Raindrop raises $35M from CRV to build AI-powered agent reliability and simulation for Fortune 100 companies

Speaker 1: and now with AI agents. Go check it out, Ben. How you doing? Doing great. Welcome to the team in Ultradome.

Speaker 2: How you doing? John's going detective mode. You guys look great. I didn't get the memo. Yeah. That's okay. You wanna wear one? You can I would? Yeah. Yeah. Enjoy. Tyler, maybe sacrifice yours. No. We we we can share. Look like you look sorry. Can't rock. Look like you're about to release a book on fear and doom. Yeah.

Speaker 1: Optimistic are you? How about the world? How are things going in your world? Are you a doomer? Are you an optimist? Are you an accelerationist? Are you cautious optimist? Getting right into it. Right right with the whole question. It's always it's always interesting to start from like, where are you? Because something I've noticed is that there are no doomers at the application layer. There are no doomers in the infrastructure trade. There are no doomers are only at the model layer. And there's a good reason, because that's where AGI and super intelligence emerging from. Think this is

Speaker 9: maybe a hot take here. I actually think that so I think part of the problem right now is that the whole AI safety conversation risk, whatever, it's like five or six different things all muddled up into one. Totally. It was like yeah. There's like Climate change from data centers is getting looped into Like, violence. Exactly. Very different things. There's like there's like the whole sycophancy thing, how it affects people's mental health. Totally. There's like Totally. There's like a bunch of different things. Yep. And I do think that for some of them, not necessarily the bioweapon thing or the data center thing, but, like, for a lot of lot like, a lot of the things I think about are at the application layer. They are like, that's where people meet models. Sure. And there there is a risk there that that that a, people at Frontier Labs, like, don't really care about or think about. Sure. Right? But, like, you know, we work with Fortune 100 companies that, like, their their products are used by millions of people and then affect even more people. Right? And so Sure.

Speaker 1: Well, start there. Reintroduce the product. Explain. I know you're here with fundraising news and product news, but give us the high level again.

Speaker 9: Raindrop is a agent reliability product. Important. So we detect issues happening in production, and we prevent those issues from happening. So the product we just announced recently, which I'm really, really excited about, is simulations. Cool. So when people make changes to their agents, we simulate what those changes are gonna do in the real world Mhmm. And can show people things they wouldn't have even ever expected to happen. Sure.

Speaker 1: So yeah. How reasonable is it to have a AI model that just reads every Trace. Thing, every trace.

Speaker 9: So Is that prohibitively expensive? Is that the problem? It's getting less so. Yeah. So I think, like, you know, we've we've sort of pioneered, like, super cheap classifiers for this thing. Jeff is another example of this. Yeah. Exactly. I'm just imagining,

Speaker 1: like, burn a classifier onto silicon that just says, is this bad or not? And have it run over the trace. You start typing things now. So there's a problem. Please.

Speaker 9: It turns out that, like, binary classification, like, is this bad or not? Is really, really hard. I can imagine. Yeah. Take it like a step, like, back. It's like, humanity has trying has been trying to define what is good and what is bad for, like, as long as we've existed. We're gonna need more than one simple truth. It's gonna be Yeah. Exactly. Yeah. So so we consider ourselves in some ways like a human alignment company. Like Sure. How do you get a company to define what is good or bad? Yeah. Yeah. Yeah. And then how do you make that as, like, easy as possible? How do you a lot of it is, like, especially now that agents can do, like, more than the people that make them have ever imagined that they can do. Like, they're, like, looking at Traces, like, oh, shit. Didn't expect that to be asked. I didn't know it could do that. Like,

Speaker 1: you need to be able to, like, surface those things and be like, hey, do you want to be able to do this? Do you want it to be Yeah. Yeah. So so creating, like, alignment within a company looks different. I always think about the tobacco industry because a lot of the models will reject to work on anything tobacco related. But, like, Altria is going to use AI somewhere. Yes. So they're but they're gonna get rejected a lot. But in some ways, they're gonna wind up deploying AI one way or another. And their sense of Yes. Legality is gonna be informed by the way they're regulated in every single market. And it might not look like the morality of the lab that trained the model,

Speaker 9: but it will fit within their business construct in a weird way. And the way we process the morality of of cigarettes is through the court system. Totally. I I think there's hundreds of examples of this, thousands of Yep. Or infinite, like Probably infinite. And especially, like, I think the way that FrontierLabs consider alignment is, like, often very different. Sure. I was, like, using, I think, like, Claude or something recently. I was, like, trying to negotiate on some product. And, like, I wanted the the it was, like, on Marketplace. And the guy, like, wanted to, like, hang out with me. Like, he was, oh, it'd be really cool if I could, like, sit down and, like, show you this thing. But I was just getting, like, weird vibes. I was, like I told him a lot. I was, like, you know, just just, like, say something that kinda, like, gets by this, but, like, in a way that's not too, like You know? Softly? And it was, like, I cannot be dishonest about this. Oh. We should not, like and I was, It's a white lie. Exactly. I was, like, come on. Like, you know, this could be my safety at risk here. Who knows? But anyway, so, yeah, I I think that human

Speaker 1: alignment is a really hard problem. And I think you should Did you hang out with a guy? No. I did not. I didn't even end up. He was just like, yeah. Actually, Claude booked us a four hour walk on the beach. No. It would have. Right? It would have. No. The deal wasn't worth it in the end. Okay. Okay.

Speaker 9: Good. Yeah. I think that, like I think that telling something, whether that's a human or an agent, what you want is incredibly hard. I think management is incredibly hard. It's just in general. Like, if it were really easy to express what you want and have the other person like, have some entity understand, like, management wouldn't

Speaker 2: half of management wouldn't be Okay. So so an interesting thing for your business right now is that consumer agents are getting more and more popular. We saw we saw the, you know, the Open Claw boom and then this cycle will be bigger because you don't need to go buy a Mac Mini or really set up much of anything. And you can just imagine every business on earth is going to start having first of all, they're gonna start realizing like, wow, our phone's like ringing We're a lot getting just like more messages. And so what that's gonna create is like tons and tons of businesses that need to create like counter agents to process and negotiate with other agents which is kind of funny because I'm sure some people will be like, why do we need a data center for this? Just like, why don't just work it out the old fashioned way? You know, but clearly people like having like an intermediary in a bunch of these different things. And so, is your bet that like there's this massive long tail of businesses that are gonna need to have their own agents and sort of process and basically grade them on on how they're performing and improve them?

Speaker 9: Yeah. I I I think that there's gonna be more agents and than than there are now.

Speaker 6: Let's go. The hot day of the century.

Speaker 2: More AI in the future. Right. But but but another question is, like, you would almost bet that a business is going to have to take some version of an agent and then fine tune it for their business and then test it. And I feel like that is where you're gonna fit it. Yeah. Totally. Yeah. That makes sense. Like a bit like, somebody might buy an aide say, I'll take one agent, but then I also want the software to understand how it's working. Yes. I think that companies

Speaker 9: will want both, like, humans and agents that are aligned to how their company works. Like, if you take, you know like, pick a random Fortune 100 company, they probably have some way that they work that, like, works for whatever industry that they're in. Yeah. And

Speaker 1: there might even be parts of it that are, like, you know, suboptimal and some, like I wonder if more culture stuff will bleed in. Like, there's a there's a couple famous examples of, like, if you go to Nike headquarters, like, everyone wears Nike. Yeah. Like, you don't walk you don't go to work. Major Haitian shirts. I'm wearing Adidas today. Like, that would be a bad look. That would be bad. And so but if you get an an off the shelf model You wear a Nike Fiddle. Yeah. But if you get if you get an off the shelf model, it's gonna be like, yeah, like Nike and Adidas, like these are both like comparable brands. They're both like pretty good. Yeah. And and you're gonna be like, this AI is not a culture fit. Yeah. Yeah. I need to and and that's like a very subtle alignment thing that actually might be real because you don't want to be building a deck and it's like, oh, yeah, like, you know, maybe we should partner with Adidas because they're similar and it's like I worked I worked at Apple for like four years. Yeah. Everyone works at Apple. Right? Yeah. Exactly. And

Speaker 9: and this is like a protest, but no. I I think that the way that AI is and agents will fit into there Yeah. Will be different than other companies. Yeah. Like, it's just like they're it's a the way it works is just entirely different. Right? Yeah. Like, if you ever worked at a big company, you sort of get that. And so, I don't know. I think there's millions of examples of this. I I think that yeah. I think a lot of the future will be how do you sort of conform Sure. Sure. Simulation,

Speaker 1: where does the simulation end? Because you can create a digital twin Right here. We're in it. But, I mean I mean, if you go to a company, you can create a digital twin of their their AWS instance Sure. And their Gmail Yep. Their Slack. Yep. But at a certain point, then you also need to simulate, like, the entire Internet because that's part of Like, the how big can you go? How big are you going now? How big will you go ultimately? This is a really, really good question. I mean, ultimately,

Speaker 9: you know, I'm joking, but also, like, you maybe know, this could be a raindrop simulation. I don't know. Yeah. Yeah. I don't know. For now, where we're starting Yeah. Is all about it's all about replaying historical things that have happened, getting really, really smart about how you choose those historical things, and then in very specific scenarios creating completely sort of like on the fly adversarial scenarios. As far as like, technically, we simulate everything. Mhmm. And it's kind of a hybrid of trying to do be as smart upfront about what you need to one one way of thinking about it is if you think about, like, let's take a pass trace, like something the agent did. Every time it called a tool, it's almost like a a hole through this, like, veil that lets you sort of see what the world behind it looks like. Right? The agent sort of perceiving that world one sort of, like, tool call at a time. So you can sort of start there and say, well, let's try to, like, fill in around the holes as much as possible. So if it goes, know, except here it goes, like, right next to it, it sort of has that data. Right? And then you also have to be monitoring that simulation in real time just like they do for this one. Right? To see, oh, you know, if something's slightly off. Right? So it's like maybe it tried to get data somewhere but that data was not filled in. You can sort of detect that as a quality gap if you're really smart about how you're classifying it. Mhmm. And then fill in that data at run time. Mhmm. So it that's that's essentially the way to do it well. How do you think about

Speaker 2: how different cultures are set up to even build agents? Because it feels like there's it's quite funny to me that that I'm looking at, like, Amish culture. I'm not culturally, we don't miss a lot. But but I I think it's very funny that I think it's very funny that Apple, like, is just getting around to cloning ChadGBT.

Speaker 1: Yeah. They have such an advantage in the personal agent thing. I know. They are the most trusted tech company with privacy. Killing it. They should be killing it. All of my passwords are in Google. Even even Google people will be like, they're gonna sell your data. They wanna advertise. Apple is the one where my family photos are in there. Oh, I know. They have very very high traffic. Test time was way behind. There's two companies that and Meta Meta is

Speaker 2: Oh, yeah. Totally. It's Google and Apple. Yeah. And Meta is specifically well suited for it because Zucks like, I don't care if it misspells like five out of a 100 people's names if nine you know, like you saw this with with one of the agents today. Somebody had an issue like getting on a flight apparently,

Speaker 1: because Yeah. He's always working like a messier, more algorithmic world than Apple where it's like, if the photo is on your phone, it must sync to your iCloud. Like, that's there is no there is no AI lineage, no stochastic lineage in the business. I think I think it's very true. I think that

Speaker 9: that Apple is first of I would literally never count them out because I think they're ridiculously well positioned and they have some amazing people there. But, yeah, like, I think that there was this, you know there there was a way of working there which was sort of, you know, we're going to, like, you know the reason why Apple is such, like, insanely you know, the the most insane, you know, attention to detail, etcetera, is that designers would sort of define, like, this is how it should work. Yep. Right? And then engineers are like, well, no. Like, that that won't work. And the designers are like, well, we don't care. Like, this is how it's going to work. Yeah. And, you know, you just kind of do that loop. Right? Sure. It's like RL or something. And the thing is that that does that kind of way of working doesn't work as well for AI. Right? Because

Speaker 1: you you can't just say, like, oh, it's like, well, you know, it should be, you know, as natural as I mean, it is entirely flipped at the Yes. At the Frontier Labs. It's like the researchers went and did something Yep. And this is what it can do. Exactly. Go market. Yeah. Like like like try to build a product. And and and but we do also see Wait. Wait. It can do math now? Yeah. Exactly. Go tell people it can do math. Exactly. Exactly. And they're like, but it can't do comedy or it can't do they're they're still six fingers but it's, you know, amazing at Python. And you're like, okay. That's the product. Yeah. Yeah. Exactly. Exactly. As opposed to the opposite, which is like 100%. We are building this particular thing. 100%. Yeah. Yeah. Very, yeah, very exciting times. What does it take to sell to a Fortune 100 these Like, what are you seeing in the enterprise? Arkarazian has this evolving take that there's sort of this duopoly in price competition at the frontier between Anthropic and OpenAI. He has said that there's less switching to open source and Chinese open source than he thought he would see. But there's tons of price competition on the frontier still. What are you seeing? What are what are enterprise leaders just, like, excited about, worried about? Anything that you can share?

Speaker 9: So, yes. Definitely way less adoption of, like, I would just say, like, Chinese models in general, especially Fortune one hundreds. I think that there's, like, definitely a narrative of, like, a risk around it even if it's not super well It's more like, let's

Speaker 1: go actually, instead of token max, like, efficient token spend will take us the next leg up. Cost?

Speaker 9: People do care about cost. Yep. And I think that it's, like, pretty reasonable when you're, like, even, you know, five five k, you know, ahead times, like, you know Yeah. Gets up. 80,000, 100,000, like, whatever. It's like it starts starts getting crazy. Right? Yeah. Yeah. So it does matter. I think that there's a lot of like, companies are definitely all building internal things right now, and I think trying to grapple with, like, well, like, when do we just use Cloud Code? When do we just build our own, I guess, Codex Now maybe. But, like, when do we just build our own thing? I I think that, like, there's more of a justification for companies to build their own kind of internal ways of working than I might have expected six months ago, actually. Yeah. Which is really interesting. Like, I think even if you look at something like Muse, it's like it's like pretty much, like, some Open Claw style thing. Heavily inspired. Yeah. Use their words. Yeah. Everyone was talking about his Open Claw Yeah. Telling him to drink more water Yeah. Strike Sessions, like, maybe four or five months ago. It's it's the the the barrier to entry there is, like, pretty low. But distribution is really important. Yes. But but if you have, you know, especially, I think, again, when you think about Fortune 100, Fortune five hundreds, think we're seeing an increasing number of just internal Oh, shit. Agents and applications that I think, like, two years ago might have felt a little silly. Like, it's kind of, like, well, like, why are you building this yourself? I think increasingly, it's like, no, like, you know, you have all these legacy, like, you know, internal docs, and like, exact, like, you know, the exact path a ticket should take, and every company Yeah. Sort of like maps to org structure, etcetera. It, like, kinda makes sense increasingly,

Speaker 1: which is Yeah. It is interesting that, like, even if you go back to .com, like, Amazon became the everything store and yet there were so many brands that didn't go all in on Amazon, maintained third party websites for decades Yep. Successfully Yeah. And and built brands even on even in retail and and big box. Like, all these things, like, lived on longer. And so, yeah, when I see, a law firm saying like, okay, we're gonna build our own models, we're gonna build our own Yeah. Harnesses. I'm like, yeah, maybe that's not that crazy. Maybe that'll stick around. Does brand design matter more or less today than it did two years ago? Oh, it matters more. Yeah. It matters more. Does it though? I think so. Because I feel like when everyone can have like a pretty decent

Speaker 2: brand, like I you see Yeah. Some teams like looking at at instincts, right, getting to to to 10,000,000,000 with the plain text website. You don't have a brand. Right? Is there a brand? No. Of course. They have like It's a name? Pretty Twitter brand. Right? Like like like brand is not always It's like what you do. The brand is super understated. But but here's another here's another example. It used to be like you could get a long way as an early stage startup designing like a really simple Yes. Cool three d asset. Yes. No. Totally. Like that, they would just instantly be like, okay, this team, they may not have a lot of resources but they like And now you can just be like, make me a cool three d asset. But this is know, the website flips around. Control. This is why it matters more though. So I think I've like I've gone back and forth. I think right now my view is that,

Speaker 9: like, some ways you could think of it as like taste is being commoditized. Mhmm. But also, the second that things are commoditized, you need to stand out. Yeah. And so, you do need someone at the, like, like, that understands the zeitgeist, understands whatever the tip of, like, whatever it is. It's like, you know, because it is not enough to throw a nice, like, three d asset on your website or something. Yeah. And so but you'd still need, like, for example, if a Fortune 100 company is landing on your page or, like, just anyone, you want them to understand that this is a real team of humans behind it. Mhmm. It's even like writing. Like, you can imagine that, like, so I I think AI writing is still bad. But also, like, if you sent, like, AI writing to people, like, ten years ago, you know, would they think it's bad? It's not obvious to me versus, like, there being a lot of tells because you are getting inundated with, like, really low quality

Speaker 1: writing that you Yeah. Obviously, like, old Paul Graham post where he uses all the AI. No. Yeah. And dashes, whatever. Yeah. Exactly. And he clearly wasn't using AI unless he's a time traveler. Which is possible. But yeah. Mine It wasn't annoying then and now it's annoying now because it's oversaturated. Yeah. Those

Speaker 2: the second I detected, I stopped reading and I just closed whatever the Exactly. What's weird is that I

Speaker 1: I feel like I'm detecting it less in the actual responses. Like, if I asked TragicPuty to go pull some data together for me Mhmm. I'm noticing they clearly are all, but it's not this. It's that out of the model on the last run. Of course. Yeah. Because people are like, it's okay. That's actually really annoying. But in general, like, I'll sit there and I probably read twenty minutes of AI text a day Yeah. Just from queries that I do throughout the day. You're lucky. That's all? I I don't know. I mean, I'm live a lot. So so much fun. Yeah. But but you're not necessarily annoyed by that. But if I tell you, like, hey, I got this blog post, and then you're like, this is AI, it's like disrespectful.

Speaker 9: We we actually But the actual quality of the writing is probably identical to what you read all day. I get what you're saying. Yeah. Yeah. I actually I actually do find the the responses from, like, Codex, etcetera to be, like, obnoxious. Yeah. Yeah. Yeah. Sometimes it can be annoying. But, yeah, we we we I had I had a funny thing recently where, like, we put up a blog and it was, like, it was, like, I I thought it was a pretty like, it's a pretty decent blog. It was about the hugging face incident. We made, like, a whole, like, simulated game called Hugging World. It's at huggingworld.com. Cool. And it's pretty funny. It's funny to watch. And so the thing is, like, I was pretty busy. I got the blog, like, 80% done, but, like, I knew that there were, like, some Yeah. Like, you could read you could read it. It was good. But it it also, I was, like, someone's gonna call like, someone's gonna see, like, the the, like, some AI written phrase here. I don't wanna get called on that. You know I mean? And so I added a little, like, stamp, like, vibe written, you know Oh, yeah. On the top. I think that was, like, a nice, like Okay. Yeah. Just being honest with the audience. Exactly. But it's like it's like there is good information in there. And was there pushback or no? No. That's great. Exactly. So I I think I think, like, there was good information. I it was not, I think, disrespectful to someone to read. Yeah. But at the same time, it's a little bit different than if I were to write every sentence. Yeah. Yeah. So there's some sort of

Speaker 1: I have one more question. Please. Then we have one big gong here. Yeah. We do. Yes. Maybe we should hit the gong first. You raised $25,000,000 CRV. 35. 35. Oh. Who's counting? We get it. Wow. Wow. We want Smash that gong. CRV CRV is any shipment up to Boston. Did you go to Boston or you raised in the Valley? No. No. Okay. Cool. Charles Rivers Ventures. The Charles. It's in Boston. We were there. It's great. Charles Rivers, I wanna call you. Completely unrelated question, but prediction in the tech world. The new Tesla Roadster is being announced October 1, and the FAA has cleared the airspace 10,000 feet up. I saw it. So the maximum height it could go is 10,000 feet. The minimum height it could go is is zero feet. Right? What is your prediction? It has to be It's gonna get off the ground. Inclusive of zero to 10,000 feet. How high do you think it'll go? Could be an inch. Could be a thousand feet, could be a 100 feet, could be 10,000 feet. Can't be higher than 10,000 feet. Can't be less than We know it's in between there. We have the bounds now. I think it's I think it'll be more than zero. Okay. Pretty sure.

Speaker 9: I'm thinking I'm thinking I'm thinking where to get 10 feet. 10 feet. I like it. Here's here's the new here's the These

Speaker 2: guys are gonna send it half I gotta get spoil under here. I've been thinking about it more, it really seems like he never wanted to act once he got into this whole thing and humanoids, he doesn't really wanna make this car. Right? No. Yeah. Exactly. He really doesn't wanna make this car. Right? Okay. People have been sitting there for nine years having their money locked up. They're like, I hope this year I get my car. He doesn't wanna make it. Right? Okay. But he we know he can make it. Yeah. And we know that he could make something so incredible. We know he could make a flying car. I don't think anyone doubts like, oh Yeah. I'm not saying it's gonna be super safe. Okay. But but let's say I'm saying it's gonna ship? Ship. Let's say this. He makes something that is no. So he makes something that is just like an insane demo but it just instantly gets like hard ban hammer by Oh. And he's like, oh no. I can't make it. It won't let me. Or it gets into another five year period of like legal battle. I gotta keep the deposits. Okay. Okay. Okay. We gotta get through these legal battles. The deposits are that big. But

Speaker 1: also there's a wild card which is that it could be more or less a normal car, but they cleared the airspace because it's doing sort of an evil kennel style jump and they set up a ramp. And that will throw all of our predictions off. Quite possible. Yeah. Because it could easily go 30 feet in the air if there's a ramp and it could just be a normal car. Just wouldn't be a flying car. You know, like, some like,

Speaker 9: he has he has he has the resources. Yeah. Right? And and Yeah. I'm sure at some point someone was like, well, what if it flies? Yeah. You know what I mean? And so he would have been like, well, let's get a team on that. Yeah. And I believe that if you you can attach engines to this thing and Yeah. Fly like I I I was going Why not? No ramp, six feet up.

Speaker 1: But after seeing the 10,000 foot news, I think you might be flying around like a plane or something. It might be getting up there. We'll see. You think so? Maybe. And even the I know why. Chase flights. That's the thing. Oh, yeah. There would have been spies or There would have been some pilots and there would have been yeah. I I think I don't know. They might have it locked down. It's a pretty secure Self driving? Yeah. I don't know. Self That's real. That's the easy part. Yeah. I I think the hard part is to actually make a make a rocket car that doesn't explode when you're Passive driving day. Yes. Exactly. Anyway, thank you so much for coming on the show. Congratulations.

Speaker 9: See you guys. Thanks being Of course, you're you're Thanks on. Coming to the show.

Speaker 1: And our next guest will be joining in just a minute. We have some other breaking news we gotta run through. We just set up our our our next guest. I think it'd be worth talking about Paramount. Let's get a quick tip of the hat for Tip of the hat. Friends over at Raindrop. Yes. Yes. Yes. Of course. So Paramount is officially not leaving California. It's been a back and forth. The company going through an acquisition with Warner Brothers Discovery has been duking it out with AG, attorneys general in California. The company said Monday, that's Paramount, said that it has reached a settlement with the states challenging its $111,000,000,000 acquisition of Warner Brothers Discovery, clearing what had become the last major obstacle to the deal. California attorney general Rob Bonta and a coalition of other states sued Paramount in July arguing that combining the two companies would give a single entertainment giant too much control over major film studios, cable networks, and other media properties to resolve the case. Paramount agreed to several conditions. The combined company will create an independent board of journalists intended to protect the editorial autonomy of CNN and CBS News, commit to releasing at least 30 movies theatrically each year, and spend an additional $1,500,000,000 on film production over the next five years. Paramount did not agree to more structural changes. They're not selling CNN or divesting one of the movie studios. Once completed, the merger will bring Paramount and Warner Brothers under the same roof along with Paramount plus HBO Max, CBS, CNN, and franchises and films including Superman, Minecraft, and Sinners. Paramount has argued that the scale is necessary to compete with increasingly powerful technology and streaming companies like Netflix. If you if you subscribe to every Paramount, Warner Brothers, Discovery premium service now, you're probably giving them, like, $60 a month, maybe more. HBO Max is an independent subscription. Paramount Plus, that gets pricey if you get the don't have UFC now. CBS, I think, has an all access. CNN launched something, but I think they shuttered it. Anyway, the settlement also follows months of political pressure surrounding Paramount's future in California. The company had warned that it could move significant production and thousands of jobs out of the state if the acquisition were blocked and had reportedly begun exploring studio space in Nashville. Governor Gavin Newsom publicly warned that losing Paramount would damage California's standing at as the center of the entertainment industry. With the lawsuit now headed for resolution, one of the more serious recent threats of a major Hollywood company pulling back from Los Angeles appears to have receded. So this is extremely bad news for everyone that was hoping that the Paramount lot would be converted into a data center in the part of Los Angeles. A lot of the pro data center crowd were optimistic about a large scale data center coming in to Los Angeles, using up a lot of the water, a lot of the energy here in Los Angeles, but looks like they're gonna be making movies instead.

Speaker 2: Well, let's check-in with professor

Speaker 1: Let's go to the top. Zhang Oh, what happened? Who went on Jack Neil's podcast to talk about AI. Oh, yes. Yes. We need to this is the funniest clip. This is like half true, half false. Let's play it.

Speaker 3: Do you really think when you put something into ChatGPT, it's gonna be generated by the AI? I guarantee you, it's not true. I guarantee you, it's being manipulated by humans somewhere in India. Human in the loop. ChatGPT.

Speaker 1: Human in the loop at scale. Billion billion weekly active users,

Speaker 3: billion people answering questions. I think what happened was a human first wrote that. And so he's

Speaker 1: sort of right. He's, like, describing training data and, like, it's half crazy. Like, no. There's no human in the loop when you ask a question. It answers with superhuman speed. How is that possible? But at the same time, like, yes, Scale AI exists. Merkur exists.

Speaker 2: There is a lot of training data. Professor Zhang is willing to actually guarantee this too. Did guarantee. He did guarantee. He's not just saying it. He's willing to guarantee it. Yes. It seems that that the new meta for governments globally Mhmm. Is to just have like a podcast guy Okay. That just goes around saying insane things, half of which is true Yeah. Half of which are are are made up. Yeah. And then it just sort of is exciting and confusing for people, but but very stimulating. John John Kiriaku is another guy Similar. Another guy like this, it seems. Although he's taken a turn into fashion. So we'll see. Oh, yeah. He walked at

Speaker 1: Fashion Week. There's an interesting thing. David Sachs is the AI czar, and he's also the cochair of the President's Council of Advisors on Science and Technology,