AIForHumans94DeepSeekR1OpenAI

Gavin Purcell: [00:00:00] Well, it was a fun run. Pack it in, friends. DeepSeek R1 has officially destroyed the AI industry. The quote unquote 6 million AI David is taking down some of the largest titans of the space. Stock markets have trembled. Meta is running panicked. OpenAI is shifting up releases. I'm selling all of my organs online.

Gavin Purcell: Wait, wait, what? Where we're going, we're not going to need any organs, Gavin. It's the future. It's AI. It's China. This whole thing was supposed to be sarcastic. There is a lot more to this story, including some breaking news that Deep Seek might've pilfered the data from OpenAI. Oh man, thankfully no one knows what pilfered means.

Gavin Purcell: Plus concerns are mounting over data privacy, data security, market manipulation, government propaganda. I am selling all of my insides. Is this the dawn of a new AI era or is this a total tech meltdown? Please buy my thyroid now. Good price, great quality, fast shipping. Plus China has [00:01:00] even more new models coming out, including another open source text model, an open source iPhone.

Gavin Purcell: Audio model. That is very good. And whatever this Octobot tentacle robot thing is. Amazon basics just launched a new glands line. I'm so cooked. That's right. This is AI for organs. Everyone,

Gavin Purcell: everybody deep seek is the story of the week. You heard it here first last week. If you were listening to our podcast, obviously you heard us talk about deep seek our one, we did not farm the story for engagement enough last week. Cause we were like, Hey, it's out. It's impressive. That's a thing. And then the earth.

Gavin Purcell: Yeah, it's so crazy. It's so funny to be on something like that and then realize, Oh, on three days later, suddenly the story gets picked up. So basically, you know, I'm sure everybody here has been following this. If you're listening to the show, DeepSeek R1 is a Chinese, um, reasoning model that came out from the company DeepSeek.

Gavin Purcell: And over the weekend, [00:02:00] Monday morning, it basically destroyed the stock market. And the real reason behind this is because people believe that, uh, the money that is being spent spent on AI, including chips from NVIDIA, including the money that OpenAI and Microsoft are spending on AI is not worth it, Kevin, because supposedly this Chinese company made a AI model that is as good as GPT 4.

Gavin Purcell: 0 or 0. 1, you know, OpenAI's reasoning model. For five to 6 million. What do you think about that? I think compared to the, uh, half a trillion dollars that some of these companies are raising for all that infrastructure that you mentioned, the server farms, the energy, the chips, the, yeah, it sounds pretty devastating.

Gavin Purcell: And then when you start to peel back the onion, Uh, you realize this ogre has layers, Gavin, that there's, there might be more to this story, uh, than just, Oh, it was just a 5 million dice roll by some amateurs overnight, and that is toppling the [00:03:00] AI industry. So, so yeah, let's, let's really quickly run through some of the very top level news around what's happened around this.

Gavin Purcell: And then we're going to get into why this matters. So first there's some breaking news just today that supposedly OpenAI and Microsoft are saying that possibly. DeepSeek or somebody connected to DeepSeek pulled a bunch of data from their API to train this model, which I know Kevin at this point seems like it's kind of a little sour grapesy, but obviously there's some stuff going on here.

Gavin Purcell: And then, of course, you have people reacting to this breaking news with the fact that OpenAI is it's a little ironic that OpenAI is asking people to be feel bad for them about their data being stolen. The analogy is that China is Luigi and open AI manages healthcare. I think like no one is shedding a tear for the massive AI company that swept everybody's data into their systems to build their amazing machines and profit.

Gavin Purcell: No one is, is playing the tiny violin for the fact that China might have breached their API in a way that's [00:04:00] against their terms of service. To distill their model. So level setting really quickly, deep seek R1. If you are really catching up for the first time here, it is a very capable AI model. It thinks or reasons through problems.

Gavin Purcell: It scores very, very well on benchmarks. They released it open source, which means we'll get to the privacy concerns later, but it does mean that you can run it locally. You can run it on your own devices in your own secured environment. And it is an impressive model. It works really well. And people are.

Gavin Purcell: are creating these fine tuned, distilled, even smaller, quantized versions of them, and they're running them on like little 30 raspberry pies. So it's a powerful intelligence model, because it is open source, is getting into very capable hands and getting all these versions made of it. So, so yeah, that it is real.

Gavin Purcell: It's not like fairy dust. It is a real thing. No. And I think that's why it threw the markets for so much. And also there's a couple other quick things, you know, Mark Andreessen called this AI spyware. Sputnik moment, meaning that the idea that this was the Russians [00:05:00] getting to space before us, which I kind of disagree with a little bit, but also meta has scrambled up Kevin for war rooms, according to the information to try to figure out how deep seek did this.

Gavin Purcell: So there's real science going on here that you think you have to make sure that you don't dismiss. It is. And not only is it a real science, it is at the top of the app store right now. So this has broken through in a way that we haven't seen an AI model breakthrough to the mainstream until like, you know, since chat GPT was released, I would say.

Gavin Purcell: Yeah. Yeah. I mean, I got a text from my mom. I've seen lots of people get texts from their moms talking about like, what is deep seek? What do I, what is this thing? And then, you know, there's a lot of people I've seen posts online. of students using it. And I think that's one thing we have to be aware of not underestimating that chat GPT was driven by a lot of students.

Gavin Purcell: And the fact that this model is free, whereas you have to pay for access to Oh one right now, at least that's a pretty big deal. Yeah. And that's going to change as well. Open AI has already, you know, Sam kind of congratulated [00:06:00] the team on an impressive model release and said that they've got better stuff.

Gavin Purcell: Don't worry, investors. And don't worry, consumers, we've got better stuff, but they're pulling their release schedules up. Right. Which might mean a little less safety training, a little less guard railing, but yeah, They're changing their plans in the wake of this as well. So it is a big deal. But to round out the, the Microsoft open AI of it all, Microsoft apparently detected that someone was hitting open AI's API in an unauthorized way, like months ago, and they're making the claim at least broadly right now, that that might be related to, to, uh, to the deep seek, uh, company now.

Gavin Purcell: When we look at like, did they really do this thing for five or six million? No, the final training run that created this model cost that much, but we don't know how much money went into it before then. This doesn't account for the money spent on the researchers and the infrastructure and everything else that goes into it.

Gavin Purcell: That was just sort of the final training run. Um, Do we believe that [00:07:00] it was done with the, like there are, we talk about it on the show all the time, but there are certain chip restrictions on China, right? Yeah, due to us, China is only allowed to have a certain number of chips at a certain, let's say, a level of power or quality.

Gavin Purcell: They managed to get this modeled in under those restrictions, according to them. Now, according to others, they're saying, no, that's probably not the case. That via Singapore. China was able to get a bunch more ships. So there's such a cloud of uncertainty and doubt over all of this. But you did say Gavin, that it's number one in the app store, which is sparking a lot of privacy concerns.

Gavin Purcell: Yeah. I mean, I think the, the privacy concerns are something that are real because people have had, uh, privacy concerns over Chinese. Models in general. Although the, the, the back, the back on that is that open AI itself has had privacy concerns and like what kind of data they can see and everything. Those are American privacy concerns.

Gavin Purcell: Gavin, those are farm to table, non GMO, Eagle screaming, electric guitar [00:08:00] playing, sawdust consuming, um, patriotic privacy concerns. We don't worry about that. That's, that's right. Of course, we don't worry about that. Rah, rah, America, keep moving here. I think, I think that there are a couple things that are really important to understand when it comes to the fact that this is out.

Gavin Purcell: One, I think this is showing people, uh, pretty close to state of the art, cutting edge AI right now for free. Also, people are seeing what reasoning models actually say and think, which I think is another interesting thing. When you actually ask them DeepSeek, what it wants a question. It'll show you thought chain of thought, direct chain of thought, which open AI doesn't, which again, we've mentioned here on the show, but I think that's surprising to a lot of people.

Gavin Purcell: Um, I think one thing that we should get into Kev a little bit is what it means to have a distilled model and why a distilled model. model can sometimes be better than a large model. Um, distillation is a process where you take a large model, sometimes referred to as a parent model, and then you get all the best stuff out of it and you kind of squeeze it [00:09:00] down into what's called a child model.

Gavin Purcell: And the child model asks the parent model stuff to learn from it. So the distilled model can actually be smaller and cheaper and much better ultimately than the parent model. And if you're a long time listener of AI for humans, you remember us talking about, uh, an AI researcher named Leopold something.

Gavin Purcell: I don't remember what Leopold's last name was a long time ago, where he talked about this, it was a very long essay about how he talked about, we were going to get to ASI and had a couple of different bumps along the way. Distilled models are a big part of that, right? Because distilling a model means you can take the original trained model and then kind of filter it down and down and down and make a much better model ultimately that is smaller and cheaper.

Gavin Purcell: And that is what deep seek R1 and deep seek V3 are. They are distilled models that are, you can run in a much cheaper way. And that, accelerates AI so much because it means that you can spend all of this other money on running inference and compute at different [00:10:00] places. Well, and so when you mentioned that, that reportedly meta has these four war rooms going on to figure out how, how deep seek did what they did.

Gavin Purcell: Um, It's not like they don't under, like DeepSeek released a paper with, with math. They show their work. They tell you how they did what they did. Now, this is Meta and others scrambling to go. How do we apply what they did to what we do? And other companies are doing that as well. And some of that is distillation.

Gavin Purcell: There is a company which we've talked about before on the show called Groq. G R O Q. Not to be confused with Elon's XAI grok. I'll digress there, but, uh, GROQ, uh, hashtag, not an ad. They've got basically these custom chips that crunch these models very, very fast and usually very, very cheap. And for example, they.

Gavin Purcell: Did a DeepSeek R1 distillation of LLAMA 70 billion. I'm going to break that down. DeepSeek R1 is this Chinese reasoning model that we're talking about. The LLAMA 70 billion model was an open [00:11:00] sourced meta large language model that they spent billions Tens of millions of dollars, probably billions of dollars ultimately to create, right?

Gavin Purcell: They scraped all the data, they created a giant model. So now Grok is taking the reasoning from deep seek and applying it to that 70 billion Lama parameter model, distilling that down and running it on their servers. And we're going to talk in a second about these use cases where you can give these incredibly complex problems or coding tasks or whatever.

Gavin Purcell: And this model. Crushes it. It just speeds through it and does what would take, you know, minutes of thinking with a different model or might take a human hours of thinking and it does it in seconds. Yeah, it's pretty crazy. And you know, one of the things that Satya Nadella, the head of Microsoft, who of course all of the heads of all the tech companies had to weigh in on this in some form or another.

Gavin Purcell: He tweeted Javon's paradox strikes again. As AI gets more efficient and accessible, we will see its use. Skyrocket, turning into commodity we just can't get enough of. So you may or may not have heard of this term Javon's [00:12:00] paradox. I hadn't heard about it before this week and you know, it's, well, it sounds like, you know, some guy in, in a backwater of Alabama who's like trying to decide if he's going to drink his moonshine or if he's going to go like kick it in the middle of the night with the crocodiles.

Gavin Purcell: It's actually a really interesting thing. Javon's paradox is the idea that you might think. That as things get cheaper, you would spend less money on those things. And in this case, the thing that would get cheaper is compute. But what Satya is saying here, which I think is probably true and again, kind of leans into the idea why we're going to speed up here rather than like have a major collapse.

Gavin Purcell: Is that as this compute gets cheaper and more dispersed, the use cases of it will skyrocket. So people will start using it for all sorts of things. And actually we're going to need a lot more compute and we're going to need a lot more data centers to run this stuff on. Now, again, Satya's talking his own bag.

Gavin Purcell: We all know this, that everybody who's got a invested interest in this makes sense, but I honestly do believe this, right? I [00:13:00] do believe that. Distillation and DeepSeek R1 are going to drive a massive, massive spike in AI investment, AI use, and all sorts of other stuff. And this is where we get the sort of transformative things that will really change our world.

Gavin Purcell: And Kevin, you know, I think going into this year, I was thinking, well, 2025 is going to be crazy. I mean, it's literally the end of January and I feel like we've already had like a major tonal shift in what's going on. I can't, I don't care. I can't expect what's going to happen, you know, once OpenAI's 03 comes out and then, you know, an 04, probably the summer or the fall, like this, is it going to be a massive year for AI?

Gavin Purcell: I feel like. Yeah. Uh, there's already been some like, uh, projects where they're using deep seek, for example, to, uh, improve the code of the project itself. Right. This is the promise of the, uh, the, the AI writing its own tools to make itself better and reportedly some of that stuff exists now, but to your point, Oh, three comes out while they, they use Oh three [00:14:00] to better itself.

Gavin Purcell: And then it helps train Oh four. And then they distill Oh four. And like the. It's going to be a wild year, I'll say that. I don't think it's slowing down. I agree with you there. And for everybody, I had so many DMs and texts about like NVIDIA stock or these other tech stocks, you know, crashing. And, and my response to all of them was like, listen, I don't, don't invest on the back of anything that I'm saying, but I don't.

Gavin Purcell: I don't think as impressive as DeepSeek is, and the reported numbers being impressive, I don't think that should cause a panic. To your point, if this stuff becomes like electricity, we're going to need a lot more power plants, right? We have video models, world models, reasoning models, physical models.

Gavin Purcell: Models. We're going to have robotics to power. We're going to want intelligence in every phone and every smart toaster and television. Like there's going to be plenty of need for servers to power all this. And even as things get more efficient, we're still going to need more to deliver it. Yeah, all of that is true.

Gavin Purcell: I think the one thing that we should [00:15:00] dive into very quickly here is that how this has become a geopolitical story, right? Because it's not just about the fact that this AI model is cheaper and better, but the fact that it's from China and there is a lot of people out there kind of waving their flag of like being scared, right?

Gavin Purcell: And there's a lot of people who for a long time in the geopolitical world have kind of held up China as this kind of evil other. And granted they, we don't know China's intentions a lot of ways. But people do believe that some of this, you know, is a psyop, right? There's many people out there who are, whether it's conspiratorial, conspiratorially thinking or not, they think that China is doing this and releasing this stuff now to try to take down American economy, to try to kind of take back some sort of elite in AI, and that they might be hiding some sense of how they did this.

Gavin Purcell: Gavin, it's time to play America's fastest growing game show craze, Who twatted that tweet? Do I Who tw Who don Who do Who tw Who [00:16:00] tweet Okay, give me the twat, give me the twat and we'll see what happens. The 5 million number is bogus! It is pushed by a Chinese hedge fund to slow investment in American AI startups, service their own shorts against American titans like NVIDIA, and hide the truth.

Gavin Purcell: Sanction evasion. America is a fertile bed for psyops like this because our media apparatus hates our technology companies, even though it's owned by them and wants to see President Trump fail. That is a serious tweet to hath been twatted. And who is the Twitter behind that twatted tweet, Gavin? I think it's a guy with a long hair and a goatee that's longer.

Gavin Purcell: And, and, you know, we love this guy because he's definitely out there doing interesting stuff. I like that. That's Palmer. That's Palmer. Lucky, right? Kevin. That's right. We're lucky. That is Palmer. Lucky. For those who don't know, former Oculus CEO, now, uh, CEO of Anderil. Big defense contractor, uh, [00:17:00] smart, smart human being by all accounts, very, very well versed in AI as well, uses it in their products.

Gavin Purcell: And I think has been pretty attuned to the space for years, but Gavin very clearly planting the flag in the, uh, the propaganda camp saying this is, this is pure China. Yeah, and you know, the, the, there's another big story that the U. S. Navy has already banned Deep Seek due to ethics and safety concerns.

Gavin Purcell: CNBC's reporting that. So you know that there's a lot of stuff going on. I mean, this is such a big deal that Trump responded to it. We get, we have a little response from what he had to say about, uh, the Deep Seek, uh, panic, let's call it. The release of deep seek AI from a Chinese company should be a wake up call for our industries that we need to be laser focused on competing to win because we have the greatest scientists in the world.

Gavin Purcell: Even Chinese leadership told me that. They said you have the most brilliant scientists in the world. Seattle and various places. Okay. So it's a massive wake up call and it's [00:18:00] also a good thing because it could save us money. It's always a fascinating experience to listen to that person talk about science stuff, but you know, here we are.

Gavin Purcell: Like, I think this is an interesting conversation point where it has become The topic of conversation for the President of the United States to respond to AI news. So clearly, Kevin, we are on a pathway to gold. I mean, here, I wanted to be a governor for a long time. I want to be a governor of some, like, I'm not, I don't need a big state.

Gavin Purcell: Like, give me, I don't know, like North Dakota. And then I want to be like Governor King. I can start here. I'm so excited. This is clearly the answer. I'm glad I was here for your origin story, Gavin. Uh, and not that we started this, uh, podcast to dive deep into politics, but they two, uh, fields are very intertwined right now, AI and, and politics.

Gavin Purcell: So, uh, on the heels of that, uh, president Trump announced the U S is going to place tariffs on all semiconductors and pharmaceuticals that are imported from Taiwan, which is, it could be a potentially very big [00:19:00] deal. Now, there are. Plenty of companies that are opening, uh, fabrication, uh, plants within the U S which would help end around that.

Gavin Purcell: And I think that is the intent behind these tariffs is to get people to start manufacturing here, but the speed with which we need to get chips and, uh, the already inflated costs. Have some people shooting flares in the sky saying this is actually going to do more harm than anything else, than any regulations ever would have on like AI safety, for example.

Gavin Purcell: So I don't know if you've formed an opinion on this and want to share it so that you can get death threats, Gavin, but I'd love to hear it. Here's a, here's a death threat opinion. I think that we really have to be cautious about Taiwan period because there's a very good possibility that Taiwan. Could change hands in the next, say, three to five years.

Gavin Purcell: And this is like a scary thing, right? Like in Taiwan is a place, obviously, that is a geopolitical hotbed. And I hope we did not just drag our YouTube algorithm by talking about it. But I think this is something that everybody needs to be aware of is that a vast majority [00:20:00] of chips are coming out of Taiwan.

Gavin Purcell: And I think that the faster that we can onboard them into American companies or companies somewhere else, in the same way that people, um, off loaded, uh, Um, clothing manufacturing or other things out of China when they started to see Chinese businesses start to get kind of pushed down a little bit by the government to places like Vietnam or, or different countries.

Gavin Purcell: I think this is a really important conversation. Obviously, you and I are not geopolitical experts, but I think it is something to track really closely. Uh, you know, I think the other thing, Kevin, to think about here is there's one specific American company who might be really happy that all this stuff is happening right now.

Gavin Purcell: Yeah, some are saying, uh, we, he's talking about, uh, Apple, specifically Timmy Apple. Timmy Cook. Who might look, yeah, who might look like a genius just for laying back in the cut to use some jazz parlance. It turns out What?! Laying back in the cut. Laying back in the cut, baby. You're going to have to do an explainer on that.

Gavin Purcell: Just give us a second. You can have pilfered and I'm having laying back in the [00:21:00] cut. You know what? Hold on a second. Hold on. I'm going to do it right now. We're going to Grok cloud. So Gavin, I'm going to console. grok. com hashtag, not an ad. And I'm going to use the deep seek R1 distilled llama 70 billion model.

Gavin Purcell: I'm going to say, um, explain to my co host Gavin what laying back in the cut means. With regard to jazz slash music, and I'm going to let it think in jazz music and jazz or music in general, quote, laying back in the cut refers to a rhythmic concept where a musician plays slightly behind the main beat, creating a relaxed, subtle feel.

Gavin Purcell: This technique adds to a sense of groove and tension as the musician is not strictly on the beat, but slightly after it. It's often used to add depth and nuance to the music rather than making it feel more natural and less rigid. So you're saying Tim, Tim Cook is back there with his stand up bass, playing his fingers.

Gavin Purcell: Yes, I got it. He's got a long filtered vape. It's not a cigarette at the end of it, but it's a long [00:22:00] filtered vape because it's more techie. Can somebody in our audience put Palmer Lucky's hair and goatee on Tim Cook and then put him behind a bass? Because that is the perfect image of Tim Cook laying behind the cut or whatever.

Gavin Purcell: One of those celebrities, if they made it, I would love to see Palmer's goat. The goat's goat on Timmy Cook, but the idea is Tim, yes, upright bass and beret, long filtered vape, laying back in the cut while the industry is marching to the beat. He's just slightly behind doing his thing nice and chill. And the reason any of this matters is that it turns out This particular model, this DeepSeek R1, runs pretty well on Apple Silicon.

Gavin Purcell: And Apple Silicon is something that is way more powerful than a lot of people maybe have given it credit for, because most people think of Apple as like, you know, MacBooks, but they've been steadily upgrading their Silicon in a big way. And, you know, this is another kind of possible issue with NVIDIA, that NVIDIA was really the place to drive a lot of this AI work, and their [00:23:00] CUDA back end specifically.

Gavin Purcell: Apple now might have a leg up in this space, which is a pretty big deal. Yeah. And when you look at, you know, the reason people, I think also saw Apple as a little bit behind it, because most of the AI tools were written for Nvidia GPUs, specifically for their software and hardware stack. Um, this is a shout out to Alex Chima over on X, who had a breakdown of the chips that can run DeepSeek on the market now, and basically he broke down Nvidia versus AMD versus Apple.

Gavin Purcell: Right. Which is weird to see again, Apple chips in that, in that race, Apple's M two ultra, which was released in 2023. It's the processor I have in my laptop is four times more cost efficient per unit of memory than AMD and 12 times more cost efficient than an NVIDIA H 100. Now. This doesn't mean the industry is shifting towards this, but it does bode.

Gavin Purcell: Well, I keep seeing all these posts, Gavin of people like stacking Mac minis on top of each other and basically saying, Hey, [00:24:00] look for six to 8, 000. I just built me a corn pewter that can run deep seek R one, the full model. Yeah, right here off the shelf. I've basically got open AI in my server closet. That is why yeah, which is again going back to those distilled models and being able to do stuff smaller than better makes a big difference.

Gavin Purcell: Okay, we should probably wrap up all of this conversation with a couple other big things that came out of China. Um, but before we do that, this is your chance to subscribe. That's right. We are going to let you subscribe to our podcast. This is, this is your only chance. This is your only chance. And Gavin is very generously allowing you to subscribe.

Gavin Purcell: This is a chance to subscribe to our podcast. It's a one week offer. It shuts down next week in which we will open it up again. Uh, please subscribe to our YouTube channel right here if you're watching this. If you're on audio, please, uh, make sure you leave us a review on Apple, uh, iTunes. We got a couple new, uh, reviews.

Gavin Purcell: That was very nice of everybody. But this is your chance to subscribe, like, share. We only get [00:25:00] seen by other people. If you share us because we are an independent podcast, Kevin, and I make this ourselves. That is the one thing you can do for us. You're also welcome to drop some money in our tip jar on Patreon, but you don't have to.

Gavin Purcell: The most important thing is like, subscribe and share for us, y'all. Thank you so much. Thank you very much. Yeah, really, it is. The only way we grow is off the back of your kindness and engagement. So thank you for that. It's true. It's true. OK, Kev, these other stories that are pretty important, again, all are coming out of China.

Gavin Purcell: Something that kind of shocked me yesterday is there's another new Chinese LLM. There's a new Quen model, which has been out already, but they updated it as the 2. 5 dash max, a large MOE LLM pre trained on massive data and post trained on curated SFT and RLHF recipes. All of that is kind of like, you know, a tech speak for a, it's a big model, not just a small model.

Gavin Purcell: And it actually is comparable, if not outpacing. Uh, V3, DeepSeek's last model, not the reasoning engine, but [00:26:00] it's outpacing V3, which is also outpacing GPT 40. So Kev, just coming off the heels of all this DeepSeek stuff, there's another Chinese model that kind of is surprisingly powerful. And I think it does make you start to think like somehow they kind of timed all this.

Gavin Purcell: I know it's also Chinese New Year's time. So I know that maybe it might be around that, but like, man, China is just coming hard right now with this AI stuff. Oh, sorry, Gavin. I just, uh, late breaking. I'm getting, uh, uh, what sounds like Suno music from like, uh, six to eight months ago, maybe a year ago, another Chinese open source model has dropped.

Gavin Purcell: This is Y U E S one seven B, which is not sexy or fun, but it is. It's actually pretty capable. This is an open source music creation model, again, coming out of China. And uh, it's the, the, the Chinese word means music and happiness. Yeah, do you want to play a little bit of it? I think you're right. It doesn't sound perfect, but let's play a little bit of that to take a listen to it.[00:27:00]

Gavin Purcell: I mean, it is funny. It sounds Oh, is that Tim Cook laying back in the cut? He's still laying back in the cut, flipping his hair back and stroking his goatee with one hand and doing the bass with the other. So what if his goatee is the upright bass? What if his goat goes so long to the His goatee? Tim Cook laying back in the cut, playing his own Go.

Gavin Purcell: His go Team bass. Yeah. Goes down to the floor and he's playing it like the bass. Oh my God. We've been unwashed by, I don't know what is going on here. So anyway, anybody can make that. Please put that in our discord, please. This is an open source music model, so I think the one thing we haven't really seen come out of China yet.

Gavin Purcell: Is a music model that kind of took the world by storm. Now, granted, it does feel like a little bit behind Suno. [00:28:00] Obviously it's not as feature rich as Suno is. And Suno is kind of laying in a lot of features and doing a lot of things. But Kevin, the thing about China, and we're going to talk about this a little bit later when I get into Kling stuff, uh, in their video model.

Gavin Purcell: China does not really respect IP or any of that stuff. So I am, I'm kind of convinced at some point that China will make a very good music generator, if not for the fact that they just, they'll train on everything. Yeah. A hundred percent. Look, that the sample that I played from the hugging face model host was a very little poppy kind of little ditty, but this thing can do all genres of music.

Gavin Purcell: It can do English lyrics. Yeah. Uh, make no mistake, it will get better. But the difference again, uh, is, is, was twofold. Yes. They'll train on everything. That's fine. Uh, two, it is open source. So people will take this and run with this and improve this, and this will worm its way into every product ever. We know we talk about 11 labs all the time on the show.

Gavin Purcell: They've [00:29:00] got great, great voice models, voice assistants. They have a sound effects generator. They don't have a music competitor yet. And what traditionally 11 labs does is take the best in class, open source stuff, bring it in house and then improve on it as quickly as possible. There are going to be dozens of music services overnight that are going to probably fine tune and custom train on their own genre and make no mistake, they'll get better and better and better.

Gavin Purcell: Counterpoint counterpoint. I object. I object to the judge. The counterpoint is in this instance. The reason why a lot of these don't exist is because of the risk of lawsuits from the music industry. And I think that honestly, you know, what's interesting, the Suno has continued to grow and build. And obviously they got a giant check.

Gavin Purcell: One company that we were talking a lot about early on that I haven't seen get another check. Isudio, right? So I think there is a lot more risk involved with music, but so I'm not entirely sure 11 labs wants to be in that business because it's really hard to litigate and understand the [00:30:00] music industry rights and they are crazy.

Gavin Purcell: The music industry is insane. I don't know if you guys saw the story, but Spotify put the biggest payout ever to music in the music industry, which is good for the music industry. I'm not saying that they should get paid, but like they are going to get their money. So I think the AI rights. of music is even more difficult than almost anything else.

Gavin Purcell: I think you misund I think you are grossly misunderstanding Eleven Labs. Their favorite thing is litigation. I know this. It's actually not voice cloning. Is that true? They're gonna leap They love lawsuits. So they're leap No, that's that's not at all. Please sponsor us, Eleven Labs. Please sponsor us. Kevin just got himself in trouble at Eleven Labs.

Gavin Purcell: Just being a dum dum. Welcome to the uh, red list. I do work with them on the same thing. I'm being a dumb, dumb. They don't like it. That's fine. Hey, Gavin, let's talk about video models. How about that? Now, later on, we're going to talk about cling AI's new feature. That is a Chinese video model that I played around with, but right now, Kevin, a company we talked about last week, mini max is dropping a new feature.

Gavin Purcell: It is only available for their early testers right now, but it [00:31:00] is very cool. It is called the director feature. And what this is, is it gives you a lot more control over the camera and it follows those controls really well. If you look at the video they released, you can see how it kind of like you can directly tell the camera what to do and it follows it.

Gavin Purcell: It reduces randomness in the movements. It enhances the control of accuracy of everything. And the videos that you see in here, like a push in shot, you see a zoom out, all these things. Keep the consistency of the characters, and so I've already seen people do some really interesting stuff. I do want to shout out Wuxia Rocks, uh, I'm sorry if I'm pronouncing that wrong, who, who used it to create the evil dead shot, which is fascinating to see, like, it's, it can be fast, it goes through the, uh, the, uh, The forest very quickly, quickly is a shot through a forest with like some smoker fog.

Gavin Purcell: Yeah. There's like someone on an A TV, but like the camera is racing with this kind of exaggerated, uh, field of view. It's a beautiful shot. Yeah, it's absolutely right. And there's [00:32:00] another great thread from Alex Petrosky who does a lot of really good stuff. He has a shot of what looks like almost like an Orson Wells sort of film where it's black and white and you see.

Gavin Purcell: Very clearly, uh, characters interacting with each other. It just goes more to the fact of like, we're gonna get so much better control over these AI models. And again, Kev, what's interesting is this is all coming out of China right now. So this is like, you know, China is stepping up their game in a big way.

Gavin Purcell: Yeah, I look, we said prompt to Hollywood within five years, we might have been wrong, but for the first time wrong, because we overestimated, we're seeing commercials now that are fully AI driven. And I think that almost counts. But like character consistency, camera control, lighting, all these directorial tools, even AI editing, this is all really coming online.

Gavin Purcell: And I'm, I don't know if you're noticing this, Gavin, but I'm feeling And I know that it's, it's my own filtered bubble of bias, but whether it's LinkedIn or X or on threads, I'm seeing a lot of, it used to [00:33:00] be a lot of never AI. And now I'm seeing like a, Hey, I experimented with this and I'm starting to make that, Hey, I'm tooling around with this.

Gavin Purcell: And I made this very quickly. And now I want to actually go make it as a. Quote unquote, real project. Are you feeling that, that tide shifting? Yeah, I definitely am. And you know, I'm speaking at NatP next week. Uh, I was invited to speak on a panel where I'm putting together a six minute presentation. So a quick thing, and my, my topic is going to be about speed to IP and how that AI is going to affect how we create IP and how IP will come out of the AI space more often.

Gavin Purcell: And again, I, you know, we've said this on the show a bajillion times, but I'm definitely seeing that. I think smart people are striving, are starting to understand the idea that, Hey, I could be an individual creator or I could work with a team of three and five and make something really cool. And, and the whole point of what, when I say speed to IP, one of the things, if you've been following the entertainment industry for a long time, like all Obviously every major movie is now based on IP.

Gavin Purcell: There is no movie that exists out there that is not based on IP and the wild robot's a great [00:34:00] example. People thought, Oh, maybe the wild robot is this cool new movie that's based on original IP. It's not, it's a book, right? So I think what's going to be interesting about this time in our creative lives is that people are going to be able to create IP much faster and then see if that IP is sticky and then double down on it.

Gavin Purcell: So I honestly think small teams. Creative people who have done work in the creative space before are going to succeed in ways that you can't even imagine. And larger Hollywood might be in trouble, but the small teams who can iterate fast and create things are gonna do very well. Gavin, I couldn't agree more.

Gavin Purcell: How do we make sure that every possible butt is in that Nat P seat? You know, I will share, I will share on the show handle, uh, on X and a couple other places, uh, where it's going to be. And if you can come, that's great. If you're attending, obviously you have to be an attendant to come, but I would love to say hi to you.

Gavin Purcell: If you're a fan of the show, pop in and say hello to me for sure. Uh, Kev, guess what? It is time. For one of our favorite sections of the [00:35:00] show, where we look at some of the stuff that people have done with AI and we shout them out and we celebrate them, it's time for AI, see what you did there. Sometimes you're scrolling without a care.

Gavin Purcell: Then suddenly you stop caring.

Gavin Purcell: Kev, there's a very great AI creator out there named AI Warper, who is the person responsible for the Travis Scott meme that kind of took the AI world by storm. They have dived into trying to create an AI influencer and it's changed now, but I really did appreciate the fact that he started out by trying to make a female version of Asmongold, the famous video game streamer.

Gavin Purcell: And there's just a very funny clip he posted where like, The female Asmongold is responding to the male Asmongold, and to me, this is like the perfect use case of AI. It's just a dumb, really interesting way to look at using [00:36:00] AI influencers to kind of comment on other influencers. Now, and part of it is a showcase of like, LipDub.

Gavin Purcell: AI technology. Again, hashtag not an ad, but have you explored? They're claiming that LipDub. AI is the highest quality lip sync. And trusted by industry leaders like HBO is in there. Um, I I've seen lip dub AI before. I mean, I think all these lip dubs, as you know, are, are tricky, right? Because like none of them are perfect.

Gavin Purcell: And it really is one of the things that's going to continue to kind of. Be a thorn in the side of AI. This is honestly why I think act one is one of the best things you can use. But in that instance, you're really having to do the acting yourself. I just haven't seen a perfect AI lip dub yet tool. I was going to say someone who, uh, loves love is blind.

Gavin Purcell: Okay. Spoiler. I do. I do. I'm still amazed that when I fire up Netflix, that there aren't like lip dub versions of the love is blind, Germany and love is blind, Brazil and love is blind, whatever, right? They've had the [00:37:00] subtitles and they have the, even the, Like the, the dubbed voiceovers, but where's just the press the button.

Gavin Purcell: And it uses AI to make the mouths move with the voice of the person. We know that tech exists. I I'm sure it would be expensive, but like what Netflix is raising prices yet again. What are they doing? I don't need more WWE, no shade to the superstars. I want to watch love is blind, Germany. But in English, I wonder if that is a big deal.

Gavin Purcell: Like, I just, I, I don't, you know, I don't think about that. Because I do watch, like, honestly, when I watch, um, foreign shows, like Squid Game, I watch with the original language in the subtitles because I can't stand dubs. But maybe there would be a better sense if the lips match the voices. But in general, I don't know.

Gavin Purcell: I, I, maybe they don't do it because there's not enough of a demand for it. Well, I'm demanding it now, dammit. It's too hard for me to It's too hard for me to read the TV while I'm scrolling TikTok during my favorite shows. Okay? I can't fragment my attention like the average viewer. Now, one thing I would watch in any language or [00:38:00] dubbed in any subtitle, Gavin, is Indiana Jones.

Gavin Purcell: Filmed in front of a live studio audience. This is a, uh, a runway project. This was on the AI video subreddit. Um, you had seen it and it made you giggle and I had seen it. And then we both completely neglected to shout it out. It is just. Scenes from Indiana Jones, but they added a live studio audience. It is exactly what it sounds like.

Gavin Purcell: And for some reason it tickles me. I think it's great. I want to watch other films in front of a live audience. Like again, shouting out citizen Kane or like, what would Scarface look like in front of a live audience? Imagine the Scarface scene where he's sitting at the desk and people are behind him, clapping and so excited for him.

Gavin Purcell: Like this might be a new genre. Give me someone scrambling for the key to unlock the thing so they can get the bomb out of their stomach or whatever. Give me saw. With the live studio audience, but they did. They just, they took scenes from the movie and extended them with runway. Um, and you could see the original versus the runway stuff, but watching uncanny audience members hunched over and clapping and giggling.

Gavin Purcell: If you're getting the audio only version of the show, go check [00:39:00] it out. The link will be in the show notes. It is worth the click. Yeah. Shout out to vague V A I G U E man. Who's the person that uploaded this and created it on Reddit. Um, so Kevin, the other thing I really want to shout out, we talked about Grokka earlier.

Gavin Purcell: There is a, uh, two X user that we've, we've loved forever. We've shouted him on the show many times. His name is at underscore a K H a L I Q. It's not the greatest name in the world, but he's always shared for like the last two years, some really interesting stuff. And he put together a hugging face space, uh, where you can try R1 plus Grok.

Gavin Purcell: And my experience with this is kind of mind blowing. So when you, and this is specifically about coding small apps. So if you go to the link that we have in the show notes, you can try this, what it is basically, it's just a box that says like, what, what, what do you want to create? You type in what you want to create and depending on if the space is up, because it might be getting hammered in again in time.

Gavin Purcell: You can create a program in once it goes through [00:40:00] 10 seconds and that is insane. I, I'm going to get to this later in my cling thing, but I will say one of the biggest difference makers of any AI tool is the speed in which you get results because so much about this is how do I iterate on the thing I got?

Gavin Purcell: Is it broken? Okay. If it's broken, how do I go back and fix it? And if you are waiting. 10 minutes between generations, it is a disaster. You lose track of where you're going, something else in your life comes up. But this implementation of Grok plus R1 feels like the absolute future. Have you, did you have a chance to play with it at all?

Gavin Purcell: Yeah, you sent me the link and I immediately fired it up and got in there. And, and even with everybody else, Poking and prodding at it probably at the same time that I was, I was amazed at just how quick it was, uh, and it reminds me of the, the llama distillation version that I was talking about earlier in the show that I was playing with.

Gavin Purcell: Like, um, the speed is a differentiator for sure. You want, obviously you want the highest quality result you can get from an LLM, especially if you're coding or something, you want it to be [00:41:00] clean and workable, but then beyond that. I would say speed even before cost for me is the most important thing because to your point, yeah, you issue a command.

Gavin Purcell: I do a lot of coding with cursor now. I will. Tell it what I need fixed or what I want to work on. And then I will sit there and I will wait and I will sometimes get lost in the sauce. And if I want one tiny correction, I'm like, Oh, I got to go back and wait again. Having things happen at this speed. It's almost like we are going to have this, this gives you a glimpse of that promised future of like, Hey, it's dinner time and we want to split the bill.

Gavin Purcell: Does anybody want to download for that? No, I'll just whisper this need into my phone and it will code the app for me. And now we'll all have. So, yeah, I did. I use this. It's the, the actually looks like it might be down right now, but this gives you an idea of how insane this will be going forward. And again, it's, it's.

Gavin Purcell: It's code at the speed of thought, right? And to your point, like not just splitting the bill, but like I used it to create something that we did back in the day, which was like a debate [00:42:00] between two characters. And I said, Hey, I'd love you to create an app where I can input two characters and then a question for them to debate about and have that happen.

Gavin Purcell: And, you know, the actual result came back like in 20 seconds. And then it wasn't perfect. It didn't do it exactly in the voices. But you can see how once you can get results like that, this fast in a coding platform. Holy cow. You could make stuff so quickly. And I just think we're going to see interactive stuff like skyrocket in an interesting way.

Gavin Purcell: I just want to interact with tentacles, Gavin, and I feel like you do. That's what you've wanted for your whole life. That's your like, I feel like Hint AI is not advancing fast enough. And then out of nowhere, Gavin, I am gifted. Uh, an amazing video on X, um, and it's from, it's from Zanshi Wang is the source YouTube channel.

Gavin Purcell: Uh, Spirobs is the, the, the channel S P I R O B S, we'll put a link in the shout outs, but this is logarithmic spiral shaped robots for versatile grasping across [00:43:00] scales. And when you look at the video, you see what looks like a 3d printed. Octopi ish arm performing tasks that are insane. I'm talking like, grasping of delicate objects, catching balls in real time motion, whipping, it can like, knock things off of things, there's even, uh, like, curling and uncurling, and then there's a version of it that's on a quadcopter with multiple tentacles, and the quadcopter will descend and pick something up or delicately drop it off, and it is, uh, Wild to see this bio inspired grasping arm flail about, but then have so much utility and they have robots that sprint and dance and do all sorts of things now.

Gavin Purcell: But when you think about these, these robots, they're trying to make articulated fingers and this, that, the other, but there's a world where we've got. humanoid tentacle bots running around that have a little thing that can extend from their belly or from their head or whatever and grasp a feather an ant or whatever it [00:44:00] just it's so weird looking but so inspiring well that's what i was going to say here is that we obviously have been building humanoid robots to exist in the world that we create and everybody's always like why do you need to build a robot that acts and works like a human well the argument is that we build the world for us to be shaped as humans so that we need robots that can do this and one of the things interesting about this is like Maybe this is what the robots would design themselves to look like, right?

Gavin Purcell: There are things in nature that are much more useful than arms. And in this case, what's interesting is it's clear that, like, it's very good at grabbing things. Like, even the fact that it could grab a tissue out of a box, you know? There are things in nature that are more useful than arms. Like, put that on the box!

Gavin Purcell: You're so right like watching it like the multi tentacle one grab things. It's like it looks like an octopus. OK, so I just texted you one that goes to the time code like I would love for you to describe in real time what you're seeing because it is exactly what you're talking about. OK, so I'm seeing a manipulation.

Gavin Purcell: Yeah, so this is what you're looking at is the tentacle. Dropping a ball into [00:45:00] very specific locations and it can get into different places, right? And that's really interesting. It's able to interact with something different than a hand could because it's so flexible. Now I'm sure there are better things for hands, but like this could be where, you know, they always talk about AI.

Gavin Purcell: Like we're thinking about itself, maybe more than for humans. Like when AI start programming themselves, they may be like, Hey, a tentacle is way better than a hand because I can do X, Y, and Z with it. Japan has run those numbers. And the answer is yes, they've known it for decades, buddy. All right. Okay. So Kev, we're going to get into a little bit about what we did with AI this week.

Gavin Purcell: I played around with Kling's new feature elements, 1. 6 elements. And people are out there talking about how impressive it is. And Kling is a very good model and it is another Chinese model. And there's a lot of interesting things you can do with this. Now elements is basically the implementation in Kling of something we saw come out a while ago and we played around with it.

Gavin Purcell: Where you can take like the coat of a person, you can take a [00:46:00] person's picture and you can take another thing like a car or a desk and kind of have them all drive the scene, right? So that's a really interesting thing is if you can take multiple things, say you could have a subject of a person and you could have a different outfit to put on them and then you could have a location to put them in.

Gavin Purcell: That gives you much greater control over the AI video. Except for, I will tell you, I had a relatively miserable experience about this, and Kling, which I paid for, I got a one year subscription for like 60 at Christmas, I have to say I've been relatively disappointed in over time, and it's not because the results are terrible, generally they are okay, we'll get into what the results of this specifically were.

Gavin Purcell: But Kevin, each Kling result takes 10 minutes to get. And this goes to talk about what I was talking about before with Grok is that when something takes 10 minutes, you do not get the sense of progression. You do not get the sense of that worked or this doesn't work. And it becomes impossible in my mind to do creative work because you can't, I [00:47:00] don't know if your brain works this way, but like, I can't like put something in and then take 10 minutes off and then put something in and take 10 minutes off.

Gavin Purcell: It just, I will get distracted and it felt. Unfortunate to me that this is the way we're working right now in AI video. So okay, assuming speed gets better over time as it should, walk me through the process though. Was it as easy as dropping in some images and then describing the shot that you wanted? So it is very easy.

Gavin Purcell: The Kling interface basically has four boxes for you to add as many pictures as you want. So if you look at some of the first results, I love the picture of Mark Andreessen with his giant pointy head. That's something that I've been, you know, I think it's hilarious in some ways. So the other benefit of Kling is that you can use real people for better or for worse.

Gavin Purcell: So I took a picture of Mark Andreessen with his pointy head. He was holding his hands together. I took a picture of a space station. I took a picture of a space suit and I said, You know, put the man on a space station floating in zero gravity. And if you look at the first [00:48:00] Marc Andreessen one, that was the initial result.

Gavin Purcell: So it got Marc Andreessen pretty well, right? That's pretty close to Marc Andreessen. And he's sitting in the same outfit. He's wearing not a space suit on a space station with a beautiful, uh, space going by him. So this result totally usable, but not really what I wanted. Um, I then. Did it again, the second result, and I even tried to go a little bit longer, and it kind of did the exact same thing, like it missed the spacesuit, didn't put him in zero G.

Gavin Purcell: It did have his hands freeze together for some reason in this one for a longer period, but I don't know. You can see these, you know, they're pretty good results. And if you had thought maybe that's what I asked for, you might be amazed, right? Yeah, I was admittedly a little more impressed until you told me the ingredients that you gave it and what the prompt was.

Gavin Purcell: It was sort of like, well, it kinda put the person in the thing, but it seems like it ignored just as much as it got correct. Yeah. But these aren't bad. Some of the other renders you sent, though, feel Yeah. So [00:49:00] I, then I decided, you know, one of the memes has been going around about deep seek is a Jane Yang.

Gavin Purcell: If you follow it, Silicon Valley, Jen Yang is a character from that a who's from China. I decided to try to put him in a similar sort of area. And I put him into this space with a picture of him kind of replacing the Mark Andreessen photo. And I said, a man is smoking on a space station. And what's interesting about this is that either it doesn't understand smoking very well, or smoking is something he doesn't get.

Gavin Purcell: Uh, but the smoking part looks bad. But again, you know, Jin Yang is there in this first one, like the one where he's sitting down. That is Jin Yang with the beautiful, like, you know, planet behind him on a space station. It got that. The second one, uh, I added the element of having a panther next to him in a tuxedo.

Gavin Purcell: So it got the tuxedo, it got the panther. But then the face isn't really gen yang anymore, and the smoking still isn't perfect. It is better than the first time. Uh, in fact he put, I like how he puts the two fingers against it. [00:50:00] Then I did one more re render with a different tuxedo picture, but unfortunately I left the other face on the tuxedo.

Gavin Purcell: Um, the model in this one, I didn't want this to be Jen Yang. Oh, sorry. I wanted this to be Jen Yang and it used the model's face. Uh, from the tuxedo and put a tiger next to him, not a panther. And you didn't know that until almost a half hour later when you came back to your digital slot machine and found out it wasn't a win.

Gavin Purcell: And that is the big thing is like how many people would actually play a slot machine if you had to wait 10 minutes between each spin? It'd probably be better for human beings and terrible for Vegas. But but you don't have the sense of iteration. And I think that's something we talked a lot about how R1 in Grok is speedy.

Gavin Purcell: I cannot wait to see something where I can say something to a model and again, I would love to just say this and not have to type it and have it be interacting with me in real time. I mean, you remember what we saw CREA and all these other image models do where you can do image generation real time.

Gavin Purcell: That's what I'm waiting for with [00:51:00] video and runway turbo is great. It's pretty quick. And Sora turbo is relatively quick, but like, I I really think this is going to change the way we make stuff when this becomes two to five second outputs. And I think that's probably coming. It's just a matter of scale.

Gavin Purcell: When I can, uh, whisper a thought, give it elements or otherwise, and it gives me a wall of thumbnails. That I could then go, Oh, let me see what that one looks like. I click, it quickly generates a turbo version of the video, like a previs. And I can look at it and go, Oh, that's great, except make the camera move a little bit faster.

Gavin Purcell: And make sure the bow tie is blue instead of green. And then I click the button and then it goes, Okay, and spits it off to the higher resolution model. Everything else is trashed, we move on. That is the speed and I think the workflow with which people are going to start building full movies, full scenes and everything.

Gavin Purcell: I think it's, Pretty incredible. It reminds me of, um, you know, like you mentioned that you want to just use your voice and talk to it. I went out of my way to make cursor, use my voice this weekend for a coding project. Um, and [00:52:00] it's bizarre to me that it's just not more intelligently integrated into all the things, even like if I write something, I want to click a button and quickly say, Actually change this to this.

Gavin Purcell: It's so much faster than typing and the, the whisper by any other name, which is a tool that will take your speech and transcribe it is almost real time now. So I want voice control over all of the things in all of the apps. Um, but I, I used, uh, Mac OS is built in. voice accessibility features to allow me to talk into the cursor code box, basically.

Gavin Purcell: So I could click, I could say, start listening. My laptop starts listening. I dictate what I want, uh, and then I tell it to go to sleep basically. And then I press send on cursor, but, um, I sent you a project that I was jamming on. And over the course of about three days, I was able to, um, using some off the shelf prompts.

Gavin Purcell: Um, I used GPT 01, their reasoning model, which could have been. Uh, you know, DeepSeek, the truth be told right now and would have saved me a couple bucks, but I [00:53:00] used, uh, uh, OpenAI's 01 model. I told it to think through this thing that I needed to build, basically this app, and give me atomic level directions for a junior engineer.

Gavin Purcell: That junior engineer, by the way, was you. It was, well, no, it was Anthropics Claude. It was Solary. Right? Yeah, exactly. I'm just the middleman. I'm a layer of unnecessary, just middleman. I'm the, I'm the octopus at the switchboard in the cartoon that's just plugging one thing into the other. Eventually I will not be needed for this.

Gavin Purcell: But so I, O1 came up with the plan. I put the plan into a file. I told Cursa to read the file and go through, ask clarifying questions. It did. And then I answered those questions and ultimately I just started churning out an app that was Off the shelf AI components to do real time web search and generate imagery and interact with the user using their voice and the speed with which I was able to get a proof of concept together.

Gavin Purcell: Again, mind blowing to me as someone who doesn't do that as their core competency really makes me reflect on like what the [00:54:00] heads of state are saying in terms of coding as a, as a, as a career in 2027. I think there's definitely going to be a need for Engineers that are competent that can plug in things and make sure they're like they're secure and that systems are intelligently integrated.

Gavin Purcell: But the getting the actual coding work done, I feel like agents are going to really crush it. Yeah, I couldn't agree more. I think coding is one of those places where, because it's technical, because you're using a lot of the system, same sort of plugins and the same sort of stuff, there will always be a place for super creative coders, right?

Gavin Purcell: There will always be a place for people who are innovating in code or trying to do stuff. But pretty, pretty much, you know, a lot of integrated code, a lot of code that people are doing is not the most creative in the world. So I think you're right. Um, okay, that's it for today. We will see everybody next week.

Gavin Purcell: Uh, hopefully we will not have another giant stock market crash based on AI. If we do. Come here. We'll always fill you in on what it is. We will see y'all next time. Subscribe to our newsletter, AI for humans. Dot show. Bye. That's right. Definitely. [00:55:00] Bye bye.