Senators Say "Shut AI Down", Mistral Forage, Pentagon AI, Google AI

**Jayden Schafer** (0:00)
Welcome to the podcast, I'm your host, Jayden Schafer. Today on the show, we're talking about some big news in the AI space. Number one, Mistral is betting on a build-your-own AI. They're taking on OpenAI, Anthropic and Enterprise. Gary Tan has a cloud code set up, which is getting a lot of people triggered, and a lot of people love it. The Pentagon is developing an alternative to Anthropic. Some new reports have shown. And BuzzFeed right now is developing what is called quote unquote AI Slop Apps. And they're trying to do this to get new revenue. Google has a personal intelligence feature that is expanding to all US users, and OpenAI is expanding their government footprint. And SeedDance, the AI video generated company coming out of ByteDance, they actually are getting some serious heat from Congress, which is calling on them to shut down over basically a lack of guardrails. So we're getting into all of this on the podcast today. We're gonna do a deep dive on the SeedDance story in particular. Before we get into all of that, I wanted to mention some huge news for my startup, which is aibox.ai. We have just launched video on our platform. So in the past, you know that you got access to over 40 of the top AI models all in one place. You could kind of chat with them in a playground. We had image, text and audio, and we have officially now added video. So we have two models from ByteDance. They're SeedDance. We have the three different models from Google, VO2, 3fast and 3 We have two different models from OpenAI, including Sora 2 and Sora 2 Pro. And we have Pixiverse V5 from Pixiverse. So there's a ton of amazing video models that are now on Open or now on AIBox.AI. If you don't have a subscription already, you can get it for $8.99 a month. Super cheap, way cheaper than any of the other platforms. You get access to 78 different AI models. Guys, in the past, you've heard me say 40 models a million times. I'm actually kind of stuck in that, but we keep adding new models. And I just counted right now, we're at 78 new models on AI Box. Everything from text, image, audio, video, more announcements coming, tons of new features, subscriptions are going crazy, and we actually doubled revenue last month, which is amazing. But if you want to check it out, it's linked in the description, aibox.ai. Check out all of the latest new video models. It's only $8.99 a month, and we have 20% off if you get an annual plan. So I'll leave a link. Let's get into everything happening in the news today. So the first star I want to cover is that Google is expanding their personal intelligence feature. They're doing this to all of their US users. So basically, they're pushing Gemini a lot deeper into the Google ecosystem. And I mean, I thought it was pretty embedded in there. I've actually been impressed because I've been calling on them to do this for like over a year now. But basically, they're going to let it personalize responses using connected data from your Gmail, from your Google Photos. I think what's interesting here is it's not just kind of this premium user experience. Google is going to widen distribution. They're gonna put this capability inside of AI mode in search and on the Gemini app and Gemini in Chrome. So it's off by default, but the product direction is very clear if you want to get that enabled. I actually appreciate this is off by default because I do think while it's great and cool and personal intelligence is awesome, I don't think everyone's going to appreciate having, you know, their Gmail and their Google Photos automatically opted into AI personalization, let's just say. I think the next phase of consumer AI is not just about better models, but it's about better context. We know, right, if you give Chet GPT better context on what you're asking it to do, aka like if you're trying to get it to write you an article, copy and paste an example of that article or a specific type of document or file, copy and paste an example, that context makes the output way better. I think Google understands this, and the company right now that plugs into your email, your files, your browsing, your history, your photos, they are going to have a massive advantage. This is something that, you know, Chet GPT had a big advantage because people kind of used them at the beginning, so it had all of that history, and it could personalize answers based off of their past Chet GPT history. Google has way more history and data on all of us, for better or for worse, and I think this is going to give Google a really big competitive advantage, and they have, you know, so much distribution. Now they have this, you know, basically they have a huge AI moat on just data that they have, so this is going to be interesting. The next story I want to cover is that the Pentagon is reportedly building alternatives to Anthropic. This should come as no surprise as the big spat between Anthropic and the Pentagon rolled out earlier in the last couple weeks. There was a obviously very public breakdown in the relationship, and the Defense Department is now developing replacements rather than assuming Anthropic is going to remain part of its stack. I think that is following the broader clash over military use, surveillance and autonomous weapons. And at the same time, I did see that Congress is pushing to create red lines on what AI could and couldn't be used for. I, for one, think that if we're going to have this stuff regulated, Congress is the place for it to happen. If we're going to make rules about what AI should or shouldn't be used for, Congress is the place for that to happen. I don't love the feeling that, let's say, we're going to go and use a model that is perhaps Anthropic, right? Totally cool American model. They make rules that the military doesn't like about what they can and can't do according to the terms of service. What happens if Anthropic gets, as a private company, gets purchased by, let's say, a Chinese investor or a Russian investor? And all of a sudden, they can kind of manipulate the terms of service of the company, which is being directly used by the military. So, I mean, I'm sure the government would block any sort of acquisition by that. But it just really, I just don't like the companies themselves creating the terms that the government has to follow through. So I appreciate that Congress is looking into this. And I hope that Anthropic doesn't, you know, doesn't get wrecked too hard financially from that decision. It'll be interesting to see what happens in the future. I know a lot of consumers are kind of supporting the company because they agreed with some of the reasons why Anthropic had a fallout with the Pentagon. So we'll see. I do like Anthropic. I do like Claude, one of my preferred models for really high quality outputs. But I just don't think it's a good precedent for American tech companies to basically make the rules that I think Congress should on, you know, military use or what the government should be doing. All right. The next thing I want to cover is that Mistral has just launched what they're calling Mistral Forage. This is an NVIDIA GTC. It's one of the most important enterprise AI product moves I think that I've seen today. Forage is basically designed to let enterprises and governments build custom models. So it's going to be trained on their own data, not just kind of like lightly fine-tuned. I think with all of this, Mistral really is betting that companies want a lot more control. They want a lot more customization. They want a lot less dependencies on someone else's kind of black box roadmap. And so I think right now, Mistral is trying not just to win the consumer chatbot race head on, but going after a part of the market where control, governance, kind of like multilingual performance, long-term ownership, a lot of that matters more than just raw consumer mind share. And I think that's important because if Mistral is really on track to surpass a billion dollars in annual recurring revenue this year, then it's definitely going to be a serious enterprise challenge to kind of having the open AI or anthropic kind of duopoly narrative that we see right now. And of course, I think we just, like Mistral isn't going to win, especially not in the United States as a French company. So perhaps in France it's the most popular, but in the United States, this is not the most popular chat bot for consumers. So they really got to focus on going the enterprise route. We've seen this from other players like Cohere. The next bit of tech drama I have for you is a culture battle for Gary Tan's Claude Code setup. It went viral, it had almost 20,000 Github stars, had 2,200 forks after he basically open sourced his workflow. A lot of supporters say that like, hey, this is super legit. Haters were saying it's basically just an overhyped prompt package. Something I thought else was interesting. BuzzFeed is launching a wave of AI-powered content apps. They're trying to basically unlock new revenue streams. So they're doing things like quizzes, content generation, a lot of personalized media experiences driven by AI. This is kind of interesting for me right now because media companies are not debating right now whether to use AI. They're basically just experimenting really aggressively with it in order to survive. So many of these media companies are launching lawsuits against OpenAI and Anthropic saying, look, you guys scraped us and now people don't need to read our content anymore. And I mean, there's all sorts of arguments that they're trying to make. But at the end of the day, they know that AI and the age of AI is shifting how people read the news, how they get information, how they see ads. I think the problem is that most of the content right now and a lot of the content risks become what the internet is already complaining about, right? There's this kind of low quality AI slop. So the business model from BuzzFeed right now and this kind of wave of AI powered content apps isn't very clear, but they're obviously experimenting. It'll be interesting to see if they're able to actually make money off of this. So the biggest story that I've seen today is it's in politics and it's basically a preview of I think what AI regulation is about to look like. There are US senators that are now calling on ByteDance to quote unquote immediately shut down their AI video app SeedDance 2.0. SeedDance basically lets users generate AI video and you can make these videos of real people or you can do things of, you know, videos of something that is like a licensed character and, you know, not inspired them by them, not kind of loosely based on them, but you can directly use their likeness. So basically we're talking about, you know, content featuring people like Tom Cruise, Brad Pitt. You could do, you know, someone from like Stranger Things, right? And you could basically generate all of that with SeedDance. Now SeedDance is a big app as it's incorporated straight into CapCut, which is one of the biggest video editors in the world. And by the way, I also have SeedDance. We just launched a video on AI Box.AI. So we have it over there if you want to try it out. But right now they're getting a lot of heat because two senators, one Republican, one Democrat, they both sent a letter to ByteDance saying that this is one of the clearest cases of copyright infringement they've seen from an AI product. And then basically, they're just saying shut it down and put real safeguards in place. I don't think it's just politicians. I think Hollywood is probably lobbying this pretty hard. The Motion Picture Association apparently sent a cease and desist. There's lawsuits that I think are probably coming from this. And ByteDance right now has already paused the global rollout where they're trying to deal with kind of some of the legal fallout of this. And I'll be honest, I've actually tested SeedDance 2 and I was really impressed with it. I was a little bit shocked that you could generate a video of Tom Cruise and there was no guardrails. Personally, as a user, I was like, wow, this is super cool. This is the first video model that feels like it should do anything I tell it. But yeah, we're about to get that nerfed a little bit and maybe it's for good. And I don't know, for me as a user, it's kind of disappointing, but whatever. Maybe that's all good, right? AI models are obviously trained on huge amounts of data and that includes copyrighted materials, right? So when you're sucking in all the video on the planet, you're inevitably going to get tons of clips of Tom Cruise and Brad Pitt and all these other actors. Up until now, I think most of the debate has been pretty theoretical because OpenAI doing something like Sora or Google V03 has been, these are very responsible companies, or at least they're trying to be in how they do that with, you know, Sora, all of the kind of quote unquote deep fake videos that you'll see coming out of those are people that are giving their permission generally.