Meta has disclosed the latest models of its artificial intelligence (AI) and announced plans to open source them. Mark Zuckerberg, Meta’s CEO, addressed their plan to make Llama 3.1 405B, their first “frontier-level” open source AI model, publicly accessible, including upgraded versions Llama 3.1 70B and 8B.
The recently developed 405B model is ‘state-of-the-art’, boasting flexibility and control equal to the finest proprietary models. Executives declared that the model, mainly employed for chatbot functions both internally and by external coders, possesses an array of fresh abilities, such as enhanced reasoning for solving sophisticated mathematical challenges or instantly creating a book’s worth of text.
The text-only model will be available within the European Union, Meta reported. These models, capable of multi-language processing, have undergone rigorous testing.
Mr Zuckerberg stated, “Llama 3 is showing its competitiveness with the leading models this year, and even outstripping them in some areas. From next year onward, we anticipate subsequent Llama iterations to become the industry’s top performers.”
Meta’s choice to provide this technology as an open-source product could draw criticism. Detractors of Meta’s openness with AI technology highlight the susceptibility for misuse, and the concern that tech firms from geopolitical competitors like China might utilise Meta’s technologies to stay on par with their American peers.
However, Zuckerberg defended the decision, arguing that open sourcing will foster growth of ecosystems around Llama. He believes open sourcing is fundamental to a beneficial AI future, ensuring the perks and prospects of AI are not monopolised by a select few companies.
Zuckerberg concluded, “The safety of open-source AI models is currently a contentious topic. In my opinion, open-source AI is safer than other options. I believe governments will decide supporting open source is in their best interest as it will generate prosperity and enhance global safety.”
In the future, there’s a possibility that malicious individuals might harness the intellect of AI models to create novel threats from the data readily accessible on the internet. Ensuring a power balance will be a critical aspect of AI security at this stage. It’s my opinion that a world with broadly applied AI is preferable, as it allows bigger players to monitor and control the influence of these smaller, harmful forces.”
Mark Zuckerberg has faced urgency to remain competitive in the rapidly advancing world of artificial intelligence, alongside tech giants like OpenAI, Microsoft, and Google. This has led to increased investment in the necessary technology and infrastructure to facilitate his ambitions, according to an additional report by Bloomberg. Detractors of Meta’s approach to making AI open source indicate towards its potential misuse- or the apprehension that technological firms from rival nations such as China might take advantage of Meta’s technology to stay on par with their US adversaries.