Latest / In the Loop AI / Inside Meta's $15B AI Crashout and $100M Employee Signing Bonus

Transcript

Inside Meta's $15B AI Crashout and $100M Employee Signing Bonus · In the Loop AI

• Ejaaz: [0:03] Okay, this might be one of the most expensive panic buys in the history of AI. Meta just dropped $15 billion on a company called Scale AI. And within 24 hours, Google, their biggest paying customer, pulled $200 million of paid services from them. And in following, OpenAI, Microsoft and Anthropic also did the same. What is happening at Menlo Park? What is happening with Zuckerberg? It sounds like he's crashing out at this point. Ejaaz: [0:31] So Meta, guys, Meta has had a tumultuous week. Firstly, the news broke out last week that Meta was planning to make a sizable investment in this company called Scale.ai. They spent $15 billion for 49% of the company. So not even a majority stake, 49% of the company. So that's a post-money valuation of what, $30 billion. But it's not necessarily to acquire or own the company. It's to get the team. It's to get the talent. And this follows a trend of Meta basically falling behind on AI completely. Their leading model, Lama, has come nowhere near or as close as the leading model from OpenAI, Google, or Anthropic. So they're getting in quite a desperate position. And Zuck is kind of like knocking on all the doors and kind of raining hell. Now, I want to focus on this point of spending $100 million, that's seven to nine figures in comp just to sign on as a sign-on bonus for Ejaaz: [1:25] one single individual, for one single employee. But before we get into discussing this, guys, I want to get into the hot news, which is Sam Altman's take on Meta trying to poach his employees for $100 million. Here's the video. Josh: [1:37] They started making these like giant offers to, you know, a lot of people on our team, you know, like $100 million signing bonuses, more than that comp per year. That's crazy. And I'm actually, it is crazy. I'm really happy that at least so far, none of our best people have decided to take them up on that. I think that people sort of look at the two paths and say, all right, OpenAI has got a really good shot, a much better shot at actually delivering on super intelligence and also may eventually be the more valuable company. But I think the strategy of a ton of upfront guaranteed comp and that being the reason you tell someone to join, like really the degree to which they're focusing on that and not the work and not the mission. And I don't think that's going to set up a great culture. And, you know, I hope that we can be the best place in the world to do this kind of research. Ejaaz: [2:25] I mean, it's just brutal. Sam is basically calling out Meta, saying it's been slacking on AI completely, despite the billions of dollars that it's spent over the last few years. And to top it all off, he says, it's just not a great culture to innovate at all. Guys, what's your take on this gut reaction? Josh: [2:44] And there's two, it's two pronged. So one of them is like, this is outrageous. $100 million for an employee is insane. That is nine figures. That's like one with eight zeros behind it, which is a tremendous amount of money. But then it's also thinking like you go back to comps and you think, okay, well, when else did people spend this much money on employees? And what did that yield for them? Well, we have most recently OpenAI where they bought Johnny Ives IO company. And it was employees, the net per employee, I think, came out to over $150 million, which is crazy. So they paid equally as much for those employees. We're yet to see if that was a return. And then going back even further with Apple, well, they paid half a million dollars to buy Steve Jobs back. So that worked out well. David: [3:27] I think- There's precedent is what you're saying? Josh: [3:28] Yeah. So I think there is precedent for this, where buy Steve Jobs for $500 million, $4 trillion company. Buy Johnny Ives' design firm, take OpenAI from maybe half a billion dollars to many trillions of dollars we don't know and then this is even further down the we don't know stack where you're kind of competing for this weird invisible thing of AGI and you're not sure what form factor that's going to take how you're able to extract value in terms of financials from that and it's it's this like kind of outrageous thing and then the other take that I had immediately after watching is well this is actually really good defense that Sam is putting up as well against his employees where by publicly supporting like hey my employees are getting all these offers but we have the best culture here. And it would be like shameful if you were to leave for this gross, nasty company. So even if he doesn't- Yeah, so deal with the devil if you take that. Yeah. Like, ooh, you're making a deal with the devil if you want to take $100 million. We won't give you that much, but like we're a much better culture and a much better company. And while I believe he probably deeply believes that, it's also a tactic as well to publicize this and make it known like, hey, this meta company is trying to throw money at this problem and they're not really doing it the right way. They don't have the culture. They don't have the right intentions. They're not working towards the right things and nor will they ever get there. So it's this two pronged thing where it's like, OK, this is kind of outrageous. And then it's Sam playing defense and kind of like, oh, well, maybe like don't don't do that. It's stupid. I don't think it's doing that good type thing. David: [4:49] So, OK, understanding that there is precedent for this, there is precedent for acquisitions or acquihires that are coming at a very steep price tag coming in at like one hundred million dollars per employee. Like Facebook did this when they acquired Instagram way back in the day, you know, very low number of employees that they paid well over a billion dollars for. Same thing with WhatsApp. David: [5:08] And so this isn't unheard of, as you said, Josh. But I think the big emphasis is, is this coming from a place of strength or is this coming from a place of weakness? And I think we can unequivocally say that Meta's AI models, LLAMA, they're mid. They're mid. And the AI teams at Meta are not the same caliber of teams that are found at OpenAI or Anthropic. There are AI native labs. Then there are AI teams inside of these already pre-existing Web2 bureaucratic gargantuans that have fiduciary duty to their shareholders to maximize profits for their shareholders. And there's just a completely different notion culture to sam altman's point as to like who you work for and why and like sam altman has been repeating we're gonna make agi we're gonna make agi and i think you can also find that same culture at anthropic too and it opening on and anthropic also are the two ai labs that are known to retain talent the best and the people the organizations that are leaking the most talent are the big web to gargantuans and so So at some point in when you are in the fight to make AGI and you're in the race to make God. David: [6:20] The price tag of like your compensation just doesn't really matter. So the difference of like Meta or Meta paying you $100 million versus OpenAI paying you $500,000, which is not anywhere near $100 million, but still quite a lot. At some point, that price tag just doesn't really mean much when the stakes for changing the world as you know it are so incredibly high. Ejaaz: [6:43] Well, not just the stakes, but the potential cost, right? Like how much would you pay to own the number one leading AI model of the future? Ejaaz: [6:52] $15 billion probably doesn't sound like too bad of a price tag given a decade from now, right? The other thing is like $100 billion is probably just a drop in the ocean for Facebook or sorry, Meta's annual rev or daily rev. So they probably don't care too much about that. But there is one major distinction between Meta and some of these other companies. And it's the fact that Meta hasn't historically been known for being an infrastructure company. They are a social media consumer company to start off with, right? And if I were to ask both of you, hey, you know, what's the number one definite bet for a consumer AI app? I don't think we have a confident answer based on that. But if I were to bet on the team that would come up with that. And by the way, consumer apps in the Web2 era ended up being the highest valued companies in the Web2 era, right? So it's safe to say that this is probably going to play out the same way in AI. And I would bet on Meta being one of the leading teams to do that. In fact, like on Zuck's appearance on Dwarkesha's podcast, he spent 20 minutes literally just saying like, I'm not trying to compete in the infrastructure game. I'm trying to compete in the consumer game. Now, this purchase of scale AI to try and acquire this talent, I don't know whether he's trying to orient a team around building the best consumer AI product, but it's something worth considering. And then to your point, Josh, like he's done this before. Ejaaz: [8:14] If he hadn't had the wins with WhatsApp and Instagram, I wouldn't have as much faith in this playing out. But I don't know, if I was a betting man, maybe he has a shot. Josh: [8:22] I trust Zuck's judgment. He has run this company as the sole founder for 20 something years, however long it's been. And he's managed to navigate all of these tide turns. I mean, he turned a social media app into a hardware company with their glasses Josh: [8:37] and advertising company with sponsors. Now they're doing an AI company. And he's repeatedly been able to pivot the company directionally correct there's this advantage that meta has that zuck might realize in this consumer infrastructure game and that's just the sheer amount of users that they have as a company they have a significant percentage of the planet as monthly active users on their company and that's something that no other ai competitor has so i assume that probably has a big role in it where if he can just convert a fraction of these monthly active users that facebook has into consumers of an ai type application that should yields significantly more revenue than they're spending so i mean zuck's not dumb he knew that when he bought the company the competitors would pull out it would lose its revenue i'm sure that was baked into it i'm sure it's just looking forward he must believe that there's some way to monetize this gigantic user base in a way that can actually earn you more than than 15 million dollars or billion dollars Yeah, Ejaaz: [9:34] He has everything to lose, but also everything to win, right? And he's playing a high stakes game of poker now, just basically betting on the talent. I'm curious what you guys think of the scale AI CEO. So like, that's truly who he's been rumored to be going after. So for context here, Meta is planning to launch a new AI labs. So that's an internal team that kind of like works on all their AI efforts. And so far they've had a team internally, but they haven't been performing very well. So his plan supposedly is to replace them with 50 of the best AI researchers and builders in the world. Ejaaz: [10:11] And Alexander Wang is the CEO and supposed new head of this new team. I don't know too much about the inner workings of the scale AI team. I just know that they are a data provider, which is traditionally very, very important to creating the best AI models. Ejaaz: [10:29] Is Alexander Wang the guy to kind of lead this new team and create frontier model developments on the consumer background? He doesn't have any background on that. He's an infrastructure guy, right? So presumably like him and Zuck kind of like co-work on this new thing. Another thing that I was thinking of that is kind of coming from left field, Josh, and I know you'll appreciate this the most being, you know, like the hardware enthusiast here is I remember calling Zuck completely nuts when he leaned into the whole metaverse thing, right? He renamed Facebook to meta, right? And when he invested in Oculus, I remember thinking, dude, didn't you just see Google Glass just completely fail? And now you're doing this Oculus thing? You're insane. Fast forward to now and, you know, they're signing DOD, Department of Defense partnerships and, you know, doing all this new Anderil stuff. And I'm like, oh, maybe you are kind of like investing in the future here. Do you think he's doing the same thing? Do you think he's got a shot at it? Josh: [11:24] Yeah. And it feels unfair to judge because it's so clear that nobody in the world has more context than Zuck. And a lot of the times these ideas, the problem with them is like you saw when he bought Instagram way back in the day, they paid a billion dollars for it. And there's actually been leaked emails around this acquisition. And almost all of the people that were giving feedback were saying, this is absurd for you to spend that much money on something. That's just, you're never going to make this back. It is, you're buying a hundred people and one product that we could copy in a couple of months. And it turned out that everyone was wrong and Zuck was right. Zuck saw a future in which like photo sharing and video sharing mattered. And now I think most of my friends at least use Instagram far more than they use Facebook. So Zuck has been living in this world at a level that we're just unaware of because he's in it all day, every day for decades at a time. And I'm sure he has this version of reality that exists five years from now, 10 years from now, that he might not be sharing publicly, but that's clearly what he's guiding for. So I want to give him the benefit of the doubt because every pivot he's made that seemed outrageous at first has paid off in the end. But again, I mean, there's, it's not always a guarantee and there's a, there's a strong chance that they are behind. I mean, it's very clear that they're behind. There are llama models. They never released the big flagship on they're not doing well so maybe he maybe this is reactionary but maybe this was all baked into the plan all along i mean there's no way to really tell David: [12:46] I do not want to give him the benefit of the doubt. There's a very big difference between the parallels of buying Instagram in whatever year that they bought Instagram versus buying Scale AI. Scale AI is not the thing. Instagram at the moment was the thing. It created its own category. There was no alternative to Instagram, and Facebook bought it. That would be like if Facebook bought OpenAI or Anthropic. They bought the thing. That's not what's going on. Mark Zuckerberg doesn't have the thing. They have to buy this data processing company, Scale AI, in order just to get the talent. And so I don't really think that there's strong parallels here. David: [13:30] Again, Zuckerberg, way smarter than me, has way more context than me, as you said. But it's just so clear that he is coming from behind and he doesn't have the same leverage that he had using Facebook inside the WhatsApp story or the Instagram story as he does today. And so we all know the meta LLM is just not as great and they just acquired this team. And so they are now starting from scratch in the year 2025, while OpenAI is moving into hardware because their software products are sufficiently competent, sufficiently profitable, and they're generating sufficient like just new user base that they are ready to expand into hardware. And so they are already multiple steps ahead of where Facebook is. And so, like, I think the emphasis on coming from a place of weakness, I don't think can be overstated. Ejaaz: [14:18] Let me offer a slightly different, nerdier perspective to counter you, David, okay? So... Ejaaz: [14:26] I might argue that this isn't necessarily a talent acquisition, but it's a knowledge and data acquisition. So to dumb it down, if you want a really freaking good AI model, easy way to get there, easy, but very expensive way to get there, is to just get all the amazing data that none of the other model producers have and bung it into your model, right? Train it on that. And the chances are the next model that you produce is going to become hyper intelligent, right? But I don't actually think that's the benefit. I think he wants the scale AI team specifically because they've been working with all the other leading AI producers and they know exactly what data they're looking for. And most importantly, how that data fits into your AI products, whether it's the model, whether it's the app interface or whatever that might be. So imagine you're Zuckerberg and you're like, shit, AI is going to change everything. We're the leading social media platform. we need to disrupt ourselves and figure out how to integrate AI into our platform ASAP, whatever that might manifest into, you would probably buy the team that knows the ins and outs behind the scenes of how to create that potential new product, right? I'm not saying he knows what that product is going to be. I'm very sure that it's still in experimentation mode. But if you were to create an experimental team, it would be the guys that kind of manage and control all the data that goes into making the best AI models. Just a slightly different perspective Josh: [15:55] Similarly to apple they're also probably not directly competing with open ai on the large model infrastructure where apple's kind of going for the the user focus we're building the application layer using smaller models i think facebook is probably going to wind up or meta is probably going to wind up pivoting towards that we saw the llama 4 model fail horribly the giant one that they were trying to do and it seems like with this acquisition, they're really pivoting the focus from these gigantic models to the consumer space and understanding more data of how these models work, how people use them and further optimizing for. Ejaaz: [16:31] So Sam Altman is going, David: [16:33] I'm trying to build AGI. And so is Anthropic. And you just think that Mark Zuckerberg is like, I don't care about AGI. I want to make AI products that are good for consumers that are where consumers are and put them into my trillion dollar company. Is that what you're saying? Josh: [16:47] I'd love to ask him the question, but it appears as if that's true, where it's very clear, like Google is going for AGI, OpenEdge is going for AGI, Apple is going for users, Zuck might just be going for users also, where the most value that he's going to be able to generate, his competitive advantage doesn't lean towards building AGI, it leans towards users and giving users value. So i'm assuming that's probably the the way that he'll go we we kind of had this there's a similar thing with the oculus acquisition too where he spent a lot of money on oculus which seemed ridiculous and absurd and stupid and these glasses and goggles were useless for the last 15 years they fired palmer the founder so they acquired him for the ip and the person they even fired the founder palmer but then fast forward to now well now they're building the glasses and the goggles they just brought palmer back and they have the defense contract and and it all kind of forms together. So I don't know. I'm going to trust the vision, Zuck. Don't mess it up. Josh: [17:41] Go for the consumers. And hopefully that works. Ejaaz: [17:43] Actually, on the topic of Meta creating an AI powered social media platform, have you seen the new feature that got heckled and absolutely destroyed this week? Josh: [17:54] This isn't threads that already failed, right? This is something new. Ejaaz: [17:58] No, completely separate social media site feature newsfeed. Okay. So, so David, actually, if you pull up this page, it might be easier for the viewers or listeners to kind of see what I'm talking about. David: [18:11] Yeah, this is meta.ai. Is this basically meta's version of ChatGPT? Because it kind of kind of looks like that. There's a little search bar and I can make queries. Right. Ejaaz: [18:20] It's exactly that, except any prompt that you type into the search bar basically gets publicly shown to the entire world. So to be clear, if you, I, or Josh has ever engaged in Meta.ai on any of their apps, which includes Instagram, WhatsApp, or of course, Facebook, your prompts are publicly posted on this feed. Now, one might think, okay, well, what's the difference between that and, you know, typing a search query into Google, well, you're not typing intimate details about your life and personal thoughts and vices into an AI search bar or chat GPT, right? You're typing very personal things. And to add that publicly is crazy. I mean, I saw a story, not a story, I saw a literal screenshot of some guy, he says, I'm a 66 year old man, and I'm looking for a younger woman to potentially be my new wife, what are the best countries to go to where they like older men? I mean, it's insane, right? David: [19:23] And that was made public? Ejaaz: [19:24] So take a look at this, right? I can read the conversation verbatim. He goes, what counties do younger women like older white men? And, you know, the meta AI response, that's a pretty broad and subjective question. Attraction is highly individual. It basically gives a very diplomatic response. Right. And then he responds, I need details. I'm 66 and single. I'm from Iowa and I'm open to moving to a new country if I can find a younger woman. Now, this is the difference of typing in something on Google, which is kind of like what everyone's getting access to. And then thinking you're having a very private conversation with an AI and then it being made completely public. Josh, I can see you pulling a bunch of faces here. what's your take on Josh: [20:10] This it's like well there's nothing new here like this nothing changes this weird stuff happens all the time except you can't see it it happens behind closed doors it's only visible to google for meta to open this up to the public as default is crazy there needs to be this Ejaaz: [20:27] Is intentional this is intentional this isn't like they leaked your dm this David: [20:32] Is like a literal. Ejaaz: [20:33] Product You've just opened up Meta.ai. This is a live product. Josh: [20:38] There needs to be a big red banner above the search bar that says your searches, your queries are public. And the fact that they aren't leads to some, yeah, pretty, pretty interesting. Ejaaz: [20:48] So I'm going to fight Meta's case, right? I'm going to go pro-Meta for a second. Social media sites have always been heralded for public interaction. Facebook would not have succeeded if you couldn't go to your friend's wall and like a bunch of their posts or see their pictures, right? So social media websites in general have thrived on being public. I just think Meta didn't really think this product through well enough before they kind of posted it, right? I'm looking at David scrolling through all of these things and two things jump out to me, right? Number one it kind of looks like the company the product mid-journey in its v1 or v2 phase do you remember this josh where basically mid-journey is this product where you can type in a prompt and it creates some kind of visual image that's what we're looking at right now for the majority of david's screen and feed right and in that perspective it actually works really well because i remember scrolling through seeing people's mid-journey prompts and their outputs and being like, oh, I didn't know I could make that. Let me make that. David: [21:49] This is a Pinterest. It looks like Pinterest. Ejaaz: [21:51] It looks like Pinterest, right? So I'm like, oh, I'm inspired by this. I'm going to try and create my own version of this, right? I can imagine Google VO3, as I was spoken about on a previous episode, doing the exact same thing, right? But when it comes to words, when it comes to intimate dealings where you're having a private conversation with an AI, you're asking it very, I don't know, intimate details about a relationship that you're in or some kind of like issue that you're going through and you're treating it as your personal therapist, you probably don't want to air that laundry out to your best friend or to your family, God forbid. And you want to keep that private. And I think Meta just didn't understand that nuance when they launched this open chat GPT. David: [22:30] So I'm looking at this one account, Garpatau135. Can you recommend the best sources of protein? I'm interested in learning more about animal-based versus plant-based. I'd like to know about their overall health benefits too. This is a totally normal prompt that I would totally ask my chat GPT. I don't know if this needs to be made public as a default. Like, you know, I want to, there are things that I want to share. And then there are things that I like are just mostly irrelevant that like, why would anyone care about this prompt? And then there are prompts that are like, I definitely do not want to share. Like, no, that is for me, not for anyone else. And yeah, it's very obvious that users of this platform are not aware that their prompts, their queries, are being made public. MARK MANDELAVYSKI- There's. Ejaaz: [23:14] Actually another feedback loop, David, where you know how we scroll X every single day, right? And how do we hear about the latest viral thing that's happened in AI? It's usually on X. So if there's like this breaking new prompt, which, you know, gets your AI to act a certain way, I'm going to copy and paste it from a tweet and put it directly in there. And I'm going to test the prompt, right? So meta isn't completely misfiring here. There is something there where I like see a prompt. There's something here. There's something here. I just think the execution completely failed. Josh, what's your take on this? Josh: [23:46] You got to know your users, right? Like I imagine the people using this are not the ones who I want prompts from. I'm not interested in like your weird cartoon, like ball pythons, I reflects like this just, it doesn't make sense to me. And I think, I mean, to, to go against the case I made earlier, the bear case for meta in the world of AI. Well, when they acquired Instagram, when they acquired WhatsApp, when they acquired Oculus, they were already on these, like these successful trends, but meta has not yet actually been able to generate a new trend. And I think this is like a core difference where meta actually tried this recently. We mentioned it earlier with threads and And Threads was the direct Twitter competitor. And Threads is now used by a total of zero people that I know. And they've tried to like force it into their product. So I see it on my Instagram feed. You see it on Facebook. But without the network effect, like everything happens on Twitter. Josh: [24:34] I have no interest in going to Threads. And I think they've tried this and failed. And they're trying this again. And it seems to me initially like a failure. I could see the sentiment around it, why they would want to do it this way. But it's just what a terrible execution. like you look at this and there's not a single thing on this page since they've started scrolling that appeals to me that makes me want to like click that and use that and i think that's probably where they're they're at now it's just like what do we do this doesn't work people don't know how it works yeah David: [25:01] There's an absence of curation here like one of the things that is front and center here free date ideas miami so clearly a user was looking for a free date idea in the city of miami I am not in Miami. I am in New York. And Meta should know that. And overall, like the average person coming to this website is not going to be of the cohort of people who are interested in free date ideas in Miami. There needs to be curation. So, like, there needs to be a Reddit element here of if more people find this one query, this one prompt useful, it should be elevated higher. And if only one person found that query, it shouldn't even be shown at all. And so I get the notion that like there are prompts that are universally useful and universally appropriate for every single user. And when ChatGPT or XAI or Meta or whatever, when they notice that one user's prompt is getting copied and pasted and shared. And now all of a sudden that one user turns into 10, 100, 1,000, a prompt goes viral, if you will. They should elevate that prompt. And that's a good social media front end. That's just not what's happening here. I'm getting free date ideas in Miami. And frankly, I don't care. I just don't really care about free date ideas in Miami. Ejaaz: [26:16] Yeah, I think curation is really important. But also as a consumer, I'm interested in the output. I don't go to YouTube necessarily to kind of like learn how to create content. I learn to like, kind of like just like, go to the subjects that I like, search the things that I like, and kind of like consume, right? So maybe I'm not interested in the Miami outdoor date activities that I'm seeing here, David, but I am curious just to like learn and see like what kind of, because like it's this weird thing that I'm trying to define here, which is like, I kind of want to know what the intimate things you're discussing, but I don't want to know who's discussing it. I don't want to know who's discussing it. Do you know what I mean? It's the beauty of Reddit, right? It's like pseudo anonymous slash anonymous, right? So I go there and I'm like getting real, insight into how the human mind really thinks, right? And I'm just like, I kind of want access to that. Anyway, Meta, the leading social media platform provider, has completely failed on their first attempt. I have high hopes for their second or third attempt, maybe. Boys, anything else that you want to discuss or get into? David: [27:21] No. I wish Zuck. Josh: [27:22] And Co all the best. David: [27:24] Yeah. And their $100 million paid packages. Ejaaz: [27:27] Yeah. And all those employees, Josh: [27:29] Exactly, who can now retire their next 10 and generations, congratulations. MARK MANDELAVYSKI- Yep. Ejaaz: [27:33] MARK MANDELAVYSKI- Yep. Awesome. David: [27:34] MARK MANDELAVYSKI- All right. For all the listeners, thank you for listening to the episode. If you are not subscribed to the channel, please hit that subscribe button. We do these episodes twice, three times a week. We're going to be interviewing a few AI experts in this space as well, so make sure that you are tapped in so you can get all that content delivered straight into your feed. Josh: [27:49] MARK MANDELAVYSKI- Awesome.