Meta’s (META) approach to releasing open source AI models combats concerns about the excessive concentration of power in the hands of a few large technology companies, according to Nick Clegg, the company’s president of global affairs. Unlike competitors such as OpenAI, Google (GOOGL) and Anthropic, Meta’s approach to the AI arms race is largely defined by open access, allowing the source code of models such as the recently released Llama 3.1 405B to be freely downloaded and modified. . In contrast to closed AI systems, Meta’s open source offerings mean you have “really democratic” access to new technologies, Clegg said today (September 24) while speaking at the Concordia Conference in New York.
While Meta has been opening up its AI models for commercial use for a long time, Llama 3.1 405B, launched in July, is the largest open source AI model ever released. According to the company, the new model’s capabilities are comparable or superior in all long-term benchmarks compared to competitors such as OpenAI’s GPT4o and Anthropic’s Claude 3.5 Sonnet.
This is not the first time Clegg has defended Meta’s approach to open sources. A former member of the UK parliament who first joined Meta in 2018 as a lobbyist and communications officer, a large part of Clegg’s role as head of global affairs includes representing Meta’s products. In a Financial Times op-ed last year, Clegg reiterated Meta’s belief that open sourcing will provide broader access to AI instead of keeping the technology concentrated among powerful companies.
Open source AI models should come as no surprise
Finding a source is not a revolutionary idea especially in the world of technology. “If you look at the evolution of the Internet itself in the early days or you look at the encryption industry or the Internet security industry, they’re all built on open source technology,” Clegg said, noting that open access has also been a tool for a long time. part of Meta’s corporate strategy. “Over the past ten years or so, our AI research team has opened up more than 1,000 AI data sources and models. So it is not new for us,” he added.
Meta’s decision to open the well is also consistent with its structure, Clegg said. Since the company’s business model does not depend on charging for access but is built on advertisements, Meta can open its products without negative financial consequences, the official explained.
Despite the many benefits of open source, its critics worry that the practice could create opportunities for exploitation by bad actors. Such fears are unfounded, according to Meta CEO Mark Zuckerberg, who has maintained that it is better to open the source of the best technology, the big players are able to check the power of the small, bad ones.
Clegg argued that wider access to AI ultimately improves the technology itself. “You get wisdom by testing people and they test again,” as opposed to letting companies “instead play ‘Whack a Mole’ with their models,” he said. “Experience, certainly over the last 20 years, suggests that an open approach often leads to innovation. big, smart, democratizing technology—and eventually, hard systems.”
