For his part, Wang wrote in a post on X Thursday that Meta is “truly only investing more and more into Meta Superintelligence Labs as a company. Any reporting to the contrary of that is clearly mistaken.”

---

In Name Only

Sure, it’s possible to look at this “basic organizational planning” as a normal and necessary reorientation after an expensive hiring spree. But the reality is that Zuckerberg has been scrambling for months, and that Meta is still confronting many of the same challenges that it was in April. Indeed, the kaleidoscopic shifts in focus from FAIR to AGI Foundations and now to Meta Superintelligence Labs says a lot about where the industry is now, and where it’s headed.

As I reported earlier this month, OpenAI’s underwhelming GPT-5 release was a moment of reckoning for both Silicon Valley and Wall Street—an overdue reality check on the feasibility and timeframe for achieving A.G.I. or superintelligence (both terms lack agreed-upon definitions and are basically interchangeable, although the industry has recently begun to favor the latter). But nothing about Meta’s restructuring seems to offer a novel path toward achieving superintelligence. In his note, Wang suggested that Meta’s roadmap involves “scaling large models to achieve superintelligence across pre-training, reasoning, and post-training.” In short, basically what every other company has been doing—relying on L.L.M.s alone to reach the promised land. (He did note that the team will explore “omni”—another word for multimodal—models, or models that can generate text, images, and video.)

And yet, over the past several years, multiple studies have indicated that L.L.M.s are probably not up to the task. It’s something that LeCun—who has explicitly argued that L.L.M.s do not provide a path toward A.G.I., and that he “hates” that term—has been trumpeting for months. Earlier this year, he dismissed L.L.M.s as mere token generators. “I am more interested in next-gen model architectures,” he said at an Nvidia event. These models should be able to do four things, per LeCun: “understand the physical world, have persistent memory, and ultimately be more capable to plan and reason.” One irony of the restructuring is that LeCun will now, apparently, be more involved in Meta’s effort to achieve superintelligence, which seems predicated on doing exactly what LeCun warned against, which is to just keep scaling up L.L.M.s in the hopes that, with enough compute, the Rubicon will be crossed. (I’m sure that was a fun Zoom.)

Of course, this is not Zuckerberg’s first moon shot—remember the Metaverse?—but this time, the bar for success is even less clear. Meta makes virtually all of its money from advertising on social and messaging platforms, and Zuckerberg seems anxious to control future commercial platforms, too—hence his interest in virtual reality, and now superintelligence. Theoretically, the technology could transform wearable products into a viable commercial platform, too. After all, Meta’s investment in virtual reality and smart glasses is set to top $100 billion this year. But that seems like a rather lackluster endgame for a technology Zuckerberg has called “one of the most important innovations in history.” Perhaps, with all of his rivals engaged in the same arms race—and Sam Altman talking about building a social network—he simply doesn’t want to get left behind.

Alas, right now, being able to brand Meta as at the forefront of A.I. research might be as important for the company’s stock price and talent retention as actually building a viable commercial product. And as long as Zuckerberg’s A.I. push doesn’t distract from the company’s core competency—generating tens of billions of advertising dollars every quarter—investors will likely stick around for the ride. Who knows, maybe in another six months, Meta Superintelligence Labs will be renamed again—with the promise that ultra-mega-intelligence is within reach.