DataTopics Unplugged
Welcome to the cozy corner of the tech world where ones and zeros mingle with casual chit-chat. Datatopics Unplugged is your go-to spot for relaxed discussions around tech, news, data, and society.
Dive into conversations that should flow as smoothly as your morning coffee (but don't), where industry insights meet laid-back banter. Whether you're a data aficionado or just someone curious about the digital age, pull up a chair, relax, and let's get into the heart of data, unplugged style!
DataTopics Unplugged
#32 NY Times sues OpenAI and Rabbit R1s LAM Device
Welcome to the cozy corner of the tech world where ones and zeros mingle with casual chit-chat. Datatopics Unplugged is your go-to spot for relaxed discussions around tech, news, data, and society.
Dive into conversations that should flow as smoothly as your morning coffee (but don't), where industry insights meet laid-back banter. Whether you're a data aficionado or just someone curious about the digital age, pull up a chair, relax, and let's get into the heart of data, unplugged style!
In this episode, we're joined by Ben Mellaerts, an expert Data Strategist, to unravel the intricacies of some of the hottest topics in the tech world.
- Rabbit R1 AI Action Model: A deep dive into the hype and reality of the Rabbit R1, a potential game-changer in AI.
- NY Times vs. OpenAI: Exploring the implications of the NY Times lawsuit against OpenAI, a case that could shape the future of AI in media.
- ChatGPT Store: Unpacking the potential of OpenAI's ChatGPT store and its impact on the tech and data landscape.
Intro music courtesy of fesliyanstudios.com.
Now we can go.
Speaker 2:All right, hello, so are we going to hit it? Oh yeah, yeah, hello, hello, hello. Welcome to 2024. Yes, another year. So well, my name is Marillo. Oh, yeah, in real life.
Speaker 2:So, part good, we are live. We're live on YouTube. This is a thing now. We're doing it. So welcome to Data Topics Unplugged, casual, lighthearted, weekly, tentative Weekly. Last week we weren't able to be here, we were still recovering from the holidays. Short discussion on what's new in data, from lawsuits to rabbits, anything goes. We're on YouTube.
Speaker 2:As I mentioned, livestreaming. Friday is at 4 pm. This is the first time. This is the second time officially we're live streaming, like live live streaming. And then one time before that we did a live stream, but we cheated a little bit. We did like a private live stream that we just made a public afterwards. We were too nervous. You know I still had gained a couple of some weight back then, so I had some time to cut it off. Today is the 12th of January of 2024. My name is Marillo. I'm your host for today. I'm Joy by Bart Hi, and I'm Joy by Ben. Hey, who's Ben? Ben is a data strategist here at Data Roots. He is a runner as well. He ran a marathon actually last year, 2023. Indeed, okay, he's going to go to Tanzania and he's a proud Rootsy award winner.
Speaker 3:Yes, yeah, we're here with a winner, Bart. It's so random when you say he's going to Tanzania.
Speaker 2:Yeah, I know, maybe the context here for everyone is I demanded fun facts and then it's not fun, so I'm not fun, it's true, I don't have fun facts. The best thing you can come up with is I travel, okay, and a marathon.
Speaker 1:Yeah, an American.
Speaker 2:Yeah, that's true. Wow, have you ever run a marathon, bart?
Speaker 1:It's an extremely long time ago, yeah, but you ran all along. I registered for the 2025 marathon and it's going to sound very impressive for the European championships, but everybody can participate, but it's going to be 11.
Speaker 2:You don't need to say that, just say you're going for the European championships, people are going to do two things and then you just don't correct. It's not lying.
Speaker 1:I'm going to do it as well.
Speaker 3:Oh you're also going to the championships, four or five people from Data Roots will join. It's really cool. They opened up for the first time. Let's go ahead. We created runners. Cool, it's going from Brussels to Lueva, so back to our home base, so it's quite nice, wow, okay.
Speaker 2:So you didn't think of that as a film. Well, okay, so let's change. Ben is just not very quick on his feet. He's fun, but all right. And well, maybe, ben, is there anything else you'd like to introduce yourself?
Speaker 3:I've been here for about two years now working as a data strategist. Indeed, I love topics like data strategy, data governance, everything which is about data, but more on the strategic side, not too technical, so yeah.
Speaker 2:Cool. So how's everyone's break, christmas, new Year's, anything fun, any cool gifts you would like to share with the internet? It was fun.
Speaker 1:A lot of festivities. Take it with gifts, cool gifts, nothing surprising. I have a new. The socks I'm wearing, oh nice, I swear gifts it's always nice, good gift, so nice.
Speaker 2:Yeah, right. So I'm excited about socks is when you know you're aging, because when I was a kid.
Speaker 1:What are you saying?
Speaker 2:Because when I was a kid, I would get socks and I was like, ah, I didn't want.
Speaker 3:You're like, I'm excited about my socks. It has polar bears on it. Oh nice, Can I see your socks?
Speaker 1:Oh, ankle socks, yeah, winter, wow, that's what I have, that's how I roll Bart.
Speaker 3:It's even worse than not having fun. Facts, you really.
Speaker 2:Oh wow, ankle socks. Yeah, that's how the cookies roll back in the motherland, in Brazil.
Speaker 1:20 years ago maybe.
Speaker 2:You got a ring right. I actually get an aura ring as well. Now I'm joining the club, bart. There we go. Yeah, just like that, cool, cool, it's a thing.
Speaker 3:It's a thing. I would buy one too. You will, yeah, no, maybe next time you have it. Yona says it.
Speaker 1:And how is your sleep? Because the aura. Can I explain it to the listeners? What does an aura ring do?
Speaker 2:So an aura ring, it's light. Well, I used I have, for less time. I know you use more of the features than I do, bart, so free, free compliment, whatever I miss out. But you can almost think of a fitness like a health tracker. A lot of them they are like watch, like, and this one goes on your finger so it tracks your sleep, it tracks your activity If you're running, it tracks heart rate, o2 saturation. There is a subscription to it, but actually I was looking and they actually even have meditation sessions and a lot of stuff. It's really, really cool and they actually track.
Speaker 2:I found out this morning as well, like if you do a meditation session, they will track how your heart rate goes throughout the session, so you can see, like, if you actually calm yourself down or anything. I thought it was quite interesting. They sent some reminders and whatnot. And the last thing, too, is that the battery life of this is much better than an Apple watch, which I also have, right, so it's less like Apple watch is always like popping up with notifications. Oh, you have an Apple watch and an aura ring. I know, I know, and I'm going to get a harm monitor too.
Speaker 1:When I was getting an aura ring, I was evaluating should I get an Apple watch or an aura ring. You just go and an end. Exactly Just all the above. And how was your sleeping score last night.
Speaker 2:Last night. I think it's fine too, because before the podcast, martin did a presentation about JPD coach and he mentioned about the aura ring and then he said he didn't want to livestream it because it's too personal. You know it has. Oh, I had a bad night's sleep. We go on the podcast. How was your sleep, bro? Let's see, let's see, let's see it was. It was good, I didn't sleep much 70. Oh, that's good, this is good. That's what I learned. Just read what's there and how many hours of sleep? Uh, six and six hours and 40 minutes. But it tracks, like so how long you actually slept. It actually tracks naps. I took a nap the other day and actually said how did you take a nap? Confirm it.
Speaker 1:What percentage is deep?
Speaker 2:sleep Rem sleep 24% and deep sleep 16%, so that was also a little too low, did you?
Speaker 3:sleep with.
Speaker 2:Well, now it's getting a bit deep with my partner and at some point my dogs jump on the bed. That's how it usually happens, but, like the night before, it was 85, so you can see there. But that was cool, that was cool. Um, what about you, ben?
Speaker 3:Did I? What gifts did you? Actually, I do have a fun fact, because my girlfriend and I we bought a house, so we got some congrats house stuff as a gift, which was kind of nice. Uh, some running gear, but nothing, nothing too special, to be honest. Cool, cool, cool.
Speaker 2:All right, without further ado. What do we got for today? So lawsuits? What do you got, ben?
Speaker 3:Yeah, we're discussing the holidays and it was in between Christmas and New Year's Eve that we got the news about the New York Times. It was on December 27th and I read an article about the New York Times suing open AI and Microsoft. So they had a claim about copyright issues. Uh, because open AI and Microsoft are using the articles of New York Times at least they used to do uh to train their models and now, of course, they are um, producing um based on the prompts, information about these articles, and the New York Times isn't really happy with that. They used to be in um negotiations with open AI to talk about licensing agreements. They didn't do it after all, and now they are. Uh, yeah, they are suing them.
Speaker 3:And if you read the article, I kind of agree with the New York Times a bit, because they are discussing three types of loss actually. Uh, first and foremost, they are talking about the fact that traffic went down on their websites since the introduction of open AI Um, and it's not really about the fact that they stole or used the articles to train it, but it's more about the fact that they are creating a substitute with it. So people need information, they have a need for information and they will fulfill the needs on chat, gpt and not anymore on the New York Times. But the problem with that is that it based on the things that New York Times are creating.
Speaker 2:Okay. So it's like is it like, if I start my blog and I write about stuff that I read, but I only read New York Times, and people start reading more my blog than they read New York Times, then that's kind of comparable.
Speaker 3:I would say so. It depends if you're. What's your revenue model, right? Um, okay, I think what they are. What they are also claiming is that, um, open AI trained their model based on paid subscription data. So when you go to the New York Times, you have the free articles, but also the paid ones for which you need a subscription. Open AI is saying that they are not training, uh, their models based on that data. Um, the New York Times is saying that they are doing it. They have some examples of that. They have evidence? Oh, they have. And then, of course, it becomes a bit tricky. They put evidence in the claim. How can you prove that?
Speaker 1:Because they ask uh about uh news facts about historical things that happened to the model and they give back more than one to one references on uh, how it was written in the article and something else. We refer to the source.
Speaker 2:Okay, and is this like uh do?
Speaker 3:you pay for the subscription, then no, I'm not a New York Times reader.
Speaker 2:So there are ways to go around it, right, Cause I clicked on the on an article here and it says here you have to pay to read it. But I'm pretty sure that, like I don't know if you're able to like incognito or something you can, that's not really the the point here right.
Speaker 2:Yeah. So I was just wondering is like is there a way in which, like, if, for example, medium I know medium used to work like this, right, they had this oh, if you want to remore, like you usually had X articles, and if you read, say, oh, you want to remore, you need a subscription. But if I went to an incognito window because there's no track, I could still read it. That was the. That's what some people not me, not naming names would do. Now is that illegal? Like, is this? I'm kind of going around the system that they put right, but is this something like am I breaking laws or is it? You know what I'm saying.
Speaker 1:If the next day, you built your own magazine with the same content, I would say that this is legal because there's copyright on these articles. Yeah, I see what you're saying. I mean the problem that I stated here and I think that's, I think, simply happened, whether or not New York Times like it happened at a very large scale. A lot of content that was copyrighted in this case tax, but also images have been taken to train a model, and that model can now to some extent substitute what they're also offering.
Speaker 2:So they use the content that was copyrighted without the permission to build something that now to some extent substitutes what they already do but couldn't I say that as soon as they use the copyrighted material, that's where the violation is, and whether they built something or didn't build something, or but what they were saying actually now it's been a while since I read the whole overview is that what they're demanding is that everything that was built on top of this that it is no longer used, which would have major implications.
Speaker 1:It would be like everything which, yeah, and it's very hard because if there is an agreement on this from court, I mean a lot of other parties would follow the same. And that's why this lawsuit, I think, is interesting, because it will set out a bit the legal contours, the legal frameworks for generating the VAI and what kind of contents you can use, because we don't have this legal framework today.
Speaker 2:Yeah, and also I think it's interesting. Well, this is all in the US, right, the US court and everything, but it could impact us as well, right?
Speaker 3:Yeah, I read something about the Fair Use Doctrine, which is like a legal principle. I'm reading this. Of course, it's a note that is primarily applicable in the United States, so it states that you can reuse stuff that is copyrighted, but there's a degree to it and there's not really a good definition about that. But actually, the fact that they are creating a substitute this is the part that they are really focusing on with the New York Times. Okay, this is a substitute, yes or no? It's a very difficult question, so it might take some time to have an answer on this, but I can imagine that for news agencies, because this is one part, let's maybe delve into the other part as well.
Speaker 3:But the first part is really about revenue laws and it's subscription revenue laws. It's also lower traffic on their website, so they have less incomes from advertising and so on. And thirdly, they were also mentioning that they have I think it's called WIRETEPPER, which is a product review site of the New York Times, and they mentioned that Bing is like reusing their product reviews and that way, the New York Times is having a loss on the affiliate marketing, because they used to have links to the product and they get money for that. So they have less income because of the reused material of Bing. I see If Bing would have product reviews themselves, it would be different. But they are reusing the content from the New York Times and still the New York Times doesn't get the affiliate marketing income.
Speaker 2:Maybe the thing like you're building something that could replace New York Times. Do you think that's a? Because it's not exactly the same, right? Like it's just a question answering thing. So it's like okay, I know I'm being devil's advocate here, right, but for example, let's say we have Stack Overflow. Then I write a book about programming and people stop going to Stack Overflow because of my book and to say, well, no, but let's take as an example, I think, what was not possible at all two years ago.
Speaker 1:If you have a subscription as a company to, for example, associate Press or Reuters and you say in between that I put a LLM, that's where you say from all these articles, generate an article in a slightly different style with this target that this audience and I'm called is theleventimescom, and you certainly have a new newspaper with very minimal effort.
Speaker 2:Yeah, yeah, yeah.
Speaker 1:And these type of things. They challenge the market and maybe for good, maybe for bad, I don't know, because it's also Today. There's two sites, I think here in New York Times, with content creators in general, there's a big question like what does this mean for content creation and ownership? And the other site if they cannot use all this data, you could question whether or not we would be so far launched today. So would it hamper development, innovation, competition?
Speaker 1:If you need to adhere to copyright laws and I think in the specific case of OpenAI, they were very early with this they also have a bit of a competitive edge because they were so early. So this was not a talking point. Today there's much more being scrutinized the copyright aspects. You also have on most websites a robottxt where you can now specify. This should not be crawled by LLMs, and people in the big companies are adhering to it. So you also have this bit of this unfair advantage where OpenAI has a base model that has a huge amount of data available and they just took everything without any consideration. But at the same time, that's why they're so good.
Speaker 2:I think that's the main thing that leaves a bad taste in my mouth there is that you're increasing the gap between OpenAI and other competitors because indeed they were the first ones. If I were to do the same thing that OpenAI did, I probably wouldn't be able to get away with it, because it is in people's minds.
Speaker 3:But I think there is a solution and I think they know about solution, because you were mentioning the Associated Press. They do have a licensing agreement with them, so they pay the Associated Press to use their data and then you can still have your advancements in the technology. Openai has an agreement with OpenAI.
Speaker 1:But the difficult thing there is that that is a solution for New York Times. From the moment that New York Times gets their, that's an quarter of the decision. New York Times is correct here. There are thousands of parties that can claim the same thing.
Speaker 3:Yeah, that's why it's so important. It's really important. I was amazed by the article and I'm really looking forward to how it will impact the world of AI and MLMs.
Speaker 1:Yeah, indeed, I feel like, if you take, For example, if you go to the imaging domain, mid Journey v6 was released not long before the holidays and it is extremely overtrained and very much overtrained on data on which there is copyrighted trademarks. So if you ask an image about the symptoms Simpsons in one shape or another, you get an exact representation of the Simpsons Really, but you can do it for a lot of different characters. So there's very clearly data used, copyrighted data, and again you can make the argument like it's good for innovation but at the same time, if you're a creator, you don't know what yours anymore. What are people allowed to do with this?
Speaker 2:Do you think there will be a New York Times lawsuit moment for Mid Journey with the copyrighted?
Speaker 1:images. I think this is extendable. I think this is really carving out the contours of what is possible. What is the legal framework for using copyrighted stuff in this context?
Speaker 2:Yeah, and also you had mentioned before we started recording that the reputation of New York Times as well is.
Speaker 3:Yeah, there's a second part. So the first part is more based on revenue and the business model, and I think Bart has a good example about the loven digestor I don't know what you the loven times, the bird part, it doesn't really seem fair right, and if you were talking about a block, I really like you, but if you would start a block I don't think it will lower traffic on the New York Times. So it's really about that. But the second part is, yeah, reality check.
Speaker 2:Yeah, it's like ah, there go my resolutions for 2024.
Speaker 3:The second part is indeed about reliability, and it's also part of the claim. By the way, the New York Times states that on Bing. When they are asking for information, often times the New York Times are put there as a source. But the problem is that Bing hallucinates, so it's not only part of the articles but also a layer on top of it, and it's often incorrect. They also added some examples of that, and what you then get is that people are finding the New York Times unreliable because they, like they associate the outcome of Bing with the New York Times because it was mentioned as a source, and people are lazy to double check the source, of course, and especially you read it and check if it's true or not.
Speaker 3:Right, I can imagine, to be honest, but that's a big problem because I'm working at a media company myself and for a news agency. The main USP order, the main important aspect, is to be reliable and trustworthy. So that's also something that they add, that it's hallucinating and that it's associated to the source of the outcome, of the reputation.
Speaker 2:Maybe to wrap this up, do you think New York Times is going to be?
Speaker 3:successful. I'm not a legal expert, so who are you?
Speaker 2:Maybe you're too rude for someone. What are you rooting for?
Speaker 1:I think we will, to the very least, learn to some extent. What are the boundaries Today? We don't know. I think that is the end.
Speaker 2:And this isn't the US right. Do you think what happens in Europe will be? They also have an agreement with.
Speaker 3:I forgot the name, but it's a German publishing agency. They have a business insider, amongst others. They also have a licensing agreement with them. So I just think this is the way to go Just have agreements in place. Lawsuits, no, maybe. Yeah, it can really be an important moment for the development of LLMs. Yeah, that's true. So yeah, maybe in five years we will say that it was a good thing that they did this. Yeah.
Speaker 2:And I thought this. I mean, I kind of thought that something was going to happen, but I'm not sure I expected New York Times to be the one to pull the trigger.
Speaker 3:Who would you expect?
Speaker 2:I'm not sure. I guess it's because for me it's not a one to one. Whatever ChagYPT outputs is not a one to one with New York Times posts, and that's why for me it was a bit surprising.
Speaker 1:But it is an organization that puts tons and tons and tons of investment in creating content.
Speaker 2:Yeah, yeah, yeah no.
Speaker 1:For sure, and you have some other entity just using the content for something.
Speaker 2:Yeah, no, it's true, it's true, it's true. I see that, that part of it and, to not change the subject, more about ChagYPT, so not the legal side.
Speaker 3:Yeah, another topic today that we can discuss is indeed the ChagYPT store. It went live two days ago, I think, and it's definitely something new, something interesting that we should discuss now. What is it? It's kind of an app store for customized GPTs. I think that sums it up very shortly.
Speaker 2:How can I so? What do you mean? Customized a GPT? Doesn't that mean that I need to have training data and retrain it and then I publish it there? Is it just prompt engineering? What is it exactly?
Speaker 3:Well, it's GPT, so it's pre-trained. But you can customize it in a way that it's focused on one or another subject. So, for example, canva as a ChagYPT app in the ChagYPT store and there you can ask, for example, to generate an image. It will come up with some proposals and then you can even click through to Canva. So that's something interesting to me in terms of business model or how businesses will take advantage of the ChagYPT store.
Speaker 2:So when you say, click through to Canva. You mean I'm on the ChagYPT Canva bot and then you will give me some visuals, I guess like images, and then I can click through and you will take me to the Canva portal with stuff, is that?
Speaker 1:what it's doing. So, functionality-wise, what you can do is you can build your custom GPTs, and that functionality existed already two months ago. I think roughly what you can see a bit as a GPT where you have a pre-existing prompt, so you already gave it some instructions and you don't see the pre-existing instructions, so you start using it and so you can define your custom instructed GPT. You can save it, you can edit it later on and with the store, you now have this worldwide overview of things that are published there and that so that you can more easily find other people's stuff.
Speaker 2:What's the pricing set up there?
Speaker 1:I don't think that I haven't seen things that are paid yet. So you can. If you have a plus account or a team's account, you can basically already create your own and be known to you. The copyright for the Data Topics podcast also comes from a custom GPT, so there's a custom GPT that also has pre-existing instructions that say this is the type of copy that we typically use. I'm going to give you these and these notes from the episode. Make sure to structure it like that. So all that instruction in the prompt is already. I did it one time and now we're using it every time for the copy of the podcast.
Speaker 1:It is well you can actually choose it. I only publish it for myself, but you can also publish it for everybody, and before the store was there I could already share it via link so that all the people could use it. But I think the store is really to more easily explore and find stuff, and probably there's also going to be a paid version of things.
Speaker 2:Yeah, like if you want to get money from your bot, kind of thing.
Speaker 1:If anyone else wants to use the Data Topics, copyright bot.
Speaker 2:Yeah business idea.
Speaker 3:But you do need to check your team plus our teams. Indeed, and I've read something about the business model as well that how they would do it is that if you have a well performing judge, you be customized, but then at least if there are many interactions with it, then you get a piece of the cake. So you will be paid out based on the interactions that people have with your app.
Speaker 2:Interesting, but then it's not yet official. Yeah.
Speaker 3:I still have to announce it.
Speaker 2:But then it's basically like a prompting on top of ChagPT that already exists.
Speaker 1:Yeah, you can also do information retrieval. I think you can also link it up to endpoints, but it's not like pre-configured stuff.
Speaker 2:Yeah, yeah, yeah. It's not like you send new data, you train more weights or you take the win, like you're not some Maybe in the future I don't think you can do it now. No, Okay, and what if you want a multi-modal stuff, like if you want, like you mentioned, kama, like if I want to create images on the data topics style, or if I want to create images on whatever? Is this something you can also do?
Speaker 2:You're specifically talking for Kama, then no, not Kama, but I'm wondering, sorry. I mean I think Kama is an interesting one because a lot of visual stuff right. So maybe can you create a bot on the ChagPT store or the GPT store, whatever the name is that will be trained to create SVG images or whatever. Is this something? Is it granular to that level that you can change the outputs and how it looks and all these things, or is it more like prompt engineering? You can maybe take some components that are already there, but it's pretty limited. That's kind of what I want to ask.
Speaker 1:I think the way to extend it is via, I think, what they call it actions. Basically, you can call endpoints of all the services.
Speaker 2:Oh, yeah, yeah, yeah.
Speaker 1:So in that extension you can go very far Just by integrating APIs of all the products and specifying how you sense stuff there and what you can do with this stuff. It comes back.
Speaker 2:Okay, cool.
Speaker 3:So the main idea or the use case is that it offers a lot of reusability. You don't have to do it over and over again. You can just reuse it and not having to configure it every time again.
Speaker 2:I mean you mentioned that topics copyright. Do you have any business idea? New GPT on the GPT store?
Speaker 3:Ben GPT. Ben GPT no, but I've already looked into it and Altrails also has a GPT app and there are some interesting apps there. I'm really looking forward to what will be the top three apps in like three months.
Speaker 2:Yeah, that's really looking forward to it. Any predictions what it will be about, or any topics, areas, anything that you Just give me something so I can come back in three months and say, oh, Ben.
Speaker 3:Difficult. I think Kanva will be one of them.
Speaker 2:Okay.
Speaker 3:But if you look it up on the internet, there are some cool apps being built, you know.
Speaker 2:And there was already I mean, there was the website like there's an AI for that that already has a lot of stuff and it can't like what you're describing. I haven't looked into the GPT store, but it gives me a similar vibe kind of thing, but now it's on the GPT chat. Gpt like centralized kind of thing.
Speaker 3:Yeah, what I was saying is that you have businesses like Kanva, like Outrails and so on, and I find it interesting to see because some of them already built their own GPT within their company. For example, at Kanva, you have a lot of Gen AI possibilities. It's also probably based on OpenAI, I'm not sure. Sometimes it will be text, sometimes multimodal, but what will they keep on their own platform and what will they put on the app on the chat GPT, and what is their strategy for that? Because you can.
Speaker 3:There are two things. You can have a good app on the chat GPT store to have a lot of brand recognition and awareness and so on, and make it a really cool app and go really deep on that. Or you can keep it to the surface and already direct them or redirect them to your own platform really soon, because what I read is that Kanva does it really fast, so they see it as an incoming referring channel. Kanva, for example, you have people coming in through socials, through YouTube, through any other referring medium, and now chat GPT will be one of them.
Speaker 1:What is interesting, so to really see it as a new acquisition channel.
Speaker 3:Indeed To bring in new people, but it's really about where you want to put a focus on.
Speaker 2:Yeah. But I guess for them right, so the GPT from Kanva in the store is a free one. You can just go and use it if you have the. If you have, yeah, indeed. So I guess they want to go to their platform and maybe they have a pro, whatever on their platform. That's then you paint them from it. Interesting, interesting, cool. And the action you mentioned, gpt, their actions they use as a REST API part. Do you have any idea how that works? I'm just kind of thinking all out here.
Speaker 1:I think you need to specify the open API spec, or the endpoint that you're integrating, based on the open API spec, knows how to use it. You still need to specify what do you, what payload do you send to it. You basically give an endpoint, a name. I think that's how it works.
Speaker 1:And then, whenever you have a prompt that is associated with that name, then you can basically link this to an, it becomes an event that you can use for stuff, so that in this way, it's very interesting how they can think even the open API, open AI API.
Speaker 2:If you were like to code it up, you can even provide functions to it as well. Right Like that. So I guess it's something.
Speaker 1:Exactly the same. That's how you do and what they also have. Maybe that's what you're referring to. So there's also a code interpreter, which is basically a Python shell which is also available in the custom GPTs.
Speaker 2:Really cool and you mentioned, these are actions.
Speaker 1:The code interpreter is not called an action. Action is the endpoints, is the endpoints.
Speaker 2:Have you ever heard of large action models? Lambs, lambs? Have you ever heard of it? No, you're going to talk to me about it. Yes, I am no. So that's actually from Rebit, rebit R1. So I didn't know about it actually, and I was on Twitter X. I have to have some side note of it. I have some feeling like it's called X, doesn't work. It doesn't work. It's not Googleable Like on my phone if I want to look for the app. It's like it's horrible. So the logo is X, but even the URL is to Twittercom. It's not even X Twitter, just Twittercom. So it's a bit weird, right? Anyways, I digress. I was on Twitter X Shitter and then I saw a couple of times I saw this popping up. Once One was a marketing guy that he was saying oh, I work for a lot of big the Fang, you know marketing, and I think this guy he's the new Steve Jobs, and I was like, okay, let's get scrolling. And then I saw again like, oh, the rabbit keynote because the keynote was exactly the same as Steve Jobs.
Speaker 2:I was definitely inspired by it, because the guy he said that and I was always a visionary revolutionary and I was like okay, whatever right. But then I saw it again on Twitter, twitter X, and I was like oh, let's, let's have a look. And I went into this rabbit, the rabbit hole, get it Nice one. And they have to do it. Did you have the labels on it? There we go, the late but. But didn't you put labels on the buttons?
Speaker 1:You said, you said that I do, man, I only saw punch.
Speaker 2:Okay, okay, cool, thanks. Next time we'll know how this like anyways. So I saw it again. So I started, I went, I saw the keynote and I looked at what the thing was about and I thought it was interesting. What is it Rapid?
Speaker 1:R1. What is it?
Speaker 2:Yes, it's. So there are two parts. The R1 is like a device, so it's a very simple device. I will link the keynote and everything on the show notes. Of course it's like it's orange, so it's like have a funny, like looking is like a rectangle with a touch display. It has one like scroll kind of thing, a button on the side and a camera and the idea they tried to kind of on the keynote.
Speaker 2:They started motivated by saying your iPhone, your phone, whatever, you can do a lot of stuff, right, it's super powerful. But right now if you want to do anything you need to kind of use different apps for it, right? So sometimes you can be overwhelmed by apps. So sometimes it's not super easy to use those, so you're limited by the operating system. So what he was saying is, with the explosion of GPTs and LLMs, he's rethinking kind of like a mobile device that will work only on natural language. It will be catered for that, right? So that's the R1. That's the device.
Speaker 2:And then they have the other, which is the LAM, actually called the large action models, which is a new foundation models. It's the same thing as like the chp, t and whatnot, but it's focused on actions, on great like on performing actions. So you still need to also give the access to the API and stuff for like Spotify and whatnot, and then, once you do, you can just kind of talk to it. So they also have GPTs. Excuse me that, but they are optimized to work on the hardware that they have. So they have an operating system the hardware I'm assuming that it's fat that is specialized for these things. So then, like, they say that if you talk, if you talk about it, then you can get an answer much faster than you would if you were to use chp t or something like this. So that's one of the arguments. They also show how you can hook it up with Spotify or whatever. You just need to give permissions.
Speaker 2:And that's why I was also fishing and started with this question. Like, how does the actions work for Chagapiti? So I think it's somewhat similar, but it's like it's handheld. There's no like if you want to use the keyboard, you can. You need to shake it a bit and then you can get a keyboard and you can type stuff. But it's made to be used with natural language. Right, it has a camera as well. So if you take a picture of the on the keynotes. They show like a fridge and say, hey, what can I cook with this? And you take a picture and then it gives you a recipe, right? So it looked really cool.
Speaker 3:But do you need different hardware for that?
Speaker 1:Depends if you. This is how Siri should work, right.
Speaker 2:Yes, that's also what I was going to get to, because after I was trying to find again for this podcast, because on the on Twitter they were like oh, the keynote in 30 seconds If haven't watched it, because apparently it's crazy long keynote 30 minutes.
Speaker 1:You think it's like there is holding a device and what can I do? And then you need to wait.
Speaker 2:Oh, yeah, yeah, yeah, but well, yeah, I'll get. Ok. I'm sorry, but then, but then they were like oh, 30 seconds for a keynote, for the keynote, and then they kind of show all the stuff. So I went to Twitter again to try to find that so I can have some cheat sheet for this podcast. I couldn't find it, so I probably forgetting stuff now. And then I saw a lot of people saying oh, this is a glorified Siri, this is glorified Alexa he also mentions this there. Right, how, basically, alexa, siri, they should work like this.
Speaker 1:Yeah, and maybe those are two separate things. So, like I think Siri is horrible If the feeling becomes worse every month, get more and more frustrated and I think Apple needs to get this together. Do you use it? I less and less.
Speaker 3:So you did you.
Speaker 1:I never use less and less, but I think that will improve. Right, that should improve, and I think we will go to to a direction where you can do much more voice interaction with your phone, yeah, where you can say I'm going to go into a meeting and have a busy day, make sure to to mute all my notifications, but just if I have a delivery coming up. So if the doorbell rings, do make the letter notification pass for only for today, and that's something that is very easy to say out loud. It's super hard to configure, like I think for these type of things. I want to go skiing, and not that far off. There is not always snow. Make sure that if you notice that there is any snow, alert me a week in advance.
Speaker 1:I can just just stay talking about these types of things that I would like to know that today require a lot of conscious action.
Speaker 2:Yeah, I see what you're saying Right, but also is like, the thing that they also defend on the keynote is that sometimes their actions you want to do are simple things, but they require more than one app, and apps don't talk to each other, not today, no today.
Speaker 2:But I think, today we have a, we have a, we have a gap there, yeah, and that's, that's one thing that they also do, and that's that's actually the most interesting thing that I thought, because he was saying, like, if you were to at least what I understood is, if you were to design the iPhone today we have GPT models. We see the natural language is prevailing, right, it's very accessible for everyone. If you were to redesign today, how would it look like? And that's what they came up with.
Speaker 1:But, but maybe those are. Maybe it is something that for some people can live in isolation, like separate from your mobile phone as it is today. I don't think everybody will stop swiping tomorrow, right, like there's still a need for screens and then to read information, but like I will be very happy to have all of examples that I just say, if I have a star truck styled batch that I can do, that I can talk to, like rabbit, a little bit is right, yeah, the it's quite big right, it's quite a device.
Speaker 3:It's bigger than a than a smartphone.
Speaker 1:Yeah, it's like it's like a square. What do we think of the design? It's a quick key.
Speaker 2:But it's like it feels like a Game Boy to me. Yeah, it seems like they're targeting children. Yeah, even the name is like Teenage Angel, I'm a major fan yeah.
Speaker 1:I like it. I like the design. So the design well, actually, the La Croix company behind it is Teenage.
Speaker 3:Engineering.
Speaker 1:Huge fellow.
Speaker 3:Because it looks like for years. It's cool. Yeah, it looks nice, but I wouldn't bring it with me because it's just too big.
Speaker 2:But then it's a size thing, it's not a design thing.
Speaker 1:Yeah, I was more talking about the design. I'm very big fan of Teenage Engineering.
Speaker 2:I thought like I think it looks fun, but it's like if you're, you go to a bank and you talk to the CEO, you don't expect him to pull something like that out. You know what I'm saying? It's more like playful. I'm a CTO.
Speaker 1:Yes, cto is OK, but they have been so they have. They're very well known for their OP1, which is something completely different. So I think they only did the design for the rabbit at the hardware maybe, but the OP1 is really their product and the OP1 is like a quirky music device and I've been falling over it for the last five years and I always think it's too expensive. I'm not going to use it enough Intense for Bart's birthday, but so yeah, major fan of Teenage Engineering.
Speaker 2:Yeah, I mean I think it looks like, like it looks playful, but actually like. So, going a bit back to the R1.
Speaker 3:Oh yeah that's right.
Speaker 2:The other, like, so, they mentioned, like the communication across, like different applications, right? So if you say, hey, I want to go to Brazil next week, can you book a ticket, book a hotel, all these things? Ideally we'll do all these things for you, I agree, yeah, yeah.
Speaker 1:By any campaign. That's very nice and just confirm, you confirm the stuff. I think they played on something that is missing today and that is easier to achieve with GPT-like technologies. I agree, I agree, you're very smart on that.
Speaker 2:I agree, and they also mentioned, indeed, that they don't see this as replacing your phone, even though they do compare it to a phone. So it's a bit, you know, it's like.
Speaker 1:And the pricing. I forgot what it was, but when I looked at it I took out this is quite accessible for very that's what's it again?
Speaker 2:$200.
Speaker 3:No subscription, that's okay to play with Right. No subscription.
Speaker 1:No subscription. How do they have? Do they use mobile? So?
Speaker 2:they have some, like they have a server for sure, because they even said like oh, we don't store your data, but like you just give access to Spotify and you need a Spotify subscription, I guess, but they don't have a subscription at all. That's what is with the advertising. So it's $200. Interesting no subscription, which means it's cheaper than most smart devices.
Speaker 1:That's probably a bit to get the actual users Right.
Speaker 2:Yeah, so now it's the pre-order, so and this is well dollars, I think, in for Europe it will be a bit less actually, but then they ship here and I think it's only pre-order now, so we'll get here by Easter.
Speaker 2:And another cool thing that they showed so they showed on the keynote that I also want to circle back a bit if you have time that they also show you how you can train a new flow. But the way that they showed, at least on the, it's like it's almost like you do a screen share. So, for example, the example they give is to create images using mid-journey. So they go on the, on the rabbit UI, they put the discord, because the mid-journey is a discord server, like a chatbot that you send stuff and it gives images back. So you put the URL there and then it's like almost like the rabbit is watching your screen and then you type whatever you want to type and you wait for the image to be generated and then you say that's the image that I want and then from that point on, it knows how to interact with it. Like you can say, hey, generate an image with mid-journey about something else and you would generate it for you.
Speaker 2:So that's how you create new workflows, yeah, you know, and this is all using their hardware, all using their the hardware, the operating system, the device and all these things.
Speaker 2:So I thought this was actually very cool. And about the keynote I know that you mentioned that's a bit long and I do know what you're saying, like how Sometimes he's showing something and they don't edit much at all. So he's like you really wait for the guy. Sometimes they have the camera on his hand and you can even see it shaking a bit because he's just holding it on his hand. So sometimes it feels a bit like you're waiting for nothing. You didn't have to wait, I just wanted to see the result. But at the same time I feel like, looking at the Gemini announcement, I feel like this is giving me a feeling more like okay, this is closer to what I can expect. Because if there was too much editing, especially after Gemini, I would be like, if I don't know right, like you're running an LM on this device, like this thing that is bigger than an iPhone but it's still not a computer, right?
Speaker 3:So this was very authentic and raw.
Speaker 2:Well, yeah, it was still polished right, like you can still see, but to me it still looked like they were literally recording on the spot, which gave me a bit more confidence.
Speaker 3:Have you heard about it before? No, because I never heard about it and suddenly it came up. True, same for me, it was a real bomb.
Speaker 2:Yeah, it's crazy, but that's the thing, like what I described in the beginning. I was on Twitter, I saw one and I was like whatever. And then I saw it again and I was like whoa, let me take a look.
Speaker 3:Someone said this is the iPhone moment for AI hardware. Who said that? Someone on Twitter.
Speaker 2:Okay, I thought I'd say that by like.
Speaker 3:No, no, no, but it was a news agency as well, I think. Yeah, yeah, I recall.
Speaker 1:I was thinking was I so inspiring?
Speaker 2:You remembered.
Speaker 1:But you actually have. You have like this device that you clip on as well. This is also very much LLM based, which was a bit of a first few months ago, and I don't know where they are actually.
Speaker 3:I don't know. I know that when it doesn't, have a screen.
Speaker 1:It was really just purely audio. It's a bit the same but a bit the same use cases.
Speaker 3:It's becoming more of a discussion point. I feel like you can see more and more about AI hardware, but what it really entails it's still difficult for me to grasp.
Speaker 2:They also mentioned, like you mentioned, the thing, the clip. They also mentioned the Ray-Ban from Meta, yeah Right, which is another thing that's me Like. I heard about it in like an announcement, but I don't, I don't hear as hype, you know, I don't see. Oh yeah, I want to get it. I don't see, you know, it's a bit, it's tougher, tougher because it was a CES right two days ago or something. The Meta glasses no CES.
Speaker 3:Like it's a yearly event for hardware announcements in the US. Sure, I don't know.
Speaker 2:But yeah, this is me. This kind of came out of the blue house also.
Speaker 3:So will you buy it, bart.
Speaker 1:The rabbit. No, you will not. Not for now. I was tempted. I don't think I'm going to use it much at this stage. It's cool for us new, it's gadget factor, but I totally agree. Yeah, but I was also like I need to buy the OP1 before I buy this one.
Speaker 2:OP1 is the teenage engineering thing.
Speaker 1:If I'm going to throw money away for some gadgets stuff, then it's going to be that before I do the rabbit R1.
Speaker 2:Yeah, but the rabbit R1, I mean to me is like $200. To me it's cheap.
Speaker 1:Oh, you're like that, I'm butch.
Speaker 3:He has a ring and a watch, exactly.
Speaker 2:Now we have the iPhone in the pocket, the rabbit R1 thing you know just be like you can buy socks, but the gadget and the money stopped at the knee. No, I was thinking about it and I was tempted. I'm going to buy still. The verdict is still allowed as well. Right, Maybe after the first.
Speaker 1:But so next month you are going to be sitting here with R1?.
Speaker 2:This is still the pre-order, so they're only going to deliver in Easter in the US, so it could be even after that.
Speaker 1:So I think, because the design comes from teenage engineering, it will only be worth more in the future.
Speaker 2:You think so? I think so Just because it's the design has nothing to do with the product.
Speaker 1:No, no, no, no product. I think it will be very quick. This will go very fast. Just looked it up the pin. I meant it's actually called the AI pin by Humane. You will see a lot of these type of things popping up.
Speaker 2:I'm interested to see what the One thing that they do on the rabbit R1 as well. If you're having a call with a thing, they do real-time translation. Oh wow, so you can have a call in another language.
Speaker 1:Exactly, oh wow, there's a really Star Trek era.
Speaker 3:Yeah, actually you can do it with Google, do-it as well.
Speaker 1:Really, yeah, in a Google Meet Indeed Wow.
Speaker 3:So what I can do is you are speaking French and I am speaking Spanish. For example, do you speak Spanish Un poco? Oh, wow, wow. And so if I speak Spanish and you can see the subtitles in French, oh, the subtitles. So the transcription, the output, is not speech, it's text, indeed no okay, okay, but still it's quite.
Speaker 1:But with the rabbit it's real-time speech. I think it's speech.
Speaker 3:It's an impressive speech. That's impressive indeed and, to be honest, the Google Do-It feature Do-It in general is Very underwhelmed.
Speaker 1:It's so powerful right yeah, if you compare it to Chatchi Pity, it's really so powerful.
Speaker 3:If you compare it to Co-Pilot. We're doing it right now and it's a huge difference. Yeah, I think that's maybe a nice topic for the next episode, indeed, indeed.
Speaker 2:Yeah, but I think, unfortunately, it's time for what we have today. Maybe we have time, we both have a bit of time to play the yeah, sure, quote or no quote, hit it All right. So what is quote or no quote? Again, not the name I chose, but that's, I guess, what we're going with. The idea is, with the LLMs making advancements, can we fool people? The to-and-tests you know. So the idea is that we have a known character, a person, musician, whatever, and we have two fake quotes and one real quote. Sometimes it flips, but the idea is that we have a mix of fake and real quotes and then it's our job to know, to identify what is the real quote, what's not. So next time, a long time ago, mariam actually was the big winner, so I asked her would you be so kind to create a few quotes?
Speaker 1:And she did the problem with this challenge is that history shows.
Speaker 2:I'm very bad at this. Yes, bart, you're amazing at a lot of stuff. This is not one of them.
Speaker 1:But this really questions my future. So much is now AI generated in terms of content I really can't differentiate. Makes me ask some questions about myself. Yeah, I don't know if we have time to get into this, but I agree, all right.
Speaker 2:So who are we talking about today? We're talking about Master Shifu from Kung Fu.
Speaker 1:Panda, I know the character, but I never saw the Okay.
Speaker 2:And actually I have to say I know which one is real. So Mariam already told me. So I will not participate. But I'm very curious to see if she can fool one of you.
Speaker 1:So I need to think without any context. I need to think about what?
Speaker 2:would. She's the Master Shifu. It is Kung Fu Panda movie. It comes from Pixar, right? Is it Pixar Really? That's the first question Like this.
Speaker 1:But if it comes from something that is kids-oriented, right, yeah, it is kids-oriented. I've never saw it, so I need to think a bit about what would be normal tone of voice for that.
Speaker 3:I saw it many years ago. What do you think? Would it be easy, or is it quite straightforward or not? Oh, I'll let you decide. So one is real.
Speaker 2:So you're going to give two or three, three One is real, two are fake. You have to find the real one. Okay, ready, let's go. The true warrior finds peace not in victory, but in the heat of the battle. Heat of battle, actually, that's the.
Speaker 1:I can see a Kung Fu Panda saying that.
Speaker 2:But actually no. So the Master Shifu is not a panda. I don't know if it's relevant, but what is Master Shifu? Let me, is it an animal? It is, they're all animals.
Speaker 1:This is an important aspect.
Speaker 2:Yes, I can show you. This is the.
Speaker 1:Oh, it's a red panda.
Speaker 2:Yeah, red panda, exactly that's actually what it says right here, so it's panda, it's Kung Fu Panda. Yeah, that's true, that's what I said. Well, that's what it says, but it doesn't really look much like a red panda, right, anyways, but it's tiny, it's a tiny thing and it is a Kung Fu Master. Yes, okay, it's kind of like Yoda. Well, you're no Star Wars, right? Yeah, so it's like Yoda, but in the animal kingdom.
Speaker 2:Can you say the quote again A true warrior finds peace not in victory but in the heat of battle. I can see him saying that. Okay, me too. Oh yeah, he said that yesterday, actually. I know him. Yeah, before the. So this is the second one. Before the battle of the fist comes the battle of the mind.
Speaker 3:That's a bit far fetched. No, I can imagine this being a model. Yeah.
Speaker 2:Okay, can you repeat? There's one more, and then I'll repeat all of them afterwards, but last one is so important.
Speaker 1:I don't think that a Kung Fu Master would speak about the battle of the fist. I think it sounds a bit aggressive. I think a Kung Fu Master is more About the mind.
Speaker 3:Yeah, hmm, okay, this is too easy, merilo.
Speaker 2:Well, it was Mariam, so there's nothing. I take no credit, or yeah, it's the opposite of that.
Speaker 1:Can you do this in the tone of voice of Master Shifu?
Speaker 2:I don't know, let's try the third one. I don't know what his voice is like.
Speaker 1:Like a rap at the wood sound. Huh Can you give me an example. No, no, go ahead, go ahead, go ahead.
Speaker 2:Let's quote in a piece is not found. It's baked like a perfect dumpling.
Speaker 1:A dumpling. Adding a dumpling Is a dumpling in Japanese.
Speaker 2:You asking me why Bart.
Speaker 1:But why would you? But you have Japanese heritage, right.
Speaker 2:It's okay, I have blood. In Japanese, that is correct.
Speaker 1:Is a dumpling in Japanese. I don't know. That's Chinese? No, I think so. Yeah, I think Chinese. Yeah, and kung fu is Japanese, right? I don't think so, is it? I hope so. You look very, merilo. You should notice Like you're 50% Japanese. You need to look this up. Yeah, if you need to look this up, that's okay, it's Chinese.
Speaker 2:What is Chinese Kung fu? Kung fu yeah.
Speaker 1:Really.
Speaker 2:At least the dumplings, this one. We just looked up kung fu origin and it came up Chinese martial arts. So I think Okay.
Speaker 1:Yeah, everybody loves kung fu. There we go, that's us there we go All right.
Speaker 2:so one last time. First quote a true warrior finds peace not in victory but in the heat of battle. Second quote before the battle of the fist comes the battle of the mind. Third inner peace is not found. It's baked like a perfect dumpling.
Speaker 1:I can see animated characters saying that, though Indeed the last one.
Speaker 3:Well it is. It's fun for children and so on.
Speaker 2:Yeah, it's a child's thing.
Speaker 1:And there are two real and one fake, one real to fake, one real to fake. I'm going to say the dumpling one is real.
Speaker 2:Okay, final answer. Final answer, just trying to get into parts heads.
Speaker 3:This is the correct one. Okay, you repeat it one more time. Which one All?
Speaker 2:of them. Yeah, a true warrior finds peace not in victory but in the heat of battle. This one, the battle. Battle of the. Before the battle of the fist comes the battle of the mind. That's two. Inner peace is not found, it's baked like a perfect dumpling. That's three.
Speaker 3:It's a bit boring if I go for three, two, so I will go for two.
Speaker 2:For two. Before the battle of the fist comes the battle of the mind. Final answer no.
Speaker 3:No, I'm going to go for three, two I want to win this.
Speaker 2:Okay, so three and you're both Wrong. Can you do the Out? You know what's even more disappointing than you would have been right. The second one is correct. The second one is the real one. I still don't know how to sound fight fits in any context, but okay. So yeah, the second one is before the battle of the fist comes, the battle of the mind. So Mariam is really the big winner. So congrats to Mariam. I'll see if she can keep the ball rolling for next time. But that's pretty much it for today. Anything else you want to say before we call it up?
Speaker 1:Thanks everyone for listening.
Speaker 2:Yes, thank you.
Speaker 1:Thanks, ben, for joining.
Speaker 2:Thanks, ben, for joining. Happy 2024. I hope to see you back soon. Sure, with checking the predictions about the GPT store, and I'll be with my rabbitR1. Yes, all right, let's do it. Thanks y'all, bye.
Speaker 3:Bye.