Meta Platforms announced on Tuesday that they will launch a commercial version of their open-source artificial intelligence AI model, Llama. This move provides startups and businesses a cost-effective alternative to the expensive proprietary models OpenAI and Google offers.
The model’s latest iteration, Llama 2, will be distributed by Microsoft through their Azure cloud service and will be compatible with the Windows operating system. In their blog post, Meta referred to Microsoft as their “preferred partner” for this release.
Previously, Meta had only granted access to Llama to a select group of academics for research purposes. However, with the new release, the model will be available for direct download and accessible through Amazon Web Services, Hugging Face, and other providers. Meta CEO Mark Zuckerberg confirmed this in a separate Facebook post.
Zuckerberg wrote, “Open source drives innovation because it enables many more developers to build with new technology. I believe it would unlock more progress if the ecosystem were more open.”
The Rise of Open-Source AI Model: Disrupting the Dominance of Closed-Source Models
Offering a sophisticated model like Llama to businesses for free could disrupt the early dominance established in the emerging market for generative AI software. This market is currently led by players like OpenAI, which receives support from Microsoft and already provides its models to business customers through Azure. The original Llama model was already competitive with OpenAI’s ChatGPT and Google’s Bard chatbot. According to Zuckerberg, the new version of Llama has undergone further training, utilizing 40% more data than its predecessor and incorporating over 1 million human annotations to enhance the quality of its outputs.
Amjad Masad, the chief executive at software developer platform Replit, said, “Commercial Llama could change the picture. Any incremental improvement in open-source models is eating into the market share of closed-source models because you can run them cheaply and have less dependency,” said Masad.
Microsoft has recently announced its plans to expand its range of AI models available to business customers, following similar moves by its cloud competitors, Google and Amazon. Amazon, for example, offers access to Claude, an AI model developed by the prominent startup Anthropic, along with its own Titan models. Google also intends to provide its cloud customers access to Claude and other models.
Impact on OpenAI and Microsoft’s Strategy
In the past, Microsoft primarily focused on making OpenAI’s technology accessible through its Azure platform. However, the decision to support additional offerings from other AI models aims to give developers more choices and further solidify Microsoft’s position as the leading cloud platform for AI work. Despite concerns about potentially diminishing OpenAI’s value, a Microsoft spokesperson emphasized the importance of providing developers diverse options to support their AI projects.
A leaked internal memo from Google titled “We have no moat, and neither does OpenAI” caused a stir in the tech industry in May. The message discussed how Meta, a growing open-source ecosystem of AI technology built using its models, could pose a challenge to competitors who rely on revenue from their proprietary technology. If developers could access equally powerful open-source systems for free, the value of proprietary technology would diminish.
Meta believes that by fostering an open-source community, it can benefit from the advancements, bug fixes, and new products that may arise from its model, becoming the default choice for AI innovation. This strategy aligns with Meta’s track record of success with its widely-adopted open-source AI framework, PyTorch.
As a social media company, Meta’s CEO Mark Zuckerberg emphasized during an investor meeting in April that the company stands to gain more by crowdsourcing ways to reduce infrastructure costs and by maximizing the creation of new consumer-facing tools. These tools can help attract users to Meta’s ad-supported services rather than charging for access to its models.