These AI experts are getting offered boatloads of cash by Zuckerberg
Why it matters: Some top executives and researchers in AI are moving from companies like Apple and OpenAI to Meta as Facebook's parent company looks for any leg-up in the race to build the top AI products.
Meta recruits top AI experts
Driving the news: Zuckerberg has been dishing out massive compensation packages to AI experts and talent like they're pro athletes or Hollywood stars.
One prospect received an offer "worth as much as $1.5 billion over at least six years," per The Wall Street Journal.
"Last year, the cost of a top, world-class deep learning expert was about the same as a top NFL quarterback prospect," Peter Lee, Microsoft's head of research, told Bloomberg BusinessWeek. "The cost of that talent is pretty remarkable."
Yes, but: The packages are often heavily weighted toward stock options over cash, which means their value could largely evaporate if Meta's stock drops
You probably familiar with the NFL and NBA stars who make bank for their talents. But identifying AI wizards isn't so easy.
Meta declined to comment on AI recruit salaries and compensation packages.
Here's a rundown of the AI prodigies getting offered boatloads of cash by Zuckerberg.
Alexandr Wang
Wang was brought into Meta when the company invested billions into Scale AI for a 49% stake. Wang, the company's CEO, was moved over to Meta in the deal to lead AI efforts and focus on "superintelligence."
Since then, Wang has joined Zuckerberg in recruiting top AI experts to build out their team.
Matt Deitke
What to know: Deitke, an artificial intelligence "whiz kid," was recruited by Meta to be an AI researcher.
The 24-year-old dropped out of the University of Washington. But he's worked at the Allen Institute for Artificial Intelligence, an AI lab in Seattle, according to the New York Times
There, he started Molmo, an A.I. chatbot that helps build images, text and sounds.
By the numbers: NYT reports Deitke's deal with Meta is "$250 million over four years, with potentially up to $100 million of that to be paid in the first year."
Andrew Tulloch
Zoom in: Tulloch is a leading researcher and co-founder of Thinking Machines Lab, an AI startup, per the Wall Street Journal.
The Australia native previously worked at Facebook on machine learning, WSJ reports. He was recruited from Facebook for OpenAI once the company started taking off. Now, Meta has sought to win him back.
By the numbers: Meta reportedly offered Tulloch "a billion-dollar package that could ... have been worth as much as $1.5 billion over at least six years," WSJ reports. Meta disputes this figure.
Tulloch hasn't accepted Meta's offer.
Shengjia Zhao
Context: Zhao was the former lead scientist at OpenAI and a co-creator of ChatGPT and GPT-4. He was hired to lead Meta's new superintelligence team, where he works alongside Zuckerberg and Wang on advancing AI models.
"In this role, Shengjia will set the research agenda and scientific direction for our new lab working directly with me and Alex," Zuckerberg wrote on Threads.
The Information reported in June that Zhao was one of several AI recruits for Meta, including OpenAI researchers Jiahui Yu, Shuchao Bi, and Hongyu Ren.
Lucas Beyer, Alexander Kolesnikov, Xiaohua Zhai
What to know: Beyer, Kolesnikov and Zhai were three of the top researchers at OpenAI's Zurich office before they were poached by Meta. The three helped launch the Zurich office after joining OpenAI from Google DeepMind.
Follow the money: Multiple reports suggested the trio was given a $100 million sign-on bonus. Beyer. himself, called it"fake news."
Ruoming Pang
Details: Pang, an AI executive for Apple who led a team running the company's foundation AI model, was one of the biggest recent Meta hires.
A well-known engineer and manager, Pang previously moved from Google parent company Alphabet to Apple in 2021.
By the numbers: Pang's offer was worth in the "tens of millions of dollars per year," per Bloomberg.
Nat Friedman
Flashback: Zuckerberg announced in June that Friedman, the former GitHub CEO, had been hired to work on the superintelligence lab with Wang.
Friedman's hire was expected since he had previously worked with AI investment firms and had served on the Meta Advisory group.
"He already has a good sense of our roadmap and what we need to do," Zuckerberg said in a memo.
Daniel Gross
Details: Gross departed startup Safe Superintelligence Inc. for Meta. Zuckerberg started recruiting Gross after Meta failed to acquire SSI earlier in 2025.
Gross was previously a tech investor with Friedman.
Trapit Bansal
Details: Bansal is an AI researcher who previously worked for OpenAI, where he was a foundational co-creator of OpenAI's first AI reasoning model, o1. He also worked with Microsoft, Google and Facebook as an intern, according to his website.
What they're saying: Bansal is seen as someone who could help Meta develop a competitive AI reasoning model, something it doesn't currently offer to the public, TechCrunch reports.
Other top AI recruits for Meta
Joel Pobar previously worked at Meta on machine learning before jumping to Anthropic.
Anton Bakhtin previously helped develop Claude at Anthropic.
Bowen Zhang was an instrumental AI researcher at Apple.
Mark Lee and Tom Gunter were two other AI experts snagged from Apple.
Tianhe Yu, Cosmo Du and Weiyue Wang were all brought into Meta. The trio worked on the Gemini model.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Gizmodo
27 minutes ago
- Gizmodo
Meta Says Its Comically Large VR Headset Prototype Can Almost Mimic Human Sight
There are a lot of lofty goals in the world of VR, but delivering hyperrealism is among the most elusive. That's because there are tons of constraints with a realistic, world-rendering face computer: there's resolution, field of view, power, weight, battery—you name it. But just because obstacles abound doesn't mean the titans of XR and VR aren't trying, and when it comes to the ol' college try, Meta definitely gets an A for effort. In its latest endeavor, Meta unveiled Tiramisu, a new headset prototype that it says is the closest it's gotten to 'passing the visual Turing Test.' In nerd lingo, that means its headset is inching towards tricking the eyes into thinking that a virtual world is actually a real one. Without using this very experimental headset for myself, it's hard to say whether that's true or not, but if we're to believe Tiramisu (a fine dessert, I may add) is actually as advanced as Meta says it is, the future of VR may be less cooked than we thought. Meta Reality Labs just showed their new VR headset prototype called Tiramisu which offers stunning realism with 90 PPD, ultra-high contrast and brightness 14x that of Quest 3. It's bulky with a narrow FOV but offers a glimpse of what could be visually possible in the future. — Nathie (@NathieVR) August 7, 2025According to Meta, Tiramisu blows the Meta Quest 3 out of the water in a few key categories. One of them is contrast, where Meta says the Tiramisu has 3x the ability of the Quest 3. Another is brightness, and Meta says that Tiramisu has a whopping 14x ability in that arena compared to the Quest 3. The last, which is more esoteric, is called 'angular resolution,' and that helps users in VR see finer details in virtual worlds. In this last category, Tiramisu has 3.6x the ability of the Quest 3. On paper, that's a pretty wild improvement over what we have available to us in this present moment, but (and here comes the approach back to Earth) there's still a ways to go before any of this comes to a headset near you. As you may have noticed by the pictures Meta provided, Tiramisu is—how do I say this—a big, chonky boy. It's the long Furby of VR headsets, which definitely puts a damper on the whole super-immersion thing. I don't know about you, but if I'm wearing an oversized trough on my head, that might break the illusion of a hyperrealistic VRChat sesh with my buds. On top of that, Meta says the FOV of its groundbreaking experimental headset is also (contrary to the size of the headset) very small. In fact, it's only 33 x 33 degrees. To put that in perspective, the human eye typically sees about 200 degrees with a vertical FOV of 135 degrees. As you may have noted with some back-of-the-napkin math, that's a pretty big difference. But this is all experimental stuff, to be fair, and the fact that Meta can achieve those improvements even with a very, very small window of visual information is exciting in and of itself. There are tons of hurdles to translating that research into something commercially viable—on top of the size, Meta is using glass optical lenses as opposed to plastic, which I'm going to go ahead and assume are a f**kton more expensive. Big and expensive haven't exactly translated well in the VR world; just ask Apple. I started this week by declaring that VR is cooked, and maybe it still is, but that doesn't mean I want it to be cooked. As much as I'm excited about the emergent future of smart glasses, I still think endeavoring to create a more immersive face computer is not just fun, but potentially groundbreaking. If Tiramisu is any indication, we've still got a while until we arrive at that hyperrealistic future, but I think I speak on behalf of all the other VR nerds out there when I say, I'd like some dessert eventually.
Android Authority
27 minutes ago
- Android Authority
OpenAI's most powerful AI model is here and free for everyone
Edgar Cervantes / Android Authority TL;DR OpenAI has announced the launch of GPT-5. The new AI model offers improvements across the board, delivering better accuracy, reduced hallucinations, faster performance, and more. GPT-5 is rolling out today to Plus, Pro, Team, and free users today, with Enterprise and Edu subscribers gaining access in one week. In July, OpenAI CEO Sam Altman confirmed that the company's highly anticipated new AI model, GPT-5, was nearing release. A report later claimed that the model was on track for an August launch. This week, the company posted on X (formerly Twitter) to tell users that a livestream would happen on Thursday, August 7. It turns out that the livestream ended up being the announcement for the rollout of GPT-5. Today, OpenAI is starting to roll out GPT-5 to all of its users. In a rather lengthy blog post, the company reveals that the model is available today for Plus, Pro, Team, and free users. However, Enterprise and Edu subscribers will have to wait one week before gaining access. There are three new models in total: GPT-5, GPT-5 mini, and GPT-5 Pro. OpenAI notes that Plus subscribers will be able to use GPT-5 significantly more than free users. When free users reach their limit, they'll be transitioned to a smaller, faster model called GPT-5 mini. Meanwhile, Pro members will be able to use the model as much as they want and will have exclusive access to GPT-5 Pro, a version with extended reasoning capabilities. According to the company, GPT-5 is a reasoning model, which would be the first of its models to offer this feature by default. If you're unfamiliar with the term 'reasoning model,' it means that the system can break down complex problems into smaller parts and work its way to an answer step-by-step. Previously, you would have to get ChatGPT to use one OpenAI's reasoning models by selecting the 'Think Longer' option in the prompt bar. In addition to being a reasoning model, the AI is said to boast improvements across the board compared to earlier GPT models. Specifically, this AI model offers better accuracy, experiences fewer hallucinations, answers questions faster, is safer, and is more honest. For example, the company claims that GPT-5 responses are 45% less likely to contain a factual error than GPT‑4o, and that number climbs up to 80% compared to OpenAI o3. In terms of honesty, it won't lie about completing a task, be overly confident about an uncertain answer, and sycophancy has been minimized. It also includes a new feature called Safe Completions, where GPT-5 will try to give the most helpful answer within safety confines. Previously, OpenAI used a different approach where the model would either comply with or refuse a prompt. On top of the GPT-5 news, OpenAI announced some new customizations are rolling out for ChatGPT. The chatbot will have four new personalities that will be available initially for text chat, with Voice coming later. These four personalities include Cynic, Robot, Listener, and Nerd. They are available as part of a research preview and can be changed or disabled at any time. Follow
The Verge
28 minutes ago
- The Verge
OpenAI gets caught vibe graphing
During its big GPT-5 livestream on Thursday, OpenAI showed off a few charts that made the model seem quite impressive — but if you look closely, some graphs were a little bit off. In one, ironically showing how well GPT-5 does in 'deception evals across models,' the scale is all over the place. For 'coding deception,' for example, GPT-5 apparently gets a 50.0 percent deception rate, but that's compared to OpenAI's smaller 47.4 percent o3 score which somehow has a larger bar. Or this one, where one of GPT-5's scores is lower than o3's but is shown with a bigger bar. In this same chart, o3 and GPT-4o's scores are different but shown with equally-sized bars. That chart was bad enough that CEO Sam Altman commented on it, calling it a 'mega chart screwup.' An OpenAI marketing staffer also apologized for the 'unintentional chart crime.' OpenAI didn't immediately respond to a request for comment. And while it's unclear if OpenAI used GPT-5 to actually make the charts, it's still not a great look for the company on its big launch day — especially when it is touting the 'significant advances in reducing hallucinations' with its new model. Posts from this author will be added to your daily email digest and your homepage feed. See All by Jay Peters Posts from this topic will be added to your daily email digest and your homepage feed. See All AI Posts from this topic will be added to your daily email digest and your homepage feed. See All News Posts from this topic will be added to your daily email digest and your homepage feed. See All OpenAI
