Meta inches toward open source AI with new LLaMA 3.1

Meta, Facebook's parent company, has announced a new open-source AI program called LLaMA 3.1 405B. Behind the name that only a programmer could love lies a new, more powerful large language model (LLM) that will power Meta AI assistant, which is baked into many of Meta's end-user programs. 

Meta CEO Mark Zuckerberg unveiled LLaMA  3.1 in a post on Threads, proclaiming, "open-source AI is the path forward." In an interview with Rundown.ai, Zuckerberg added, "Open models are going to be the standard, and I think that it's going to be good for the world. It's a bit soul-crushing when you go build features that are what you believe is good for your community, and then you're told that you can't ship them because some company wants to put you in a box so that they can better compete with you."

Also: Copilot Pro vs. ChatGPT Plus: Which AI chatbot is worth your $20 a month?

Does this mean that LLaMa 3.1 is really open source? The answer matters. As Neal Gompa, a Linux developer, asked on Mastodon, "Does that mean LLaMA 3 is under an OSI [Open Source Initiative]-approved open-source license, unlike its predecessors? We couldn't ship LLaMA 2 in Fedora because of its lack of use of an OSI-approved open-source license."

The answer is no. OSI executive director Stefano Maffulli told me in an interview, "In theory, we agree with all that Zuck wrote and said. If only Meta's license would remove the restrictions and Meta released full details about their training datasets and the training instructions, we'd be 100% in sync." 

"As it stands now," Maffulli  continued, "Llama is a liability for any developer; too opaque to be safe to use and with a license that ultimately leaves Meta in charge of their innovations."

Stephen O'Grady, a RedMonk industry analyst, agreed. "It's welcome news that Meta has dropped some of the use restrictions around Llama, but as long as it still restricts which companies can use the software, as the new license does, it's clearly and inarguably not open source. If Linux, for example, were released under this license, Meta would be entitled to use it, but companies like Amazon, Google, and Microsoft could not. That's not open source, nor would we accept it."

Also: The best AI image generators: Tested and reviewed

According to Zuckerberg, moving beyond the open-source issues to the technology, the new model is designed to be more efficient and powerful than its predecessors. It offers enhanced capabilities for various applications, from natural language processing to complex data analysis. In particular, he believes the Meta way is better than Apple's AI approach. 

The new LLaMA boasts a 405 billion parameter model, making it one of the most sophisticated AI models available. That's still much smaller than ChatGPT 4.0, with its 1.8 trillion parameters. That said, it's still a significant upgrade. It is expected to improve performance in language translation, content generation, and scientific research. 

Zuckerberg claims LLMA 3.1 is "competitive with some of the leading closed models, and in some areas, it's even ahead." Meta's benchmarking shows that LLaMA is competitive with leading foundation models, including GPT-4, GPT-4o, and Claude 3.5 Sonnet, across various tasks. He also states, "By our estimates, it's going to be 50% cheaper, I think, than GPT-4 to do inference directly on the 405B model." 

The new model is available today from more than 25 partners, including AWS, NVIDIA, Databricks, Groq, Dell, Azure, and Google Cloud. Meta also stated that it comes ready with support for such popular AI tools as vLLM, TensorRT, and PyTorch, so developers can immediately get to work with LLaMA 3.1

You can try out the new engine for yourself. Developers can check LLaMA 3.1 for themselves here.