Meta hired Alexandr Wang to build the AI. Zuckerberg’s job is to sell it.

One year after spending more than $14 billion to bring in Alexander Wang and his group of top-scale AI engineers to revamp its artificial intelligence efforts. meta It’s back on the map, at least in the AI ​​space, but still OpenAI, Anthropic, google At the market.

Wang’s big accomplishment was delivering the Muse Spark AI model in April. This marks Meta’s first foray into a proprietary underlying model, moving away from strict adherence to open source or open weight, as it is commonly referred to in AI. Wang’s group, Meta Superintelligence Labs, was created to energize the company in the hottest areas of the technology industry.

CEO Mark Zuckerberg has a new model and it’s up to him to make it financially successful. This means the company can demonstrate that its AI tools can be used to attract paying users, not just to enhance and enhance its core advertising business.

“Meta needs to provide more evidence of both adoption and commercialization,” said William Blair analyst Ralph Schuckert, recommending the stock as a buy. “Investors are looking meta to monetize new AI-first products beyond the material positive impact AI has on enhancing advertising models.”

Wall Street, at least for now, is unimpressed. Meta’s stock price has fallen 18% over the past 12 months, making it the worst performer among the mega-cap group. microsoftAI has its own challenges. This is even after Meta reported 33% revenue growth in the first quarter, which is the fastest rate of expansion in any period since 2021.

For Mehta, the problems began with what some industry experts are calling, at least in hindsight, a strategic failure. The company entered AI with its Llama family of models, offering an open-source approach that developers were free to tinker with while other major model makers charged for access.

Last April, Meta’s release of Rama 4 was a failure, failing to attract developers and prompting Zuckerberg to rethink his company’s approach to AI development. Two months later, Mr. Zuckerberg shocked the tech world by announcing that he had invested $14.3 billion in roughly half of Scale AI and, more importantly, had brought Mr. Wang and his executives on board.

Wang developed and deployed Muse Spark in April of this year, and it took off. Rather than focus on third-party developers, the new model was designed to easily connect to Meta apps like Facebook and Instagram, as well as AI-powered devices like the Ray-Ban Meta glasses, said Thomas Randall, an analyst at Infotech Research Group. This is on top of the standalone Meta AI app and site.

“Many of these frontier model providers are going to fundamentally change in different ways, and Meta needs to have a consistent, reliable proprietary model that it owns,” Randall said. He added that Meta would be “lost” if Zuckerberg didn’t move aggressively to hire Wang and other high-profile AI talent over the past year. This is what Randall calls a “strategic restructuring” of the company.

Randall said Meta hasn’t taken the “most optimized path,” but at least it “has a vision of what they’re trying to accomplish and what Wang is trying to accomplish.”

Since the release of Muse Spark, Meta has announced new AI and business-related subscription plans as part of its efforts to expand its business beyond online advertising. Historically, that hasn’t worked. Meta still relies on advertising for 98% of its revenue.

Shuckert said he would like to see “concrete evidence that Muse Spark is creating a growing list of new AI-first products, even if monetization is delayed.” That’s “what investors are looking for,” he said.

No matter how good Wang’s model is, Zuckerberg still has a high mountain to climb with developers recovering from Llama’s failure.

“I think the AI ​​community is pretty much ignoring the meta at this point,” said Rob May, CEO of Neurometric, a startup working in the field of token engineering.

Mei said it’s difficult to gauge how much success Wang has had leading MSL because the company has released only one AI model so far, and characterized it as a “yawn” among the AI ​​community because the technology is not widely accessible.

While Meta has been actively reaching out to third-party developers with Llama, Mei said the company’s efforts under Wang appear to be aimed at internal use. Mr May said he had previously been in regular contact with Mr Mehta on Rama-related issues but was now “unable to get his messages returned”.

May acknowledged that Meta has a $200 billion-a-year business to protect, so it makes sense to focus on AI in its core advertising products.

“That company made that machine,” he said.

Andrew Moore, CEO of enterprise startup Lovelace and former head of Google Cloud AI, said it’s not too late for Meta to find its lane.

Meta focuses on making models more efficient through training techniques. Moore said this could be a big differentiator for developers concerned about the rising cost of the underlying model.

“If they develop their own computationally efficient models, it’s going to be very different from what’s happening in this battle to the death between these big guys,” Moore said. “They might really benefit.”

Moore added that the meta has to show an advantage somewhere, such as cost, latency, and other technical nuances that are important to developers.

Krish Subramanian, CEO and former head of product at consulting firm KOI AI IBM Consulting says developers are more excited about Google’s AI models than what Meta offers. Llama’s appeal was that it was specifically targeted at developers looking for an open-weight alternative model, but with Muse Spark, Meta has made little effort in that direction, he said.

Subramanian noted that it took Microsoft years to regain the trust of open source programmers in the early days of Azure, and said, “If we don’t focus on third-party developers, that lack of developer trust will come back to bite them.”

“If they just focus on a walled-garden ecosystem and advertising revenue as their main source of revenue, they’ll probably never become big,” he says.

A Meta spokesperson pointed to Wang’s recent comments about the company’s continued support for the open source ecosystem, and said Meta plans to continue to provide access to Muse Spark’s underlying technology to outside developers via APIs, as previously announced.

“We are already testing with several early partners and look forward to launching it this month,” a spokesperson said.

Adding to the challenges for developers is low morale. Meta has been cutting its workforce throughout the year, laying off about 8,000 employees in May. The cuts are across departments, including teams with roles related to trust and safety, and raise concerns about potential problems with AI development, said the people, who requested anonymity to speak candidly about the issue.

Mehta declined to comment on the layoffs. Regarding safety-related issues, the spokesperson pointed to Mr. Wang’s comments on the matter. “One of the things that’s very important to me is the safety of these models,” he said on the Core Memory podcast last month.

There are also tensions at the top of AI organizations. While the Muse Spark release was well-received internally, pressure is on for Wang and former GitHub CEO Nat Friedman, who also joined the company last summer as part of the AI ​​spending, to deliver meaningful revenue growth with this model and future releases, the people said.

Metatech head Andrew Bosworth is a 20-year veteran of the company and a confidant of Mr. Zuckerberg, and someone the CEO could take on a larger role in AI if the newcomer is deemed to be failing, the people said. Wang denied the reported internal conflict in a May podcast.

Wang called Muse Spark an “appetizer” for the future, saying more powerful and “bigger models” are coming.

But the AI ​​community is used to constant updates and new features. That’s what you get from OpenAI, Anthropic, and Google.

“What I’m concerned about is the frequency and pace of launches,” said Howard Yu, a business professor at the International Institute for Management Development in Switzerland. “When you launch something, can you build more momentum?”

Info-Tech Research Group’s Randall said it’s ultimately up to Zuckerberg to decide on that strategy and show “what a superpower they are in all of their products.”

Yuu agreed.

“This is really a leadership issue, right?” He noted that in high-tech companies, especially when they’re spending billions of dollars, the CEO defines and articulates the vision.

Mr. Zuckerberg’s metaverse and virtual reality ambitions have collectively generated more than $80 billion in losses since late 2020, making AI an even tougher sell, Mr. Yu said.

“He’s running out of space to maintain his credibility,” Yu said. “I think his foray into virtual reality has burned out a lot of his goodwill with investors.”

Attention: Meta is “tone deaf”