Transcript
This is an AI transcription and may contain errors
Justin: Hello and welcome to The AI Argument with myself, Justin Collery. Join us, everybody, the ever-careful Frank Prendergast. Howdy Frank. Another cracking week in AI. Loads of releases of tools mainly, and commercial goings-on, not so much of models this week. Elon is, or not Elon, but Sam Altman is promising new, faster coding models soon. I’ll keep an eye on that. This has been the week of the battle of the titans. We’ve seen a lot of stuff that’s telling us about the future of AI. So we have had, I’m going to start with my favourite release this week, Claude Cowork. We’ll come back and talk about it in a bit, but basically they put Claude Code behind a web interface.
Justin: The clue is in the name Cowork. This is for people who work in companies. Then we had Google have released hyper-personalisation. They’re trying to get their knits into you so that they know all about you and they will win the AI race because they have all your data. So Anthropic are going for companies, Google are going for people. Google also signed a deal with Apple. We knew about this already, but it’s come up again where it’s going to power the new Siri. So it doesn’t matter what phone you use in the future, a Google model is going to be the model that’s behind it.
Will ads ruin ChatGPT before it makes money?
Justin: And OpenAI, if you shoot for the king, you better not miss. OpenAI have released a new model, Frank.
Frank: Not that I’ve heard of.
Justin: No. Did they release a new product?
Frank: Well, that depends on what way you look at this. I guess they are. Everyone’s going to be so excited about this. Everybody’s going to love it. I think this is going to be the thing that’s really going to bring people into ChatGPT. They’re going to show ads in ChatGPT.
Justin: And okay, so I think that’s interesting because if you look at OpenAI, right, people are worried about bubbles and what’s going to kill Sam Altman is not privacy. It’s going to be money that’s going to kill him, and the breadth of his ambitions to try and take over Google. So Google are fighting back with their hyper-personalisation and getting Apple on board with their models.
Justin: But OpenAI need to make money. They’ve made promises to bring in billions in ad revenue, and that’s starting this week with putting ads there. What do you think about the ads, Frank?
Frank: Well, I mean, everybody hates ads. Everybody hates marketing. In fairness, right now, it’s only on the free plan and also the kind of, they’ve had this low-cost plan that they’ve had in certain developing countries. They’re now going to bring that, I think, to certainly to the US. I’m not sure if it’ll roll out everywhere. So the free and the low-cost tier will have ads. If you’re on Plus or Teams or any of the other plans, you won’t see ads.
Frank: So I can understand that they, you know, we’ve heard time and time again from Sam Altman that most users are probably not actually paying for what they’re using. So they’re subsidising all of us massively by the sounds of it. This is a way for them to claw back some little bit of cash from the low-cost and the free tiers. So I can totally understand it, but I definitely have some concerns. But before we get to those, what do you think?
Justin: I think it’s phase two of the slopification of the internet. And I hope that what happens in the future is that you see a divergence of the models from the free ones, that people are maybe amazed by this, but the models that you have in those free tiers will be optimised to find ads and present you the ads.
Justin: Right now, I’m old enough to remember when Google had no ads and when you did a search, you got the thing that you wanted right at the top of the search. Then over time they introduced ads, and so now you couldn’t trust what was at the top of the search. And then what happened after that was that, you know, one of my bugbears on the internet is if you go to search for a recipe and you find the page and you click on the page and the recipe is actually down at the bottom of the page and it’s really hard to find. You have to scroll through the page, and this is in, we’re into your world here. The reason that exists is because Google monitors how long you spend on the page and how much of the page you’ve scrolled through. So they’re gamifying Google so that they appear higher in the results. And so I worry that what’s going to happen is that the models are going to optimise for displaying you ads. So instead of getting the right answer, you’re getting the answer that has been paid for the most.
Frank: Yeah. Now to be clear, they promise that they won’t do that.
Justin: And?
Frank: Yeah, so my concern is kind of similar to that, except that I would say that certainly right now I think we could see a drift. They promise they won’t do that. I don’t think they will do that right now, certainly, because there would be too much uproar. They’re under too much scrutiny. But who knows what the future holds, for sure.
Frank: But my concern would be the unintended consequences that could be very, very similar to exactly what you’re describing, except that they won’t even set out to do it. Now this is all kind of hypothesis and conjecture because we haven’t seen the interface yet. They’ve announced it’s coming, I don’t think anyone has actually experienced an ad yet. They’ve shown some screenshots of what this is likely to look like and talked about what you might be likely to be able to do.
Frank: So they’re saying it will not train, they won’t give your data to it or anything like that. But if the chat, if the AI sees that something would be relevant within the chat, then you’ll get a separate little ad underneath the response. Here’s the problem though. What if, you know, they’ve said that you could ask then about that product if you wanted.
Justin: Hmm.
Frank: So if you then ask about that product, and you know, all the models are going to this kind of personalisation thing and we’ll talk about Google’s hyper-personalisation in a minute, but if you have memory turned on in ChatGPT, if you ask about the product that’s now going into your memory, so in the future, you know, ChatGPT could end up thinking, oh, you know, Frank really likes Brand X because I asked about it, but I only asked about it because an ad triggered it.
Frank: So what I would like to see would be, at a minimum, when you get an ad that there would be a little button saying, “Ask about this product”. And if you clicked on that “Ask about this product”, it would open up a new chat where you could ask about that product and have a separate chat and it would be ring-fenced the same way your health conversations in Health are ring-fenced, so that you don’t get any accidental, unintended consequences from chatting about ads with ChatGPT.
Justin: I think that’s a really good idea. I like it. They should do that.
Justin: Now I will say that one of the really annoying things about Google is, you know, you need a pair of wellies. You go onto Google, you search for a pair of wellies, you get ads for wellies, you buy the wellies. But for the next six months, you’re getting ads about the wellies, even though you bought the wellies. You don’t want, I don’t need the wellies anymore. I bought the wellies. For the love of God, I don’t need any more wellies, got more wellies than I know what to do with.
Justin: Now, if what you say happens, at least in this setup, I can just say to ChatGPT, can you remove the memory about the wellies, please?
Frank: Yeah, I bought them. Leave me alone.
Justin: Right, and then you’ll stop seeing the ads for the wellies. So that might be a positive unintended consequence.
Frank: Yeah, that is true. That is true. We might get away from that feeling that, well, I was going to say we get away from the feeling that our phones are always listening to us, but in this case it would be they actually are listening to us, actually listening to what we consciously tell them.
Justin: Yeah, yeah. Now, the other thing I guess around this, the purist in me worries about this. Does the ad go into the context of the conversation? Because if it does, then I do have a problem with that from a point of view of I’m asking a question. Let’s say I’m asking a question about, I don’t know, I want to buy a new computer, and I’m asking a question about a computer. And so an ad for Dell pops up. Now, if that persists in the context of the conversation, then that really has steered the conversation and I get a less than optimal result. So let’s say I say, no, I really don’t want that. Actually, what I’m looking for is this, this and this. Well, the Dell thing is always going to be there.
Frank: My understanding is it’s not in the context of the conversation until or unless you actually chat to ChatGPT about the ad. My understanding is, yeah, it’s a separate system that kind of analyses the conversation, finds a suitable ad, pops the ad in, and then it is separate. But if I said then, “Oh, tell me more about this Dell computer that you’re showing me,” then obviously that brings it into the context and that’s kind of the unintended consequence that I’d be talking about.
Justin: All right. Okay, that sounds fine. Actually, I might, look, I don’t have a big problem with this.
Frank: If they said, yeah, you know what, this has been really good, this is working really well, we’re now going to put it in Plus and Teams accounts… I can’t see them doing it in Teams accounts, but Plus accounts I could actually see it. So if they brought it into Plus accounts, what would you reckon? I know I wouldn’t be…
Justin: I wouldn’t be too happy with that. That’d be another excuse for me to leave OpenAI.
Frank: Yeah.
Justin: Like, there’s no need for it. And Plus specifically, you’re paying money to get straight to the truth. You don’t need, you know, that seems like a kind of a social contract thing that you’re breaking. If I was able to pay, like you can do it within Google already with YouTube. So, and that’s actually exactly the parallel. In YouTube you can pay a tenner a month to not see ads. And so the same thing is in Google. I should be able to pay 20 quid a month to not see ads. Now, if you start putting ads in there, you’ve broken our contract.
Frank: Which in a way, I think what you’re pointing out would say to me that, yeah, the roadmap will definitely be this. This will come to Plus and there’ll be a Plus Plus tier. There’ll be a Plus ads-free tier as well.
Justin: Oh, okay. Well, in my head, I’m already paying for the ads-free tier. The 20 quid a month is the ad-free tier. I’d feel bad about it, but listen. Okay, so that’s that one, that’s the ads. The other thing then I would say, this is just another reason for me to leave OpenAI.
Is Claude Cowork what Copilot should have been?
Justin: So Anthropic released Cowork this week. Have you used it? Do we want… let me, will I explain what it is first?
Frank: Yes, please do. Absolutely.
Justin: Cowork is cool. Go use Cowork. Go get yourself a subscription. It’s going to cost you a hundred bucks or 200 bucks a month. It’s worth every penny. So what Cowork does is, in my other life, I’ve spent a lot of time, and even at home I spend a lot of time, where you gather information. So loads of documents and whatever that you’re working on, not code, other stuff, put it into a folder and you run Claude Code on top of it. And now you can do stuff with the documents. It can write presentations for you. It can write documents. If you have, you know, standards or procedures or whatever, it’ll read through them and understand them and you can work with them and then write more documents.
Justin: You can do, for our podcast here, I’m interested in pulling down the transcripts and then putting them all into a folder and I just run Claude Code on top of it and it will give me ideas for new stories and tell me about old stories and how they link. Selling why Frank is wrong and Justin is right. That sort of stuff. So what Cowork is, is it’s exactly that but with a web interface. So you open up basically a browser window, it’s in Claude.ai, you upload the files, it stores the files, and then you can go and do work in the files.
Justin: This is now, I think it’s still, I think it’s still one, it’s still the fact that I have to upload files is a point of friction for me. I think it needs to be integrated more closely into your machine. This is what Copilot should have been. Right. But it isn’t. It’s not.
Frank: But this is the… I mean, the reason that you’ve uploaded the files, presumably, is for security, right, and not to just give this agent full access to your entire machine, which would be useful but also dangerous.
Justin: Yeah, I mean, so when you run, so to be clear, when you run Claude Code on your machine, it only has access to the folder that it’s running in. And if it tries to do anything, if it tries to break out of that sandbox, it will ask you, are you okay with me doing this, and you can say no.
Justin: So I think it should be, and people organise, that’s what I’m saying. It should be in SharePoint, it should be on your file system. It should always be like, you know, you go to a folder where you’ve organised stuff, you right click on it and it’s like “start a session”, and then you can start doing work in that session. That feels to me like the interface, because the files are already there. Having to upload them is a point of friction. Skills are also super important. I’d be interested to see how they implement skills within this.
Frank: So when you say skills, you’re talking about the ability in Claude Code to write a file that kind of explains how to do something, and then you can call it into a project. So the way that I’ve heard it described for lay people like me who don’t code is, it’s a bit like in The Matrix where they say, “Hey, teach me kung fu,” and you just download it and you can do it.
Justin: That’s a pretty good explanation. Yeah. And the skills are really practical. The skills can be like, you know, read a Word document, read a PowerPoint, write a Word document, write a PowerPoint, talk to a data whatever, write some code, whatever. It doesn’t have to be whatever.
Justin: I’ll give you an example in my professional life. I was flying to Canada. I was doing some work on the plane, and then I was told, “Oh look, there’s this whole procedure that we need to integrate into what you’re doing. Can you do that please?” And I’d been flying, working, I was exhausted. It’d been a 16-hour day. I just fired up Claude Code and said, “Claude Code, here, look at, here’s what I’ve got, here’s the work I’ve done. I’ve been told to integrate this procedure into it. Can you do it for me please?” Half an hour later, it was done.
Frank: Brilliant.
Justin: That’s… you know, so they’re the tasks. That’s what I’m saying. They’re the practical tasks. So, have you used Cowork? Have you looked at it?
Frank: No, I haven’t. I mean, for one thing, even so, I have to admit, I have terribly neglected Claude and I keep saying to myself, I need to go back to Claude, I need to do a big deep dive into Claude. When I heard about Cowork, I was like, oh, okay, this is the perfect time for me to go and look at Claude again, because I had been hearing a lot about people using Claude Code for all kinds of things that are not code related, because very little that I do involves coding.
Frank: But I do a lot of content workflows and I’ve been hearing a lot of really good things about using Claude Code for content workflows and this Claude Cowork. The problem with using Claude Code is you have to use it in the command line interface, which, like I’ve said to you before on the show, to me is like entering The Matrix. I just see those luminous green numbers. So it’s not a skill I particularly want to develop.
Frank: But I was like, okay, I’m going to have to do it. But that was part of the friction of bothering to go back, that I’d have to learn this. So I heard about Cowork and I was like, great. You use Cowork in the desktop app for Claude. It’s only available at the moment, I believe, on Mac, but I’m a Mac user, et cetera. So I thought, this is perfect. This is the brilliant way for me to get back in now and look at Claude for content workflows with a Claude Code type approach.
Frank: But then I saw that it’s only on the top tiers, between a hundred and 200 dollars a month, and so I was like, well, I haven’t taken that leap yet. I may do so just to explore it, but I’m not sure that I would get the ongoing value from it to actually be paying it on an ongoing basis. We’ll see.
Justin: We’ll see. And then what I think is interesting about those two stories, and we’re going to move on to the third big one this week, is that it tells me a lot about the direction of the companies. OpenAI are introducing ads while Anthropic are introducing a tool for workers. So OpenAI are a B2C company and Anthropic are a B2B company. That’s what these two things are telling us.
Do we want Gemini hyper-personalising us?
Justin: Third big story of this week, Google hyper-personalisation.
Frank: Yes. This…
Justin: You love this, Frank.
Frank: I haven’t… I don’t know what I think about this. I don’t know what I think about this. So first of all, hyper-personalisation is basically where you can opt in. Thankfully you have to opt in, it’ll be off by default. You can basically opt in and you can tell Google, or tell Gemini, that it can now access anything on your Google account. So it can access, you know, it could access your emails, it could access your YouTube viewing history, it can access your photos. That’s actually a fascinating one. So basically everything in your Google universe it can access. So now it’ll be able to do incredibly hyper-personalised responses to you.
Frank: One example, and I don’t know if this is an example that came from Google or a hypothetical one I read somewhere, but it made me think. They were saying you could say to Gemini, “Recommend a restaurant in the area for me,” and Gemini could analyse your photos and look for your photos of food and food selfies to figure out what cuisine you like in order to pick the right restaurant.
Justin: Interesting.
Frank: Mind-blowing. And now again, I don’t know if that’s a hypothetical or a Google use case. I did see Google saying things like a guy was renting a car and they needed his licence plate and he couldn’t remember it, and he was able to ask Gemini, you know, go to Photos, find a picture of my car, grab my licence plate.
Justin: Yep. Interesting one. Now, tied in with that, and I didn’t put it on the board, but it was Google, I think, pretty sure it was Google this week, released UUCP and it was a shopping protocol. So it’s a protocol where vendors, sellers of products, there’s a standard way for an AI model to interact and then complete the transaction. There was a similar standard released last year and it was done with the big vendors. So Stripe were on board, Etsy were on board, and there were a couple of other vendors on board as well. So you can see, the two use cases they often give are, “I want to buy a pair of runners,” or “I want to buy flights.”
Justin: Well, with hyper-personalisation and a protocol to actually complete the transaction, you can see that Google are putting all the pieces together to enable all of that. I think it’s okay. Again, what’s interesting to me is that OpenAI are trying to kill Google and Google are fighting back and they’re doing a pretty good job. So Google are trying to get their knits into you, basically. They’re trying to make their product really sticky by knowing everything about you.
Frank: Yeah. Which, you know, they talk a lot about the security and the fact that it’s not shared, it’s not trained upon, it doesn’t go into training data, all that stuff. But there’s just that part of me that can’t help but think about the dystopian sci-fi movies that we’ve all seen. Do we want an AI that’s helping us really knowing that amount about us?
Frank: Right now as it stands, I think it’s just going to give you, you know, cool, a lot of the time, results that are more tailored to your taste, and it’s pretty cool. I would just be concerned a little bit about, you know, it’s still a tech company having an AI system that now can model you in a creepily accurate manner and there’s just something about that that puts me on edge.
Justin: Do you know what AI is also very good at?
Frank: What?
Justin: Let me push you over the edge. Do you know what AI is also very good at? It’s anonymising data. So they may not train on your data specifically, but I’d read very carefully to see how it’s worded because they could take your data, anonymise it, and then train the model. They’re not using your data specifically, but they are really using your data.
Frank: If that…
Justin: And you could do very clever things with that. If I had a billion users, I can anonymise the data and then amalgamate the data and say that out of this billion users, these are the sort of queries we get, these are the sort of links we have. You effectively do train on your data, but it’s not your data.
Frank: Yeah, and now, you know, now you can shift that back to, say, Google Ads and, based on particular things that people do online…
Justin: Yep.
Frank: …you can use that massive amount of anonymised data to be able to say, well, these kinds of people who do these things will be interested in this. And it all becomes, yeah, it just all becomes that little bit more, that little bit closer to the dystopian future that we’ve seen with, you know, hyper-personalised ads based on everything they know about you, popping up on screens as you walk down the street.
Justin: Who knows, but this is the same company that still gives me the ads for the rubber boots when I bought the rubber boots about six months ago.
Frank: Very true, very true.
Justin: It may not be that bad.
Why does Apple need Google for AI?
Justin: Now interestingly, the other thing that came up, and we won’t talk about it for long, but it was Apple signing a deal with Google. We talked about this before, but the new Siri is going to be released and it’s going to use Gemini. So Google are totally killing it in the B2C space. So no matter what type of phone you use, it’s going to be a Google model that you’re going to be talking to behind it.
Justin: But the reason I bring this up, Frank, is not because Google are killing it. It’s to ask you the question, why do you think that Apple cannot write AI software? They’re a huge company with loads of money. Do you think that they’re sort of leaning into this privacy thing and leaning into “secure your data” and their secrecy and security aspects? Basically, the regulations internally within Apple have prevented them from writing a decent AI model.
Frank: Short answer, I haven’t a clue. Longer answer, I think they were kind of rightfully wary about AI and, you know, to a degree still are. We saw, for example, Mark Zuckerberg was relatively late coming to the AI party and then he struggled initially to put a team together. Because if you look at the AI landscape in the States, it’s all the same people come up time and time again and they’re just moving around from company to company and setting up their own. This person left and did this company and, by the time Mark Zuckerberg entered the race, there were very few people left to hire, and he was very lucky to, at the time, eventually get Jan Koon on board.
Frank: We know how that turned out. But, you know, I just wonder, did Apple, you know, did they wait far too late and there was just literally nobody left to head up an Apple AI division that could do what they needed to do? But honestly, I have no idea, Justin.
Justin: Well, I think, to me it signposts the danger of regulation, because it dampens, it dampens innovation. It’s like throwing water on a fire. You need to let people run, do things, and then rein them in afterwards.
Frank: But what are you talking… sorry.
Justin: Yep.
Frank: I’m glad you brought this up because, you know, it hasn’t been a very argument show. So…
Frank: Were Apple not under the same regulation as OpenAI? Well, Google is a better example because Google are not solely AI focused, but were Apple not under the same regulations as Google, and Google have gone all in on AI and Apple have not?
Justin: The more insightful of our listeners will have seen that I’m trying to draw a line between the internal rules of Apple and the internal rules of the European Union. So yes, yes.
Frank: I see what you’re at. Okay. I see what you’re doing. Okay.
Justin: So yes, they had the same regulations, but actually Apple imposed on themselves regulations because they had a particular view of how data and the world should be treated, similar, I would say, to the European Union’s view. And I’m saying, look at the outcome. The outcome is that they couldn’t develop a model because they weren’t able to do the things that they needed to do.
Frank: So in the… you brought this up very late in the show. In the short term, I think Apple is taking a little bit of a pummelling about, like, “Oh, you haven’t figured out AI, why haven’t you figured out AI,” et cetera. In the long run, I don’t know if it will affect them. I think I’ve made this argument before, that I actually think their caution might end up being fine for Apple because they’re not like an AI company. They are a “we produce products that people love and they just work”.
Frank: We almost put in, we didn’t put in today’s show, the story about Salesforce kind of slightly backing off their whole “LLMs are going to do everything for us and we’re going to fire everybody and just use LLMs.” They’re slightly now backing away from that and they’re saying, yeah, actually we’re doing a lot more deterministic automations that include LLMs rather than just saying to an LLM, “Here, do everything.” Why? Because LLMs don’t just work. They do amazing things, but they also go completely off the rails, do crazy things, and it can be a bit dangerous to give them full autonomy.
Frank: And I think Apple are kind of looking at that and going, yeah, where does AI actually fit into a product that just works? And I think they’re figuring that out. They’re figuring it out now with Google Gemini, and in the long run, I think it’ll all work out fine for Apple.
Justin: Okay. I disagree with you. And if you look back at, you brought this up, if you look at Cowork, which Anthropic wrote… oh, another interesting story actually related to this, but anyway. Cowork took how long to write?
Frank: I think I read somewhere that they developed Cowork in 10 days. Did I…
Justin: A week and a half.
Frank: That sounds crazy. That sounds crazy, but…
Justin: That’s how long it took them to develop it. The other interesting thing that we didn’t put on the board is Anthropic have cut off Claude Code access, API access, to xAI. And they previously had cut it off to OpenAI because they’ve got the best coding model, and why would you give your best coding model to your competitors to write code that’s going to compete against you?
Justin: Apple are in the same hole. They don’t have a model. So when all the other companies are driving ahead and writing stuff in weeks, Apple are still, they’re still in the dark ages. They’ve got humans writing the code and it’s taking them months, and it’ll become a gap which gets wider and wider.
Justin: To my point this week, there was a dude on the internet, I can’t remember his name now. He had hooked Claude Code into Thinking Machines, which is a Mirati company, and used it to fine-tune a model. It did a better job than a human. This is models improving models, it’s self-improvement. Apple are going to be left behind because they don’t even have the tech to get started, and they won’t be allowed to buy the tech because the other companies aren’t going to give it to them.
Frank: This is one of those things we should place a bet on, isn’t it? A 10-year bet. But okay, you mentioned…
Justin: I mean, I’ll do my standard terms.
Frank: Yeah. What are your standard terms?
Justin: One pint, please.
Frank: Okay. Right.
Justin: I bet you one pint that within the next five to 10 years, I’d even say five years, somebody will do to Apple what Apple did to IBM.
Frank: That’s really interesting. I will take the bet, though.
Justin: You will take the…
Frank: I will. I will. Yeah, I will. Just out of, just out of, just out of obstinance. You mentioned there about Anthropic cutting off access to xAI, which just leads nicely into, let’s just touch upon this very briefly.
Did OpenAI really defraud Elon Musk?
Frank: What is Elon up to?
Justin: We should have a theme tune for this. “What is Elon up to?”
Frank: So, you know, he brought a case against OpenAI. OpenAI kind of said, this is ridiculous, this should be dismissed. Apparently the judge said, nope, it’s not ridiculous. Actually, Elon has some very sound arguments, so the case is going ahead. Can’t remember when. April 2026, I think it’s set to start.
Frank: And Elon, I’ll have to look up this figure, Elon basically is saying that OpenAI defrauded him because it was set up as a non-profit and he invested, and now they’re trying to turn it into a for-profit, and he wants 79 to 134 billion dollars in damages for that upset of being defrauded by OpenAI.
Frank: And just to add to the drama and the gossip and everything with all this, and, you know, we’re definitely going to see a movie about this, OpenAI put out a blog post, I think it was yesterday, and the title was something like “What Elon Didn’t Say” or “What Elon Left Out” or something like that. Basically going through some of the evidence that Elon provided to the court of the fact that he was defrauded. They went back to Greg Brockman’s original journals and meeting notes, et cetera, and they published everything and highlighted, you know, this is the line that Elon is saying as evidence of what he’s saying, but here’s the three paragraphs that contradict it that were also in the meeting.
Frank: So yeah, just…
Justin: I presume they put a line that, you know, “This report is being produced by, you know, ChatGPT Deep Research.” I mean, that would seem to me…
Frank: Yeah.
Justin: …to make a lot of sense. I mean, the obvious question that comes to mind, I think Elon might win this case, by the way. But the obvious question that comes to mind is if it was set up as a non-profit, Elon didn’t expect to make any money. So why is he being defrauded? I mean, you expected zero, you got zero.
Frank: I guess, philosophically defrauded. He invested in something that was going to benefit society and not be for-profit, et cetera, and now they’re, now it’s just a… now they’re trying to turn it into just a business. And, you know, Elon, he’s so philanthropic and he has society’s best interests at heart always, so, you know…
Justin: He does. He does. Yeah. Genuinely, I don’t think, I don’t know that you could say that Elon… I mean, I again would lean towards Anthropic. They have a department, they study the effects of AI on the least educated people, and they have a department and they have specific KPIs to use the Anthropic models to help the lowest educated people gain the most from AI. That’s the thing within Anthropic.
Justin: If I’m to look at the big AI companies, I really think only Anthropic could be the only ones to really claim that they’re trying to do it for the good of everybody. I don’t know that you could say the same for any of the other companies.
Frank: No. No, I don’t think so. I don’t think so.
Was Stranger Things written by ChatGPT?
Frank: So, final story of today. This is, I think this is hilarious. Do you watch Stranger Things?
Justin: I watched the first season. It was good.
Frank: Only the first one? Okay.
Justin: I didn’t like the start of the second season and I sort of dropped it. But yeah, people like it.
Frank: I think I’ve lost track of the numbers of the seasons, but I think I watched up to season four and I will, I will watch season five. But they broke it up into parts over Christmas and I just couldn’t. I just couldn’t be watching an episode and then waiting for another episode, so I’ll binge it at some point.
Frank: But apparently fans were not that happy with it. They didn’t feel like the quality was the same as previous seasons and they kind of felt like the resolution at the end was a bit meh. And then what happened was a documentary apparently came out with some behind-the-scenes footage, and the Duffer Brothers are the people responsible for Stranger Things.
Frank: So there was footage of the Duffer Brothers working on their laptops with multiple tabs open. Apparently the laptop wasn’t even in focus, so it was kind of a blurry shot, multiple tabs. One of the tabs had a monochromatic round logo and fans, Stranger Things fans, leapt on this and said, wait a minute, that’s ChatGPT. And so this whole thing blew up on the internet that season five was written by ChatGPT and that’s why it was no good.
Justin: Yeah. And I bet you they used 4.0. Everybody loved 4.0, that was better at writing than everybody else. Do you think it was written with ChatGPT? Do you think you could write a TV series with ChatGPT?
Frank: I mean, you’re not going to go into ChatGPT and say, “Write season five of Stranger Things,” but you might use it to aid your writing, certainly. And if you’re a good writer, I think you could do that well. I think there’s very few shows that manage to go five seasons without a dip in quality at some point. It happens, but I think it’s pretty unusual. So I don’t think ChatGPT is the issue here. I certainly don’t think you can just go, “Write season five of Stranger Things,” and then submit that and shoot it.
Frank: But, you know, did they use ChatGPT as part of their process? I don’t know, and I don’t care. The question is, is it good or is it not good? End of story as far as I’m concerned.
Justin: Hello, Frank. Very good. That’s great, because that’s a big turnabout from you wanting to know that everything that was created by AI should have a stamp on it to say “created by AI”. Delighted you’ve come over to my side. It’s been a great week. Thank you, and we look forward to seeing you all next week.
Frank: Got me good. Got me good. Nice one.
Justin: Take it easy.
