At Tuesday’s Microsoft Inspire event, Meta unveiled the next generation of its open-source large language model (LLM) Llama 2.
Partnering with Microsoft to make the model available on Windows and Azure, Llama 2 will be available for free for research and commercial use.
Llama 2 is trained on 40% more data than its predecessor and is available in three versions with pretrained and fine-tuned models with 7B, 13B, and 70B parameters each. The models are available in fine tuned or trained forms.
Speaking with Radio 4 earlier this week, Meta’s president of global affairs, Nick Clegg, said making LLMs open source would make them safer and better.
The original version of Llama was released in February this year, but with a non-commercial licence that was only available to academics.
However, the files containing the learned training data, commonly referred to as weights, were leaked on a torrent site and became widespread across the AI industry.
With Llama 2, this data will be legally accessible, although Meta appears to have blocked its tech-giant competitors from using it.
Meta specified that licensee’s with more than 700 million monthly active users in the preceding calendar month must seek special permission to use the models.
Critics of open-source AI models say they contain risks of misuse and are almost impossible to regulate.
“Open-source drives innovation because it enables many more developers to build with new technology. It also improves safety and security because when software is open, more people can scrutinise it to identify and fix potential issues. I believe it would unlock more progress if the ecosystem were more open, which is why we're open sourcing Llama 2,” said Mark Zuckerberg, Meta’s CEO, in a Facebook post.
“Responsible AI is at the heart of Microsoft’s approach to AI and how we partner,” Microsoft said in a blog post responding to the partnership.
Expanding, it said “Microsoft mitigate potential risks presented by the use of large language models through an iterative, layered approach that includes experimentation and measurement.”
