Meta Platforms, the parent company of Facebook, is making waves in the race to build the best artificial intelligence (AI) models with its unique and differentiated approach. While competitors like OpenAI, Alphabet’s Gemini, and AI start-up Anthropic are also developing top-performing models, Meta’s open-source strategy sets it apart.
Last week, Meta celebrated the release of its Llama 3.1 model, boasting an impressive 405 billion parameters. This marks a significant leap from Meta’s previous models, which had 8 billion and 70 billion parameters. Llama 3.1 outperforms many advanced large language models (LLMs), including ChatGPT 4-Omni and Anthropic’s Claude 3.5 Sonnet, on several parameters. Moreover, Meta claims that inferencing the Llama 3.1 model costs only half as much as running ChatGPT 4o.
What makes Llama even more unique is that it is the only open-source frontier model in the market. Open-source software allows outside developers to freely access and modify the source code, enabling them to improve functionality, fix bugs, and enhance security. This approach not only accelerates product improvement but also facilitates rapid scalability. Additionally, Llama’s versatility allows developers to run it anywhere, eliminating the need to send private data to closed models or specific clouds.
Meta’s CEO, Mark Zuckerberg, envisions Meta AI becoming the world’s most-used AI assistant by the end of the year, surpassing OpenAI. In a blog post, Zuckerberg drew parallels between the development of LLMs and the evolution of operating systems like Linux and Unix. He believes that Meta’s open-source model, similar to Linux, will eventually become the standard for AI.
While Meta may not generate direct revenue from Llama in the near term, the company plans to build services on top of the underlying model. Zuckerberg hinted at potential business messaging customer service AIs on WhatsApp, introducing ads into AI interactions, or charging for access to larger AI workloads. Other open-source software companies, such as Red Hat and WordPress, have successfully monetized their models through customer service, consulting services, and deluxe paid versions.
However, Meta’s open-source approach also poses risks, particularly concerning privacy and security. The company has faced criticism in the past regarding these issues, which may have contributed to the breakdown of talks with Apple to include Llama in the new Apple Intelligence platform. Apple, known for prioritizing privacy and security, may have reservations about opening up the source code and model weights to potential bad actors.
Zuckerberg argues that even in a closed system, countries adept at espionage, such as China, Russia, or Iran, would likely gain access to leading AI models. By embracing open innovation, he believes the U.S. and its allies can maintain a competitive advantage in AI development. However, this argument may face scrutiny from the U.S. government, which may be concerned about the potential misuse of open-source AI models.