Meta has announced that its new artificial intelligence model is the first open-source system capable of competing with products from rivals like OpenAI and Anthropic.
In a blog post, the company introduced its model, named Llama 3.1 405B, and claimed it is “competitive” with other models from OpenAI and Anthropic across a variety of tasks.
If Meta’s claim is accurate, this would mark the debut of one of the world’s most powerful AI models available without an intermediary charging for access or controlling its use.
Meta stated, “Developers can fully customize the models for their needs and applications, train on new datasets, and conduct additional fine-tuning.
This allows the broader developer community and the world to more fully realize the potential of generative AI. Developers can tailor the model for their applications and operate it in any environment … all without sharing data with Meta.”
The company also mentioned that users of Llama on Meta’s own apps (currently available only in the US) will benefit from additional “safety” features.
These safety measures are also open-source, but Meta does not have the ability to enforce their application by others using the model.
Mark Zuckerberg, Meta’s co-founder, expressed his belief in the necessity of open source for a positive future in AI.
He noted, “AI has more potential than any other modern technology to enhance human productivity, creativity, and quality of life, accelerate economic growth, and drive progress in medical and scientific research.
Open source will ensure that more people worldwide have access to AI’s benefits and opportunities, prevent power from being concentrated in a few companies, and enable the technology to be used more evenly and safely across society.”
Zuckerberg acknowledged the risk that “bad actors may use AI models to create new forms of harm,” but he argued that widespread deployment of AI would allow larger organizations to counteract smaller malicious actors.
So far, Meta has evaluated its model’s capabilities internally. The model’s size places it among the largest systems available, but it remains uncertain whether its sheer size translates to effectiveness comparable to leading competitors like GPT-4o, until independent evaluations are conducted.
Currently, the AI model is accessible to regular users in 22 countries, including the US, through Meta.ai.
Meta reportedly postponed a European launch due to concerns about the EU’s regulatory framework on AI and data protection.
However, the fully open-access version of Llama 3.1 is available globally to those who can implement it.