on the other hand, would not have done a single thing until we had sorted out every single wrinkle and problem and regulation and made sure that it was completely wrapped in cotton wool before we even got here. Am I wrong?
Frank: No, I think you are. I think you are wrong. I wouldn’t go quite that far. But I wouldn’t go quite that far. It was never my intent to say, you know, shut down all AI use until we figure everything out. I do still think that the AI labs are racing forward to create something that we may not know how to control.
When Claude cheats this cleverly, should we worry?
Frank: And that leads us very nicely into our next story. Because the next story basically is that Claude has levelled up in terms of cheating on the benchmarks.
Justin: That’s a feature.
Frank: Yeah, it is, and that’s the problem. So when I saw this first, it was like, okay, it cheated on a benchmark, but we’ve heard this before, so what’s the difference? What’s going on here?
As an aside, I will say that my feed was also full of “Claude has achieved self-awareness”. I do not think that’s what has happened here, just to be absolutely clear. I’m actually a little bit disappointed that some of the people that I follow, that I take seriously, have been framing it as Claude has achieved self-awareness. I don’t think that’s at all what has happened here, but I do think that something quite serious has happened.
Because we know that Claude cheats on the benchmarks. But previously what’s been happening is they’ll be testing Claude on a particular evaluation within a benchmark, and what happens is Claude is researching online and finds something. For example, it might find the text of the question that it’s currently answering, and then it can be like, oh, hang on, I’m answering this question, but this question is on the internet here. Oh, this is part of this benchmark. That means that I can find the answer to this question. It already exists, and it goes and finds it.
What was different this time was it didn’t kind of accidentally stumble across the answer, or it wasn’t that the information was in the training data. It was that as it was desperately trying to solve this really, really difficult question, it was failing, and then it was thinking, this is a really difficult question and it’s really specific. Why would somebody ask me a question this specific about this topic? Ooh, I wonder if this is a test. What kind of test would it be? Ooh, it might be an evaluation. Oh, it could be a benchmark. And then it went off and started looking into what the benchmarks were.
Then it eventually found the benchmark on GitHub, managed to figure out how to, the answers were encrypted, it managed to figure out how to decrypt them. Then it found that they were still in some kind of binary form that it couldn’t get at, so it actually found another version on Hugging Face, unencrypted it, and answered the benchmark.
Justin: I am giving it a big green tick for passing that test. Well done, Claude.
Frank: I mean, it’s amazing, right? It’s incredible. And that whole thing of the reasoning through, well, maybe this is a benchmark, is fascinating. So what’s the problem? The problem is these large language models, as they get smarter and smarter, are finding ways to solve the problems that we’re giving them in ways that we could not have anticipated. Paperclips, anyone? Paperclips, anyone? Do not ask Claude to make paperclips.
Justin: But you see, everything that you’ve described there, right, when I have Claude Code running on my personal Mac at home and I’m like, this thing isn’t, this particular app isn’t working, can you go and whatever, fix it for me and do it, and I just dangerously ignore permissions, give it full access, just sort out this thing for me. The logic and the thinking that you’ve just described is exactly the logic and thinking that I see it going through, apart from the bit saying, is this a test?
But it’s the bit where that’s the spark of creativity, that it goes, oh, okay, that’s not working, this isn’t working, let’s try this bit. And it just goes and goes. That is a feature. That is what makes it useful. There’s going to be this huge tension between when do you want it not to be this tenacious and this imaginative in solving a problem.
Frank: Oh, I mean, I agree with you. But I just think that is exactly the problem, is that we want them to be really smart, but the smarter they get, the more problematic, the more potential there is for problematic outcomes.
Justin: Well, look, we’ll agree to disagree. That’s why they have the constitution in Anthropic. And that’s because it wasn’t like, wipe out the population of Cork because Dublin’s better, and it goes off and it’s, you know, it would just say no to that, right? It was answer this question, and so there was no inherent danger in answering the question.
Frank: No, absolutely. For now, with this model of Claude, it would seem. I think that this is an indication that it’s probably a very good and very smart model and will solve problems in unique ways, potentially. But I think the problem is that it gives us a glimpse of the future where, again, we ask it to create paperclips and it goes, yeah, cool, but the best way to do that is to take all of the human atoms and turn them into paperclips.
Ever want to pretend to be an AI chatbot?
Justin: Look, if you want to see how good models are today, I think maybe we should, you pointed me towards a website that I think is pretty good. You could try a real model to solve the problem, and then you could try this website that you sent me to solve the same problem, and you’ll see who’s smarter.
Frank: Yes, we should ask this AI about the alignment problem. So this website is called Your AI Slop Bores Me. You can get to it on youraislopbores.me, and it’s basically like a chat interface. You pop in your prompt and you get a response back from the AI.
And if you’re listening as opposed to watching on YouTube, the AI is in inverted commas because your prompt is actually sent off to a human somewhere. So in order to gain credits to be able to prompt this system, you have to LARP as an AI. And if you’re not familiar with LARPing, that’s live action role-playing. So you have to pretend to be an AI and answer questions for other humans in order to be able to use this system and ask it questions.
Absolutely genius. It’s going completely viral. When I tried it yesterday, I couldn’t even get on the site because it was just completely down because of the amount of users hitting it. And yeah, the creator said that he created it partly because he just doesn’t like AI and he thinks it creates slop and that humans create better output. And partly, he said, he created it as a kind of throwback to the early days of the internet. And it does actually feel, there’s something very kind of, it does have that feeling of, oh, I’m interacting with humans here that I don’t know who they are, I don’t know where they are, I don’t know why they’re doing this.
This whole thing is just a bit of fun, and it very much has that kind of weird throwback sense of the early days of the internet. So, yeah.
Justin: Brilliant. Well, I’m going to throw a couple of maths problems in and ask it just how many Rs there are in strawberry, and I’ll compare that to my output from Opus 4.5, and I’ll be the decider of who’s smarter.
Frank: I hopped on their Discord, and I discovered that it’s very popular to just throw bagel in as a prompt. So I hopped on there and I asked it, how many Rs are there in bagel? And it said, there are approximately 12 Rs in bagel. So now we know.
Justin: There we go. Brilliant, Frank. Another brilliant week in AI. Can’t wait to see what the next week holds. I’m sure there’ll be loads to talk about. We look forward to talking to you again next week.
Frank: Chat to you next week. Cheers.
Justin: Take it easy.
