Most tech moguls hope to sell artificial intelligence to the masses. But Mark Zuckerberg is giving away what Meta considers one of the world’s best AI models for free.
Meta on Monday released the largest, most capable version of a large language model called Llama, for free. Meta has not disclosed the cost of developing Llama 3.1, but Zuckerberg recently told investors that his company is spending billions on AI development.
With this latest release, Meta shows that the closed approach favored by most AI companies is not the only way to develop AI. But the company also places itself at the center of the debate over the dangers of releasing AI without controls. Meta trains Llama in a way that prevents the model from producing malicious output by default, but the model can be modified to remove such protections.
Meta says Llama 3.1 is as smart and useful as the best commercial offerings from companies like OpenAI, Google, and Anthropic. In certain benchmarks that measure progress in AI, Meta says its model is the smartest AI on Earth.
“It’s very exciting,” says Percy Liang, an associate professor at Stanford University who follows open-source AI. If developers find the new model as capable as the industry’s leading models, including OpenAI’s GPT-4o, Liang says, many could switch to Meta’s offering. “It’ll be interesting to see how usage shifts,” he says.
In an open letter posted on the release of the new model, Meta CEO Zuckerberg compared Llama to the open-source Linux operating system. When Linux took off in the late 1990s and early 2000s, many big tech companies invested in proprietary alternatives and criticized open-source software as risky and unreliable. Today, however, Linux is widely used in cloud computing and is at the heart of the Android mobile OS.
“I believe AI will evolve in a similar way,” Zuckerberg writes in his letter. “Today, several tech companies are developing leading closed models. But open source is rapidly closing the gap.”
Meta’s decision to give away its AI isn’t without its own self-interest, however. Previous releases of Llama have helped the company gain a foothold in the AI research community, developers, and startups. Liang also notes that Llama 3.1 isn’t truly open source, as Meta imposes restrictions on its use, such as limiting the scale at which the model can be deployed in commercial products.
The new version of Llama has 405 billion parameters or customizable elements. Meta has already released two smaller versions of Llama 3, one with 70 billion parameters and another with 8 billion. Meta has also released upgraded versions of these models today under the name Llama 3.1.
Llama 3.1 is too large to run on a regular computer, but Meta says that many cloud providers, including Databricks, Groq, AWS and Google Cloud, will offer hosting options that will allow developers to run customized versions of the model. The model will also be accessible via Meta.ai.
