Meta has turned the tables on the traditional approach to artificial intelligence, releasing its advanced language model, Llama 3.1, to the public for free. Unlike most tech companies, which keep their AI models behind closed doors and proprietary, Meta is making its cutting-edge technology available for developers and researchers to build upon. 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.
Mark Zuckerberg, Meta’s CEO, has compared the release of Llama 3.1 to the rise of Linux, an open-source operating system. Just as Linux eventually outpaced its proprietary counterparts, Zuckerberg predicts that open-source AI will eventually overtake closed models. By releasing Llama 3.1, Meta is positioning itself as a major player in the AI landscape, while also fostering innovation and collaboration among developers and startups.
The model, which is Meta’s biggest and most capable release to date, boasts 405 billion parameters, making it as powerful as commercial offerings from companies like OpenAI, Google, and Anthropic. Meta trains Llama 3.1 with safeguards in place to prevent it from generating harmful output, but developers can choose to remove these safeguards if they prefer.
Percy Liang, an associate professor at Stanford University, believes that developers may find Llama 3.1 to be as capable as the best commercial AI models, potentially leading many to switch to Meta’s offering. However, he notes that Meta’s release of Llama 3.1 is not entirely altruistic, as it also maintains restrictions on commercial usage and requires developers to adhere to certain guidelines.
Despite these limitations, Liang suggests that Llama 3.1 will still have a significant impact on the AI industry, driving innovation and experimentation. Meta is making the model accessible through various hosting options, including those offered by Databricks, Groq, AWS, and Google Cloud, as well as through Meta.ai.
