In a bold move first reported by tech news outlet The Information, Meta is preparing to issue a commercial license with the release of their next open-source language model. Sources in Meta confirmed that the company plans to encourage other companies to not just develop but also profit from their open-source AI software.
Meta is focused on finalizing an upcoming open-source large language model (LLM), which it plans to make available for commercial purposes for the first time. This approach would represent a vast departure from closed-source language models like Google’s Bard and OpenAI’s ChatGPT currently in commercial use, and it could spark a ripple effect of widespread adoption by companies that are on the lookout for a more versatile and affordable AI alternative.
However, the benefits of this move won’t be confined to the companies that adopt open-source models. Should a growing number of developers gravitate towards Meta’s open-source AI ecosystem, Meta itself would benefit from the expertise of countless AI engineers worldwide contributing improvements to its core models.
Meta’s plan to release an open-source commercial-friendly model is not entirely unexpected. The company previously launched a highly-capable open-source LLM, known as LLaMA, in February 2023. Although initially licensed for research use and distributed to a select group of users, LLaMA’s code leaked its way into the broader tech landscape and has served as a foundation for numerous new open-source AI models developed on top of its core technology.
These open-source models are swiftly closing the gap with their closed-source counterparts. The 13-billion parameter Vicuna LLM model, built on top of the LLaMA foundational model, made waves for its claim to deliver 90% of ChatGPT’s quality in March. Google AI engineer Luke Sernau shone a spotlight on Vicuna’s rapid progress in a leaked internal memo, claiming that Google and OpenAI’s moat with closed-source models was non-existent as open-source models continued to make fast and substantial strides.
Since then, the open-source world has seen even more progress. Hugging Face’s Open LLM leaderboard now lists an array of even more sophisticated models, many of which are derivatives of the original LLaMA LLM, and continue to expand their capabilities at an impressive rate. Among these, open-source models like Falcon, developed by Abu Dhabi, have also seen growing popularity among developers.
With the burgeoning ecosystem of open-source language models, OpenAI has found itself in a tricky spot. The Information disclosed last month that OpenAI was readying its own open-source LLM as it grapples with the escalating pressure and competition from open-source contenders and aims to protect its market share.
Meanwhile, as AI continues to form a significant part of Meta’s strategic roadmap, the company is doubling down on its commitment to an open-source strategy and regularly releasing research and code into the wild at a time when Google and OpenAI have walled off much of their research.
When asked this week about whether the rapid advancement in AI could potentially pose a threat to humanity, Meta’s Chief AI Scientist, Yan LeCun, dismissed such a notion as “preposterously ridiculous.“