Elon Musk threw nine-figure promises at top AI researchers

Elon Musk sought to recruit top AI talent to his new artificial intelligence company with equity in the venture he claimed was already worth tens of billions of dollars, according to people familiar with the matter.

If xAI was worth $20 billion — a valuation Musk came up with — before it was born, then each 1% offer in stock options was like a $200 million signing bonus, the people said.

His recruiting efforts succeeded. The startup, launched Wednesday, boasted researchers from Google, DeepMind, and academic powerhouses like the University of Toronto, where major breakthroughs in deep learning have taken place.

Musk and his team plan to reveal more details about xAI during a Twitter Spaces chat Friday afternoon. He didn’t respond to a request for comment.

Musk previously said that he wants to create his own competitor to ChatGPT, which he told Fox News would be called “TruthGPT.” But a number of the high-profile researchers he hired at xAI are excited about something else: building an artificial intelligence capable of reasoning about the hardest math problems in the world.

Before he co-founded xAI, Christian Szegedy led a Google research project that sought to create an “automated mathematician.” His former co-worker Yuhuai “Tony” Wu, who also joined Musk, previously told The New York Times he wants to use machine learning to “solve mathematics” by creating an AI that can decipher theorems on its own.

Greg Yang, another mathematician who joined xAI, hinted that it planned to create an AI that could help humanity better understand the mathematical laws of the universe. “​​Any mathematician/theorist excited about this needs to DM me!” he tweeted.

For decades, computers have helped mathematicians recognize patterns, making their work more efficient. But AI is not yet capable of the reasoning required to solve advanced theorems, said Alex Kontorovich, a mathematics professor at Rutgers University.

That’s partially because there currently isn’t enough data available for an AI mathematics model to train on. For a computer to process a theorem, it needs to be translated from natural language into a formalized type of code that can be verified as logically sound. Kontorovich estimates there are around 10 million lines of this data, whereas ChatGPT was trained on trillions of lines of human language.

xAI could help solve the problem by creating its own database of formalized mathematics, or it could contribute to existing open source projects like Lean, which is led by Amazon researcher Leo de Moura.

Whatever Musk’s public relations issues are at Twitter, he remains able to recruit at the very highest level of technical talent. And there is some precedent for his ability to deliver. In 2015, he recruited a similarly elite group of researchers and co-founded OpenAI, which is now worth close to $30 billion. But Musk left before OpenAI turned a corner and prioritized the large language model research that led to ChatGPT, as Semafor first reported.

For xAI, the pressure to compete is intense. OpenAI’s ChatGPT kickstarted a wave in artificial intelligence development, spurring the world’s biggest companies into action and leading to a shortage of the souped-up Nvidia graphics processors used to train AI models.

Musk’s firm appears eager to leapfrog those deep-pocketed competitors. One industry insider told me Musk asked Nvidia to sell him 40,000 GPUs for training AI models. The company instead promised him 20,000 over 12 months.

Musk’s goal is to create artificial intelligence that can “understand the true nature of the universe,” according to xAI’s web site. The bet is that the mysteries that science can’t solve — the inner workings of the human brain, for instance — are equations beyond the grasp of human understanding.

By setting a goal that sounds a lot like science fiction, who needs a business model? Unlocking the mysteries of the universe would mean you could, for instance, simply prompt it to patent every new technology that will ever be created and charge licensing fees for all eternity.

But there’s one wrinkle in this plan. It might take some time to unlock the secrets of the universe with AI. Musk can be a demanding boss, and while he has been recruiting talented researchers, Google CEO Sundar Pichai and others have been doing the same thing.

If history is any judge, Musk won’t be content to let his team of elite eggheads sit around in an ivory tower coming up with lofty theories. They’ll need to perform, or things will go the way of OpenAI, with Musk shutting it all down or walking away, leaving promises unkept.

I asked a lot of tech people about xAI this week and there’s a duality when it comes to views of Musk, his accomplishments, and prospects for his future success.

I don’t know anyone who thinks Musk’s often juvenile tweets serve any purpose other than to entertain the world’s sometimes richest man and his loyal followers. And I don’t think anyone — including Musk — believes buying Twitter was a good business decision.

But he has had one of the greatest entrepreneurial careers in history, having helped create PayPal, found SpaceX, and built Tesla into what it is today (Not to mention founding OpenAI and yet-to-be-proven companies like Neuralink and The Boring Company).

And there are two ways of looking at where he stands now. One is that he’s lost it, as one veteran entrepreneur put it to me off the record, and his Twitter foibles marked the end of his great run.

Another is that Twitter is a distraction, but is ultimately a meaningless outlier on his record. By that logic, his ability to recruit talent and build new companies is limited only by his will and attention.

Looking at it through that lens, xAI fits nicely into the broad array of companies he controls. Tesla relies on AI for its autonomous driving features and is also building a general purpose robot. Neuralink, which makes computer brain implants, is adjacent to AI in the sense that it is gathering data from the most complex device we know of: the human brain. And SpaceX employs complex automation and will undoubtedly continue down that path.

Wired Staff Writer Will Knight argues that Musk’s new startup may have bit off more than it can chew. “The huge goals Musk has set for himself — challenging existing AI giants and protecting humanity from harmful AI — make his tiny new AI company look even smaller,” he wrote.”Many AI researchers who are also concerned about the trajectory of AI seem to view the problem as one that requires greater transparency and collaboration, rather than a lone genius with a small band of all-stars.”

• Science journalist Siobhan Roberts chronicled how mathematicians are grappling with the ways artificial intelligence is disrupting their field in The New York Times earlier this month.
• Columbia University math professor Michael Harris invited readers of his Silicon Reckoner newsletter to disclose the starting salaries Musk was offering recruits at xAI. “Anonymity guaranteed!” he wrote.