Open Source AI was the Path Forward

Almost exactly a year ago, Mark Zuckerberg took to Meta's blog to write a long post entitled "Open Source AI is the Path Forward".¹ It was a good articulation of why Meta was choosing to go down the "open source" (read: open weight) path with AI even when their main rivals were going in a different direction. It was in the midst of a press push he did around the topic – a clear call-t0-arms to try to rally the industry around Llama.

What a difference a year makes.

As Eli Tan reports for The New York Times:

Meta’s newly formed superintelligence lab has discussed making a series of changes to the company’s artificial intelligence strategy, in what would amount to a major shake-up at the social media giant.

Last week, a small group of top members of the lab, including Alexandr Wang, 28, Meta’s new chief A.I. officer, discussed abandoning the company’s most powerful open source A.I. model, called Behemoth, in favor of developing a closed model, two people with knowledge of the matter said.

Per Tan's reporting, this isn't a done deal yet and there seems to be a debate going on internally about it. The fact that Kalley Huang of The Information backs up this report might suggest that some of those factions may be attempting to use strategic leaks to pressure Meta not to move away from this ethos.

Of course, for their part, Meta will spin this regardless of which way they go. The gaslight-y statements from their PR team on the matter already suggest as much:

“Our position on open source AI is unchanged,” a spokesperson for Meta said in a statement. “We plan to continue releasing leading open source models. We haven’t released everything we’ve developed historically, and we expect to continue training a mix of open and closed models going forward.”

Ah yes, the old, "Oceania had always been at war with Eastasia" approach. Nothing to see here. Nothing at all.

Obviously, that's bullshit. If Meta's new very high profile and very highly paid new Superintelligence Lab goes down the closed model path, it will be a massive change in AI strategy from the one Zuckerberg outlined a year ago. And it also, perhaps, would be the right move.

A couple weeks back, while thinking through the ramifications of Meta's AI reset, I wrote a post entitled, "Meta's Open Source AI Mistake". As I wrote:

All we've heard for months and months from Zuckerberg is how the open sourcing of Llama is what was fueling Meta's AI growth and would ensure they won the market. We kept hearing how it would play out similarly to how it did with the open sourcing of some of their database infrastructure. That the work of the masses, building on Meta's Llama base, would propel their AI to heights that the competition couldn't match and make it the industry standard.

Again, that has not happened. It doesn't seem fair to blame open source for that – hell, Llama is not even really "open source", but "open weight" as everyone is well aware by now – but it seems clear that it hasn't really helped. At least not in the ways that Zuckerberg and Meta had hoped.

In fact, it may have hurt in at least one way that Meta clearly failed to anticipate: that other model-makers would be able to distill Llama in ways that make perhaps better models than Llama itself as DeepSeek may have done.

The open source strengths that Zuckerberg touted, turned out to perhaps be weaknesses when it came to competing at the highest end of the market. It doesn't mean open source (again, open weight) is bad – it just means Meta's strategy here may have been flawed. And I suspect where they'll net out is the same strategy that Anthropic, Google, and soon OpenAI are doing. That is, keep the cutting-edge models closed and open source older and smaller models more selectively.

I went on to compare this strategic mistake to the one Meta – then still called Facebook – made almost 15 years ago with HTML5. This is the call that Zuckerberg would go on to say was his "biggest mistake". As I wrote:

Then, as now, they thought they were getting ahead of an inevitable curve. But the turn never happened. That's not to say that it won't with open source models, it's just that trying to build your own AI business around those while allowing others to do the same as you incur the cost may not be the way to win in the market right now.

Meta had thought that others would step in to help with such a heavy lift. And even apparently went around to partners holding out their donation cup. No takers, sadly. Because why pay for the llama when you get the model for free? Meanwhile, the stance seemingly hasn't helped them bring in the right talent for the job. After some early promise, Llama 4 is clearly a mess and that's in part because a lot of people who started the project no longer work at Meta.² OpenAI has somehow been able to handle such turnover, clearly Meta has not.

All eyes remain on Yann LeCun, Meta's face of AI before Meta paid $15B to bring Alexandr Wang and Nat Friedman on board. As Tan notes:

Any move toward a closed A.I. model would be a philosophical change at Meta as much as a technical one. Meta has won plaudits from developers for open sourcing its A.I. models, and one of its top A.I. executives, Yann LeCun, had said, “The platform that will win will be the open one.” This year, the Chinese A.I. company DeepSeek released an advanced A.I. chatbot thanks in part to Meta’s open source code.

LeCun keeps retweeting posts about open source AI over on Xitter. While on Meta-owned Threads, he's much more of a company man. "Big plans" is his most recent post, quoting Zuckerberg's post about the new "superintelligence" efforts. Zuck doesn't mention open source.

One more thing: About a year ago, amidst his press flurry around open source AI, I gave a note of caution:

But if Zuckerberg takes this whole 'open vs. closed' stance too far, he'll seem disingenuous or worse, hypocritical. I'm old enough to remember when this was the approach Google tried to take against Apple in the earlier days of the Android vs. iPhone wars. It started out sounding like a "winning" thing to say, but the reality was far more complicated and nuanced.

What happens if, say, Meta has to (or decides to) shift their stance on the openness of Llama down the road? Such things have a funny way of happening over time...

Indeed.

¹ You know, the Meta blog that resides at 'about.fb.com/news'...

² Many have taken their open source AI talents to Mistral...