For Meta Platforms, An Open AI Policy Is The Best Policy

For Mark Zuckerberg, the decision by Meta Platforms – and way back when it was still known as Facebook – to open much of its technology – including server and storage designs, datacenter designs, and most recently its Llama AI large language models – came about because the company often found itself trailing competitors when it came to deploying advanced technologies.

“We’ve done a lot of open source work over time,” said Zuckerberg, Meta’s founder and chief executive officer. “I think part of it, just bluntly, is we got started after some of the other tech companies in building out stuff like the distributed computing infrastructure and the datacenters. Because of that, by the time that we built that stuff, it wasn’t a competitive advantage. We’re like, ‘Alright, we might as well make this open and then we’ll benefit from the ecosystem around that.’ So we had a bunch of projects like that.”

Thus was born the Open Compute Project in 2011, founded by Facebook as well as Intel and Rackspace Hosting, as a way to standardize datacenter infrastructure designs on the company’s hardware configurations. In subsequent years, that open source push by Zuckerberg included infrastructure software tools like React, for creating interactive UIs, and PyTorch, a machine learning framework. Given all of this, opening the Llama LLM for generative AI workloads, which first appeared in early 2023, was a natural step.

“By the time that Llama came around, we were sort of positively predisposed towards doing this,” he said.

Sharing The Stage With Nvidia’s Jensen Huang

Zuckerberg’s comments came during his widely hyped talk with Nvidia founder and chief executive office Jensen Huang at this week’s SIGGRAPH 2024 show in Denver. Meta and Nvidia have worked closely together as the former has rolled out the various iterations of the Llama LLM.  Llama 3.1, which was introduced last week, is the latest rev. Much of the hour-long discussion was focused on the drivers behind Zuckerberg’s lean into open sourcing some of Meta’s technologies and what it means to the industry to have a major open LLM at a time when most other generative AI models are proprietary, such as OpenAI’s GPT and Gemini’s Gemini, just to name two.

Huang pointed to Meta’s release of Llama 2 in June 2023 as “the biggest event last year,” giving organizations in every industry and of any size the tools to create their own AI models based on their own specific needs. It came out in three models – 7B, 13B, and 70B parameter sizes – was trained on 40 percent more data than its predecessor, and doubled the context length.

“The reason why it was the biggest event was because when that came out, it activated every company, every enterprise, and every industry,” Jensen said. “All of a sudden, every healthcare company was building AI. Every company was building AI. Every large company, small companies, startups – were building AIs. It made it possible for every researcher to be able to reengage AI again, because they have a starting point to do something with.”

Llama 3.1 swamps Llama 2, with models in 8B, 70B, and 405B parameter sizes, the latter being what Meta is calling the first open “frontier-level” model that can outperform OpenAI’s GPT-4 and GPT-4o and Anthropic’s Claude 3.5.

Maybe They Should Have Called It Open IT, Not OCP

The first step in Meta’s open technology push was the OCP, which showed Zuckerberg and other company executes that the benefits could reach beyond Meta – at the time Facebook – itself.

“We took our server designs, network designs, and eventually the datacenter designs and published all of that,” Zuckerberg said. “By having that become somewhat of an industry standard, all the supply chains basically got organized around it, which had this benefit of saving money for everyone. By making it public and open, we basically have saved billions of dollars by doing that.”

He also noted the changing nature of the tech industry during Facebook’s lifetime. When the company first formed, the first version of Facebook was open and on the web. Then it transitioned to mobile devices, which put individuals’ computers – their smartphones – and Facebook’s app in their pockets, making them constant companions. However, for Facebook executives, it meant having to bend to the will of Apple and Android device makers. And Apple and its closed system set the rules.

But it hasn’t always been like that.

“Apple was doing their kind of closed thing,” Zuckerberg said. “But Microsoft, which obviously isn’t like this perfectly open company, but compared to Apple, with Windows running on all the different OEMs and different hardware, it was a much more open ecosystem. Windows was the leading ecosystem. In the PC generation of things, the open ecosystem won.”

Zuckerberg is hopeful in this new AI era that open will continue to be the dominant model, though there always will be closed environments.

“There are benefits to both,” he said. “I’m not a zealot on this. We do closed source stuff and not everything that we publish is open, but I think, in general, for the computing platforms that the whole industry is building on, there’s a lot of value for that if the software, especially, is open.”

Looking Across To The Open Horizon

Meta is not only doing that with the Llama LLMs, but also with its Horizon OS, an operating system for mixed reality systems like its Meta Quest headsets. The goal is to create it in the fashion of Windows and Android, making it able to work with multiple hardware companies and create an open ecosystem for the virtual and augmented reality realms, Zuckerberg said, adding that he is “pretty optimistic that in the next generation, the open ones are going to win.”

“This is sort of selfish, but after building this company for a while, one of my things for the next ten or fifteen years is that I just want to make sure that we can build the fundamental technology that we are going to be building social experiences on, because there have been just too many things that I’ve tried to build and then have just been told, ‘Nah, you can’t really build that’ by the platform provider.”

With Llama 3.1, the open nature of the LLM includes a reference system, components like Llama Guard 3 to more easily detect bad content and cyberattacks, and plans for a so-called “Llama Stack” of APIs to better enable third party developers to build software for the LLM. Such features, along with expanding partnerships with tech and other companies to offer services based on Llama, will pull the LLM along.

“We’re building it because we want the thing to exist and we want to not get cut off from some closed model,” he said. “But this isn’t just like a piece of software that you can build. You need an ecosystem around it. It almost wouldn’t even work that well if we didn’t open source it. We’re not doing this because we’re kind of altruistic people, even though I think that this is going to be helpful for the ecosystem. We’re doing it because we think that this is going to make the thing that we’re building the best by having a robust ecosystem.”

Also, when something becomes an industry standard, other companies work with it, with the silicon and systems being optimized to run the models well, which benefits everyone, Zuckerberg said, adding that “that’s just one example of how this ends up being really effective. I think that the open source strategy is going to be a good one as a business strategy. I think people still don’t quite get it.”