The Journal. - Why Meta Is Offering $100 Million for AI Geniuses

Episode Date: July 3, 2025

In the battle for AI supremacy, Meta’s models have lagged. Now CEO Mark Zuckerberg is racing to hire new AI talent to close the gap with rivals. He’s dangling huge pay packages to lure experts away from OpenAI, Google, and Anthropic. WSJ’s Meghan Bobrowsky explains how Meta’s AI efforts fell short, and who will be joining the company’s new “Superintelligence Labs” to turn things around. Annie Minoff hosts. Further Listening: -The Battle Within Meta Over Chatbot Safety  -Why the New Pope Is Taking on AI  Sign up for WSJ’s free What’s News newsletter.  Learn more about your ad choices. Visit megaphone.fm/adchoices

Hundreds of top AI researchers recently found themselves in an unusual position. They were getting personalized messages from one of the tech industry's most famous people, Meta CEO Mark Zuckerberg. His pitch? Come to Meta. He's sort of become Meta's recruiter-in-chief. That's our colleague Megan Bobrowsky. Come to Meta. He's sort of become Meta's recruiter in chief. That's our colleague Megan Bobrowsky.

She covers the tech industry. In some cases, people didn't even believe it was actually Mark Zuckerberg, because why would you, right? I mean, can you imagine the CEO of this company reaching out to you? You know, one instance where a researcher didn't respond for several days because they thought it was a hoax. I mean, fair. Yeah, right? Zuckerberg is on a mission, to win over AI's brightest minds, to supercharge Meta's AI efforts.

Do we know what's in the message? Is it just like, hey, I'm Mark Zuckerberg, can we get coffee? I don't know exactly what the messages say, but generally once people do respond and are conversing with him, he invites them to his homes either in Palo Alto or South Lake Tahoe for a meeting and then they can go meet him and sort of talk about this.

We do know that he is offering some people up to $100 million to come join his company. Holy cow. Zuckerberg's offering these sky-high salaries because for him and for Metta, the stakes are really high. Whether Metta keeps its position at the top of the tech world could hinge on how Zuckerberg's recruiting targets respond. Welcome to The Journal, our show about money, business, and power. I'm Annie Minoff.

It's Thursday, July 3rd. Coming up on the show, Meta's fallen behind in the AI race. Can it catch up? Tech companies, from Google to Microsoft to Meta, are betting big on AI, wagering that AI supremacy will be the key to their future success. And AI is already peppered across many of Meta's products. You can ask questions to an AI assistant in Facebook Messenger. AI technology helps moderate content on Instagram. And in some countries, you can get AI customer service on WhatsApp.

But Meta's ambitions for AI are even bigger than that, especially when it comes to advertising. AI can be used to really advance its ads business. At its core, Meta is an ads business, and AI is already being used in some cases to optimize these ads to better target them. In some cases, it's being used to actually make the ads, to optimize these ads to better target them. In some cases it's being used to actually make the ads, not entirely from scratch, but it can make alterations to the ads,

it can turn photo ads into moving videos. So this is a big opportunity for Meta. If they can make their ad business even better than it already is, they stand to make a lot more money from companies who want to advertise and reach people on their platforms. And they're already scary good at targeting ads. I'm like, yes, I do want that skirt. You are correct. Exactly. So imagine if these ads were 10 times more effective.

That's essentially what Meta thinks it can do. And so that's where they're trying to go. Central to Meta's AI plans is its flagship large language model called Llama. That's where all this technology comes from, essentially, right? So you have the chatbots, you have improved advertising. What is powering all these things? That's this model called Llama.

And it's kind of a play on words for large language model. Oh, I just got that. So that's why it's called Llama, yeah. Meta debuted Llama in February of 2023. This was in the feverish months after OpenAI had rolled out chat GPT, kicking off an AI arms race. Tech's biggest players vied to develop

the most advanced, capable models. And Meta was right there in the mix. How did the release of Lama go? I mean, I think it was a big deal. Meta was one of the first companies to sort of open source this technology, meaning they opened it up to everyone. They're like, hey guys, we're here as well. We are in this game. Here's our version. But since Llama's debut, Meta and its model have fallen behind their AI peers. That became clear in April. This was supposed to be a big month for Meta.

The company was hosting its first ever AI Developer Conference. All right, welcome everybody to LlamaCon. To coincide with the event, they were planning to release the biggest, most advanced version of Llama yet, called Llama Behemoth. The problem? Behemoth wasn't ready. Here's Zuckerberg on stage at Llamacon. We have the pre-train of the Behemoth is done and we're working on the post-training, but even just kind of getting it. on stage at LamaCon. biggest model better and they're not seeing results. They're not seeing the improvements that they want to see

that would justify how much money they're spending on it. Right, so they're spending billions of dollars on this thing and it's not better enough than the old models. And so they decided they can't release it. Behemoth's release date was delayed from April to June and then delayed again. It's now expected to be released in the fall, or even later. According to Megan's reporting, researchers and engineers inside Meta worried that Behemoth's

performance wouldn't live up to expectations. Senior Meta executives blamed the team behind the latest model. A Meta spokesperson declined to comment. Instead of releasing Behemoth, Meta released two smaller versions of the Llama model. But those releases also didn't quite go to plan. The problem emerged when Meta submitted one of its smaller models to what's known as a benchmark test.

It's really hard to know how good is a model? What does that mean? What is it good at, right? And so this is a big question in the AI industry is like, how do you rank these things? How do you know? How do you tell who's number one? Yeah. How do you know which one is the best one? That's the big question. And so there's this whole ecosystem of third party tests that test these models and have rankings for which ones are better. The test that Meta submitted its model to is called Chatbot Arena.

Different companies submit their models and they're pitted against each other almost in like a battle. So what happened was Meta submitted a version of this model to Chatbot Arena version of this model to chatbot arena that was optimized to do well on the test. It was not the same version that they submitted to the general public. Interesting. So they sent the model to AI arena that was like specialized for these AI cage matches and then they served everybody else a different model. Yes. And so this is what happened. People find out about this and people's view of the

model kind of goes down, right? Because in their minds, Meta has gained the system to try to be better than it is. And people sort of question like, why are you doing this if your model is as good as you say it is? Zuckerberg acknowledged in a podcast interview that the model Meta submitted to Chatbot Arena was optimized for the test. The delayed release of Behemoth and the debacle over Chatbot Arena laid bare an uncomfortable truth for Meta. It was behind. Meta currently is one of the top tech companies in the world. If Mark Zuckerberg can't figure out Meta's AI problems,

Meta could lose a lot of relevance here. So Zuckerberg knows he needs to turn things around. What's his plan to do that? So Mark Zuckerberg decides there's a problem here. We need to solve it. And he's going to hire his way out of this. How Zuckerberg's recruiting blitz is going, that's next. To tackle Meta's AI problems, Zuckerberg has announced a new team. He's calling it Meta Superintelligence Labs.

The new division will house existing teams, like the one that makes the llama model, plus a new team focused on developing even more advanced LLMs. And the guy leading this whole thing is one of Zuckerberg's new hires. Tell me about Alexander Wang. Who is he? He's the CEO of this AI startup that is not itself building these large language models and this technology, but what they do is they label the data that goes into the models. So they're almost like the picks and shovels of the AI industry.

Got it. Mark Zuckerberg hopes that Alex Wang can kind of turn things around at his company. In an internal memo, Zuckerberg called Wang, quote, the most impressive founder of his generation. To bring him on board, Metta shelled out $14 billion for a stake in Wang's company, making him one of the priciest hires of all time. The new team still has a lot of seats to fill. And to help in his recruiting efforts,

Zuckerberg and his team have put together a list. It compiles all of the top AI researchers in Silicon Valley. It's everyone who he thinks is going to be able to help him turn around his company's AI efforts. And what kind of people are on this list? Yeah, so the types of people who are on this list, they typically have PhDs from schools like Berkeley or Carnegie Mellon or MIT. They've typically worked at places like OpenAI or Google DeepMind. They're usually in their 20s or

30s and it's such a small community that most of these people know each other and they're talking to each other about the fact that they're all on this list and trying to figure out if they should go to Meta or not. A lot of things factor into that decision. Of course there's the money, which for some could be in a hundred million dollar pay package. But there's also the glory. Many researchers want to be at the company where the big breakthroughs happen. And some are choosing to take their chances with Meta. So far, the company's announced 11 new hires for its superintelligence team.

Several from OpenAI, two from Google DeepMind, and one from Anthropic. But not everyone is taking Zuckerberg up on his offer. According to Megan's reporting, OpenAI co-founder Ilya Sutskova got an offer but didn't take the bait. So, for the people who are not biting, why not? Some of the people are getting counter offers. They're getting counter offers to stay at their current job. You know, Mark Zuckerberg is the one who's publicly out there recruiting these people,

but you know, certainly this is not a new thing. There is this talent war that has been going on for several years in Silicon Valley. And you know, we do know for a fact that OpenAI, for instance, has been offering people counter offers to stay. Really seems like the real winner here are these AI researchers playing these offers against each other. Yeah. OpenAI CEO Sam Altman addressed Meta's hiring efforts at an event recently. So it's like, okay, Zuckerberg is doing some new insane thing, what's next?

Earlier, he'd said that he was happy that his best people hadn't left for Meta. It's not clear yet whether Zuckerberg's hiring push will ultimately turn things around for Meta. What is clear is that the company is pouring money into AI. It's planning to spend up to 72 billion dollars on capital expenditures this year alone. Largely on their AI development, so that includes

acquiring all the hardware they need to build this stuff. It also includes powering these data centers, these huge data centers where they actually train the models. And now we'll also include the salaries that they're paying people. Let's say that Zuckerberg fails in this recruitment effort, this recruitment drive, and Meta does not catch up on AI. What would that mean for this company? It's not immediately existential for Meta.

They're a huge company that brings in billions of dollars in advertising revenue. Where this could become a problem for Meta is years down the line, if other AI companies are able to be more successful and are able to implement ads of their own, Meta could lose ground in the ad space and it could be a big problem for them.

A lot of these tech CEOs believe that AI is going to be a game changer to their business going forward in all aspects. Some CEOs have talked about AI writing code for the company. You know, there's also all the chat bots. And so if Meta is not on the cutting edge of this, you could see them start to slip down the tech totem pole. That's all for today, Thursday, July 3rd. The Journal is a co-production of Spotify and The Wall Street Journal.

Additional reporting in this episode by Ben Cohen and Berber Jinn. The shows made by Katherine Brewer, Piya Gadkari, Carlos Garcia, Rachel Humphries, Sophie Coddner, Brian Knudsen, Matt Kwong, Colin McNulty, Jessica Mendoza, Laura Morris, Enrique Perez de la Rosa, Sarah Platt, Alan Rodriguez-Espinosa, Heather Rogers, Pierce Singie, Jeevika Verma, Lisa Wang, Katherine Whalen, Tatiana Zamis, and me, Annie Minoff. Our engineers are Griffin Tanner, Nathan Singapok, and Peter Leonard. Our theme music is by So Wiley. Additional music this week from Katherine Anderson, Peter Leonard, Bobby Lord, Nathan Singapok,

So Wiley, and Blue Dot Sessions. Fact-checking this week by Kate Gallagher and Mary Mathis. We're off tomorrow for Independence Day, but we'll be back with a new episode on Monday.