Meta just launched its latest cutting-edge AI model, and CEO Mark Zuckerberg is touting its open source capabilities.
The race to build the best artificial intelligence (AI) model is definitely on, with current leader OpenAI set to make waves in the second half of 2022 and up-and-coming competitors alphabetof (Google -0.28%) (Google -0.17%) Gemini and AI startup Anthropic are also building top-performing models with hundreds of billions of parameters.
But investors Meta Platform (Meta 2.71%)has the most unique and differentiated approach to building generative AI models.
Will this approach enable Meta to lead the industry, or will the risks outweigh the benefits?
When competitors close, Meta opens up
Last week, Mark Zuckerberg and Meta celebrated the release of Meta’s Llama 3.1 model, which features 405 billion parameters.
That’s a large model. Until now, Meta had only released Llama models with 8 billion and 70 billion parameters. But Llama 3.1 405B is Meta’s first foray into “frontier” models, pushing the limits of the most advanced generative AI large language models (LLMs) on the planet. Meta says the new model outperforms the most advanced LLMs, such as ChatGPT 4-Omni and Anthropic’s Claude 3.5 Sonnet model, on many (but not all) parameters. Zuckerberg also claims that inference costs for Llama 3.1 are just 50% of what it costs to run ChatGPT 4o.
Not only does Llama perform as well or better than many of the leading models currently on the market, it also has the distinction of being the only open source frontier model.
The benefits of open source
Open source software is software whose license holders allow outsiders to freely access and modify the software’s source code, allowing external developers to make changes to the software to improve its features, fix bugs, or strengthen its security.
By distributing software code for free, developers have the potential to improve their products more quickly than with “closed” software systems where only a company’s employees can modify the code. External developers prefer “free” open source software over expensive proprietary software, so the open source model is a great choice if you want to scale usage quickly. In addition to these general open source benefits, Llama can run anywhere, so developers don’t need to send private data to a closed model or specific cloud.
Because of this, Mark Zuckerberg believes Meta AI will overtake OpenAI by the end of this year to become the world’s most used AI assistant.
In a blog post accompanying the release, Mark Zuckerberg said LLM’s development is proceeding in a similar way to the development of operating systems like Linux and Unix in the 1990s: Unix, a closed system, initially took the lead, but the benefits of Linux’s open source model eventually paved the way for it to become the standard operating system for cloud computing and mobile devices.
Zuckerberg also noted that Meta has benefited from open source tools in the past, such as the development of its open source data center architecture and AI software, such as Pytorch, which Meta first developed. Because Meta’s core business is advertising, not directly monetizing its software or data center architecture, open sourcing did not impede its ability to generate revenue and invest in research and development. On the other hand, Meta ultimately saved billions of dollars by developing these tools with outside help and using them to run its core social media platform.
The same is true for Llama AI: Meta is not a software provider or a public cloud provider, so it doesn’t need to sell models directly to generate revenue.
But Meta should benefit from Llama… eventually.
That’s not to say Meta doesn’t expect to eventually make money from Llama, but it’s more likely to do so by using the underlying model and building services on top of it. On a recent Q1 call with analysts, Zuckerberg hinted at building a business messaging customer service AI on WhatsApp, introducing ads into Meta AI’s AI conversations, or even charging for access to the biggest AI workloads with more compute.
Other open source software companies such as Red Hat have previously monetized AI models by selling customer and consulting services, and WordPress uses a dual-licensing model to monetize its website building software, offering a “deluxe” paid version to large enterprises, so there is precedent for profiting from the open source model.
While Meta won’t be deriving any direct revenue from Llama in the near future, it should begin to benefit from rolling out ancillary services built on top of the underlying Llama model over the next few years.
Will Meta Platforms win the AI war? Image source: Getty Images.
Disadvantage
But open source also comes with risks. especially In the context of AI, these risks relate to privacy and security, which Meta has been criticized for in the past. Privacy and security concerns are likely why negotiations with Meta fell through. apple (AAPL 0.22%) Plans to incorporate Llama into the new Apple Intelligence platform fell through earlier this year. Apple has perhaps the most closed business model in the tech industry and has always placed a premium on privacy and security, so its disagreements with Meta and its exclusion from Apple Intelligence could be a red flag.
Specifically, there may be resistance to releasing the source code and model weights for public viewing, as this could allow malicious actors to modify the code for nefarious purposes, and there is a risk of intervention by government authorities, which could limit the distribution of Llama to prevent it falling into the hands of hostile governments such as China, Russia, or Iran.
Zuckerberg’s answer is that those countries are good at espionage, and even if they have closed systems, they will likely have access to cutting-edge models anyway. This means that there is a risk that only a few “good guys” (only a few large companies) will have access to the top models, to the disadvantage of the United States. Moreover, Zuckerberg believes that stifling open innovation risks the United States and its allies not having the best AI at all.
But that seems like a vague argument: Bad actors will gain access to cutting-edge AI models anyway, so why not just give the code away for free? It’s unclear whether this argument will actually sway the U.S. government to go along with that thinking.
A major factor in the AI race
Meta’s open-source approach to AI could position it to become one of the world’s leading AI companies and potentially one of the most valuable in the world in the future, but it also risks government oversight and regulation, threatening its very ability to compete in the AI wars.
For Meta investors, and investors in AI technology in general, the rollout of Llama 3.1 and the government’s potential response to it in an election year will be a key point to watch in the coming months.
