Madrid, Jul 18 (EFE).- US technology giant Meta on Tuesday introduced Llama 2, the next generation of its large language model (LLM), which will be open source and free for both research and commercial uses.
An LLM is an artificial intelligence (AI) model that has been trained with deep learning algorithms to generate and generally understand text in a manner that is similar to humans.
Llama, which can be used to generate content in more than 20 languages, has more than 70 billion parameters to deliver better results and a more natural-sounding language, the company said in a statement.
As part of its “expanding partnership” with Microsoft, Meta is making the second iteration of Llama openly available in order to “benefit everyone,” the company said in a statement, after more than 100,000 requests were made to access the first version.
“We believe an open approach is the right one for the development of today’s AI models, especially those in the generative space where the technology is rapidly advancing,” Meta, the parent company of social media platforms Facebook, Instagram and WhatsApp, said.
Providing access for businesses of all sizes, as well as entrepreneurs and researchers, to AI codes that would otherwise be out of reach, will “open up a world of opportunities” for people to “experiment, innovate (…) and ultimately benefit from economically and socially,” the statement added.
This open source approach also means the model will benefit from rigorous testing, allowing the development and research community to quickly identify and resolve issues, which will in turn allow the company’s own teams to learn, improve and fix vulnerabilities, Meta said.
Supporting open approaches to LLM development will not only help spur innovation and further improve the technology, it will also make generative AI “helpful and safer”, the company said.
Despite the “huge advances” AI has brought to society as a whole, Meta acknowledged that it “also comes with risk.” LLMs can be abused or manipulated to generate harmful content or provide inaccurate and biased information.
That is why Meta says it has tested its models for safety, and will continue to do so as Llama 2 develops and improves.
The company led by American billionaire Mark Zuckerberg is also committed to transparency by explaining their methods for fine-tuning its models, and has provided a “responsible use” guide to “support developers with best practices,” the statement said, as well as a policy that Meta says will “help ensure that these models are being used fairly and responsibly.”
As part of its community-driven approach, Meta is looking to engage with innovators and researchers to help iron out any performance issues with its next generation LLM by launching a new partnership program for academic researchers.
The company will also launch a challenge to the AI innovator community that hopes to use the model to find solutions to “hard problems,” such as environmental or education challenges. EFE