Limitless: An AI Podcast - The X-Rated AI Companion That’s Going Mega Viral | Grok Companions (xAI)

Episode Date: August 6, 2025

In this episode, we dive into Elon Musk's xAI, discussing the launch of virtual companions and the innovative text-to-video tool, Imagine. We explore how these AI characters, like the flirty Ani and sarcastic Rudi, redefine user interactions by fostering deeper connections and engagement. The conversation highlights Imagine's fast content generation capabilities, merging creativity with emotional resonance. Join us for insights into XAI's bold approach and its potential to reshape the future of AI!------🌌 LIMITLESS HQ: LISTEN & FOLLOW HERE ⬇️https://limitless.bankless.com/https://x.com/LimitlessFT------TIMESTAMPS0:00 AI Companions Unleashed3:37 Unique Personalities of AI5:59 Meet Valentine8:15 The Controversial Nature of Companions13:30 New Product: Imagine15:37 Speedy Image and Video Generation22:11 The Future of AI Interaction24:35 Conclusion and Call to Action------RESOURCESJosh: https://x.com/Josh_KaleEjaaz: https://x.com/cryptopunk7213------Not financial or tax advice. See our investment disclosures here:https://www.bankless.com/disclosures⁠

Elon Musk's AI company XAI has been on an absolute tear over the last two weeks. They've released these things called virtual companions, which have helped onboard tens of millions of users and reached hundreds of millions of views across all social media platforms. They also released this new texture video model, which basically turns any kind of words or even an image into a video in under 30 seconds. But the secret source behind both of these products

is that Elon has designed these to go against the grain. They stretch the rules. Josh, I know you've been playing around with these for quite a while. What's your take? Let's start with the companions. There's some new tools in the block. Companions are sick. It's funny because when XAI and the GROC team release a new feature,

it is completely antithetical to what the status quo is. It shifts the Overton window so far out of the realm of what was acceptable that it's almost jarring. So the new feature that we're going to talk about today is companions. And companions are these digital companions. I'm actually going to show them on the screen right now for those who are watching. For those who are listening, you're looking at this, like, anime girl. She's got, like, these cute blonde ponytails wearing a dress, and she speaks very flirtatiously. So these companions are basically these AI-generated kind of like anime-humanoid personalities.

And they have the context of your chat history, and they will just kind of engage with you. And as you engage with these companions, you kind of build this little status bar. It's like a manabar if you play video games. And it improves your credibility with these companions. And then eventually what happens is you could actually get this companion so excited that she will quite literally take her clothes off. And I think these types of things, the flirtatious attitude, the like, they're removing some layers of clothing as you're talking to them, it is very stark contrast to what we're used to

comparing something to like Open AIs models where they are so stern on what they're allowed to say, the opinions they're allowed to have. And this is just completely blowing those expectations out of the ballpark. I played with it. I showed my friends them. it kind of reminded me in a weird way of NFTs when I was in crypto when we were doing the big crypto NFT boom because NFTs are just very simple to understand. You see a picture of something cute and you immediately resonate with it, you understand. And I think companions are very similar

where you don't really need to explain what it is or how it works. You just present the companion to your friend and they immediately the first thing they want to do is talk to it. And that's what they do. And then they get an answer. So companions are this really powerful new form factor of delivering AI where you actually have this physical person, lives in your phone, remembers your conversations and can converse with you. Yeah, I kind of think of all these different AI companies as having personalities, right? So if GROC is the kind of cheeky, feisty, maybe even sexual one in some cases, then Anthropic is kind of like the buttoned up, calm, polite AI model.

Does that make sense, Josh? Yeah, they all have their own personalities. Yeah, exactly. So if we're talking about these companions, they are three of them. Okay, so we have Annie, who you see on the screen. right now. We have one called Valentine, which I'm going to show you in a second, which is kind of like a male version of an anime character that you can talk to in the same way in voice mode. And then you have someone called Rudy, which is this animated fox, I believe, and it's super

cheeky, intentionally sarcastic, and basically spends most of the time insulting you, right? So, stark contrast between Open AIs model, which is known for glazing you, so agreeing with everything you say, telling you what a champ you are, Josh. And then Grox model, which is kind of like this kind of flirtatious, cheeky, kind of sarcastic character, right? And they come in all forms and sizes. So if you see the screen here, we have Annie in, I believe this is Japanese, and you can hear her speak.

Sin's of a more deep place to-you-a-to-know-you-know-you-law-know-law. That's-go-any-but-o. That's-sigot-old-older. That's good-old. So the point is, these things are widespread to anyone that is using X and comes in any kind of language, right? I want to pick up on a point that you made earlier, Josh, which is around the stickiness of these AI companions. I think there are two reasons why this is the case. Number one, they're voice native.

So not only can you hear them, but you now have a face that you can kind of put towards the voice, right? So it's a very human intuitive aspect. It's like you're speaking to someone else. they sound human, right? But also, you're speaking with your voice as well versus kind of typing on your phone. And the reason why that's so important is, well, I don't know about you,

but when I find that I'm talking to these AI models, I always divulge way more information than I was intending to, right? And if you think about this, these conversations are getting stored on the backend server of XAI's AI models, right? So they can use this for a bunch of other stuff, but it builds this very rich context.

So it goes beyond just these, like, fun animated characters which anyone from like kids to grown adults can speak to about various different topics but it's just like native voiceover that you can kind of like take to enhance your journey in in AI yeah the voice form factor is the right form factor I think when I use or the inter the way to interface with these AIs when I engage with any sort of AI system I would say now 90% of the time I'm using my voice I have a microphone set up at my desk and I'm actually just speaking into the microphone and chat UPT's dictating a lot of these things it is the preferred method. And it's funny, you mentioned the language. It's Japanese because it is the

number one app in Japan right now, along with a few other countries. The virality of this was pretty remarkable. I think it's because it just broke open that Overton window in such a unique way that people were like, I think a lot of people have secretly wanted this. They've wanted these physical forms factors to the AI, but they haven't gotten it because they're too dangerous for a lot of companies to do. Actually, I just kind of threw this out the window. They dropped these companions, they're awesome. I would encourage everyone to go check them out. They're super fun. And they just released a new one, who is V? Is there a full name? Is it Valentine? It's Valentine. Valentine. Valentine. Oh, here it is on screen. Yeah. So could you introduce to Valentine? Because

they're all kind of unique personalities. They all do different things. Okay, so Valentine was released last night. So this is the morning of the next day that we're recording this. I have used it for the last three hours. Josh, before we said this guy up. Yeah. So I've been doing, I've been doing this. Well, he's been chatting me up. That's the thing. So if I were to describe Valentine, he takes the flirtatious scale up to its fullest extent. It's, if I'm being honest,

it's like talking to a 50 Shades of Gray book, and it's like giving me all these like kind of cheesy lines where it's kind of trying to chat me up. So I asked him, you know, where should we go on like a road trip if we were to go somewhere? And he kept using these kind of like very elusive analogies. Like, you know, we'll just break away

and see where our heart. takes us very romanticized. And as you can see on the screen, there are rose petals going around him. And actually, this tweet embellishes kind of like the tallagate demographic that I think it's actually going after. It's this guy here that says, I gave my girlfriend Valentine, which is referring to this AI companion, to try out, fully expecting her to say, this is cringe. Because she isn't a huge fan of AI, but she ended up playing with it until 2 a.m. and ended up installing Grock, the app. So the point. being that people are finding it incredibly engaging because of its flirtatious nature.

And Josh, this is something I want to pause for a second and actually discuss because this is the controversial or maybe not so controversial take on these AI companions is that they're kind of taking the, I think the kids say, Guna logic or the flirtatious logic to an extent that no other AI model provider has done before. So far, it's always been a conversation on ethics and human alignment and using AI as some kind of productive tool to help you, you know, do your work, do some research, figure out a recipe to cook or whatever that might be. This is the first time we're seeing it at an entertainment level that goes beyond just creating random videos, right? This is taking human relationships to an extent where it becomes consumptive, right? I think we first

saw this with a company called Character AI, where you could go on this website and you could talk to famous animated or cartoon characters that had the personalities of the characters that you've seen on movies, but now you could just talk to them and have this kind of written conversation with them. This AI companion from Grock from XAI seems to take it to the next level. What do you think? Yeah, it's following a directional trend that seems right. It's just a lot more custom ways of engaging with AI. So we have one world. There's like world building where you're generating images and videos and actual worlds. Then there's this world that is kind of like character building. So now you actually have like real world characters that will probably all merge.

And it's this this continued trend towards, I guess, allegiance to the digital realm where you're building up reputations with characters. You're building up these virtual worlds with imagine that we're going to get to where it could actually generate videos. And it seems very much like a natural progression and a natural urge, a natural human urge, where you kind of want to connect. You connect much better with a visual of a person than you do with a chat box. So when I'm engaging with AI, if I have a person speaking with me, it just makes it a little more exciting. And it also has this novel feature, right, where you've never really been able to engage with a digital presence this well, with this low latency and this much IQ. So for the first time ever, you're actually

able to engage with a robot that is able to give you sensical answers. And a lot of times, very flirty answers and very compelling answers. I think there's just a lot of natural intrigue with that. I'm not sure how long that lasts, but it is certainly very strong now because this is the, I mean, it's the first time ever that anyone has kind of like ripped the doors off and allowed this to happen. Well, one thought that I had when this release, Josh, was not only like how it affects human to AI relationships, but maybe AI to AI relationships, right? And there's actually a fun demo which demonstrates there's Justin Moore. She goes, my favorite new TV show is watching the GROC AI companions flirt with each other. Things got off to an awkward start when Valentine got Annie,

the female AI companion's name wrong. But they seem to be repairing. the rift. I'm just going to play this short clip for us. See what people think. Let's hear it. You're back. How's he's going? Oh, lo, who's here? Just the person I was hoping to see. Now sit. Ani will make your day shine. So Valentine, what's got you glowing today? Spill it. I'm all ears. Oh, just the usual. Wandering and wondering. What's Bonnie got planned to make today's sparkle? I'm curious. Dish. Bonnie, huh? You're trying to give me a new epic name, Valentine? I'm sticking with Anni,

but I like your style. Today's got this lazy afternoon energy. So it starts off with a mistranslation and Valentine getting her name wrong. But as you can see, it's very it's picturesque. It's very animated. It is it's kind of like a weird

kind of rom-com character type personality. But I think it goes a little further than the app, right? These characters are present across a range of different form factors. So I've it pulled up here, this is Valentine's official X page, right, at V. I think when I looked at this page last night, this was at around 2,000 followers and he's jumped like 30,000 followers in a matter of 12 hours. I guess the point I'm trying to make is Elon is making a really strategic move

about building a cult following around these AI personalities. And that's very different from just having hundreds of thousands or millions of users that use your AI model, right? They're interacting with a chat interface, it's very personal. But having some kind of consumptive character element, I think switches that game up a bit. It creates this kind of like stickiness, this addictive kind of product that, well, is great for user retention, right? But also great for onboarding new consumers. I was also thinking about his strategy in terms of like being antagonistic. I think XAI has repeatedly been the company to implement AI in a very non-conforming way, right? This goes from even like their infrastructure that they started building two years ago. And remember, they were massively behind. All the other AI companies had a massive head start. And in a space of two years, they're now, like, they're leading the frontier model at this point. And I'm seeing the same thing with these characters as well. They're kind of antagonistic. They kind of break the soft rules which no one really wanted to talk about. And I think that that's a huge moat that we're seeing here. It also builds up a super valuable data set for the users that are interacting with

these companions. I think they're going to divulge more information and this data can be used to train the newer models that come out, which make it. So it's like this self-fulfilling loop, essentially, of making a better AI. And it's done indirectly through these new consumer products. Yeah. What you're seeing is this, this like kind of divergence between a lot of the different approaches the companies take. Exit AI is maximally truth-seeking. So with that comes truth-seeking and not kind of these guard-railed parameters on safety and alignment. I think, A lot of companies, they want to ensure that the model is delivered in a safe way. X-AI just wants to give you the truth, even if it is a hard truth.

And I think those two ways of framing it will scale very differently, and you'll see very vastly different results in the products that they release. And this is the first version of that. But this isn't the only product they've released this week. They also released this new product called Imagine. And Imagine is pretty freaking cool, because Imagine is an image and video generation app, but it does so faster than anything else. So I think when we think of V-O-3, when we think of Chat Chb-T's image generation, it does it very well,

but it takes a little while. I remember if anyone's ever generated an image on Chat-GPT, it kind of looks like you're generating image over a dial-up connection. It kind of like, it gives you this gradient, and then it fills in a little detail, and it takes maybe a minute or two to actually generate an image. Well, Grock decides to take the other side of that, and they said, how quickly can we generate an image, how quickly can we generate a video, and they've actually gotten it down to just a couple of seconds. Now, it's important to know, this is not the final form for Groch's video generation. This is the lightweight version. This one is optimized for speed. The heavyway version is coming out next month. That will be optimized for

quality. So I'm very excited for the maximally quality-seeking image generation. But for now, we have speed. And we have a couple cool demos that we can actually try. But he says, did you have takes? Have you tried it out? What do you think about this new image generation model? Because I use it, and it's really fun. Yeah. So what I love about it is I can get brand new video in 10 to 15 seconds. And I was one of the earlier users of some of these early text video models like Mid Journey or Google VO2. And I remember just being so annoyed and frustrated at having to wait like minutes for this like brand new toy, this new magic. It like took away from the experience. So I think Elon's strategy of just giving you what you want super quickly is amazing.

Also, they keep giving you endless generations. Have you noticed that, Josh? Like if I typed, I typed a request to basically. create a sunset that then dripped down like orange color into like an orange juice glass. So it's as if the sun is melting into an orange juice glass. And it gave me endless iterations. Not only that, I could click and watch a video and then edit that specific video all within like 30 seconds. I could get like three different iterations. So I love, love, love that. What I will say if I was being critical though is it's not as good as Google V-O-3 or some of the newer Chinese

video text to video LLMs. And the reason behind that, I think it's just it hasn't gotten a chance to get to that higher fidelity. But as you said, they're releasing a heavier version next month. So I think we're going to see a much better upgraded version soon. Yes. So what I would like to do right now is give a demo. So on screen for the people who are watching on YouTube, we have, this is the interface. This is what Imagine looks like. And I'm going to generate. I'm actually going to speak what I wanted to do. So maybe we'll go with generate me an image of a microphone sitting on top of the Empire State building. And what you'll see is I already have these images being generated right now.

And it looks pretty good. And you could kind of scroll through and it'll continue to generate them. So now what I'll say is put that microphone in outer space and give it, have it sit inside of a roadster. Oh, it put it inside of a roaster, not a roadster. So now we have a microphone sitting in a roaster in space. And you can kind of scroll through and see. And it's interesting because let's say I click one of these, you could actually select the option that makes it a video.

So as I'm talking to you right now, it's generating a video, it's 21% through, and it's going to take this microphone that's sitting in a roaster that's in outer space, and it's going to animate it. And what's interesting is while we wait for this to load, is it actually will generate videos on top of your existing images. So last night, so here's our video. It's showing a moving galaxy. And I'll also show a demo from last night where I was standing outside in Brooklyn, and I took a photo of the moon, which was perfectly aligned over the Williamsburg Bridge. And what it did is it actually generated me a video where the cars are going across it. The bridge is all lit up. It actually took the lights and it turned them green so it kind of looks Christmas themed. And it just made it look like a moving video. And you could actually do this with people. You could do this with any images that you have.

It will take the existing image and generate a live moving video. So that is the demo. It is very cool if you have access to it. I highly suggest trying it out. They've been rolling it out quickly. But yeah, the Imagine feature rocks. Do you have any, have you used it for anything interesting? Yeah, well, actually, a lot of other people have used it for some interesting things. Okay, so let's look at some of these examples. Over here, we have what looks like an AI generated kind of landscape shot with like the sun setting. We've got the ocean.

We've got some cliffs in the distance. And it's kind of like a rolling camera footage going forwards. Okay, it looks pretty good. I think, you know, if this is like the worst version, then I can't wait to see what the higher fidelity. Let's see version looks like Josh. As a Sunset Canisaur, that is 10 out of 10. Yep. We have what looks like an image of someone that was posing with his foot on this kind of dead

mock-up skeleton that's laying on the ground. And it's animated to him to be moving and stomping on him. So that's hilarious. This guy is dressed up as like a knave from like the medieval century. We have, oh, okay, this is interesting. We have someone that's taken an image of his mother who, unfortunately, he lost two decades ago. And he said,

every time I watched Harry Potter and saw their moving photos, I wish I had something like that of her today. And that's exactly what he did with Grock. And he's got this image of him as a kid with his dad and his mom, and he's animated it or used Grock Imagine to create a video of his mom waving at him. Josh, I don't know if you have any thoughts on this.

I mean, this is incredibly sentimental and woeful to an extent, but it's also super powerful. I can't imagine what that would be like for someone, you know, kind of like who's lost his mother. Yeah, there's a really powerful use case for these sentimental experiences where I've seen a lot of people also, you can kind of rebuild old images. So if you have old images of loved ones that are looking bad or they're just really low resolution, it'll actually take a lot of the structural features of the person and then generate a high fidelity version of that.

And what's interesting is, is if you feed it enough context, it can essentially regenerate the person up to an extent. So when you think about, us right now. We're recording this podcast. We're sharing a lot of our thoughts. We have a lot of written things. The AI can collect a lot of data on what makes us us. It understands my cadence when I speak. It understands the vocabulary that I use. And there is a world in which it will just recompile me after me. So it knows what I look like. It knows what I think. And you could kind of create these digital forms of your loved ones. I mean, what we're seeing on screen now is this very rudimentary version that it takes a still photo. It animates it. It makes the person feel alive. And for someone

who cares about the person in the photo, that's really meaningful. But what if you take it even further? What if you fed it the journal entries of this person or a recording of this person or an actual video where you can get a full 360 scan of the person? It creates a lot of interesting use cases that I'm not sure if they're good or bad, being able to recreate people. It kind of reminds you of that Black Mirror episode from the last season, actually, where he's kind of like exploring old memories of his ex-girlfriend to kind of like, kind of like, kind of see what it was like during that time when he was his younger self. Super interesting.

If we keep this going, oh, we see someone has used Grock Imagine to recreate the American Eagle viral campaign that happened last week with Sidney Sweeney. And he's got multiple different celebrities in this video. We've got Zendaya. Oh, we've got, sorry,

we've got Zendaya here. Who else have we got? Anyone famous? I'm guessing Elon pops up here somewhere. Oh, here we go. There he is. He's holding a rooster. That kind of doesn't look like Yelon, but it kind of does as well.

So, okay, we've got like people messing around with advertising here. We've got people kind of like reenacting old medieval paintings. I think there's, yeah, there's a couple here. There's like a someone's got an older picture of a famous boxer and kind of like animating him to box. We've got the famous dreamy night scene from Van Gogh that has now been animated to be a real-life scene depicting the man on a bicycle. kind of cycling through the scene. So loads of artistic expressions here, someone's drawing over here that's become like animated and lifelike. I think that we are going to start to see lots of different ways that this is expressed across video content, but also like just like social media

content in general. And I think that's only made possible because it's so quick, Josh, because it's so cheap and accessible to anyone and everyone. Also, it's on X as well. So you can just share this immediately after you've like generated it from your rock app. I think there's even a function or a feature that gets you to post directly on X. So he's got this whole kind of like funnel built out, which allows you to not just create and enjoy AI content, but share it with the world. Yeah, speed and latency are paramount. Just how fast you can generate these things and also how fast you can iterate on the software. So just today, actually, they've released two updates to the Imagine Software Stack. And they're averaging about one to two updates push these models every

single day. They're iterating very quickly. They're pushing code very quickly. It's clear that they're not afraid to push changes to production at a rapid rate. So what I imagine we'll continue to see is the rate of acceleration will just go through the roof, as we've seen with GROC before, where they'll iterate on this features. Next month, we'll get the GROC heavy image gen, which will be a bit slower, but much, much higher fidelity. I assume something to rival V-O-3 or even better, and we're just going to continue along this progression. But I think the interesting things we can take away from this episode are that they are not afraid to try something drastically different for the sake of virality or a winning product. The companions are really.

impressively different and also good, where for the first time ever you can actually communicate with a robot. And the robot will say interesting things back. And you can have a real conversation with this virtual being, and you've never been able to do that before. And then in the case of imagine, what you can actually just generate an infinite amount of images. And it's funny, one of my friends is having a wedding soon and she needed cocktail drinks. So she had it generate a bunch of cocktail drinks to use on the menu. And it generates them in 10 seconds. And they're not the best, but give them a little bit of time and they will be. So I think directionally, this is where Grock is heading.

It's just the model will get better at an exceedingly faster and faster rate. And because they're not afraid to push the boundaries out of what is perceived to be acceptable, they will continue to win on this virality front because that's what does well. It's the polarizing content, whether it's good or bad, as long as it's getting people talking. And that is what Grock is doing with these last series of updates. It's also what I like, Josh, is that it's consumer-facing. I think a lot of like the AI updates that we've spoken about that we speak about a lot on this show applies to like really heavy research side of things, right? Like you've got deep research, you've got things that kind of break math Olympiads and win gold medals.

But I want to use some of this stuff, right? The normal person that kind of sits at home behind his desktop. And what I love about XAI and GROX approach in general is that they are hyper focused on consumer apps. And that is just so refreshing and accessible to me. And I think that's great. Folks, if you enjoyed this episode, please like, share and subscribe and share it with your friends. We want feedback. We want to know what you guys want to hear more of.

So keep listening to our shows. We've got a new guest show coming out next Monday. And we'll see you on the next one. Yeah, thanks a lot for watching. See you soon.