Meta Introduces Large Language Model LLaMA as a Competitor for OpenAI
Meta has shared a new large language model called the Large Language Model Meta AI (LLaMA). The model is designed to aid researchers working on generative AI for language to solve some of the problems that emerge in products like ChatGPT.
Meta is releasing LLaMA, a collection of four different models, under a license explicitly for research and not commercial use. Research universities, non-profits, and industry R&D labs will have access to LLaMA, but not Meta’s commercial rivals. The LLaMA collection is impressive if it can match Meta’s claims. The tech giant says the LLaMA-13B, referring to its 13 billion parameters, outperforms GPT-3 in most metrics despite being the second smallest of the LLaMA models and capable of running on only one of the admittedly advanced Nvidia Tesla V100 GPUs. The biggest LLaMA, at 65 billion parameters, is supposedly a match for the highest end of Google and DeepMind’s models.
“Smaller, more performant models such as LLaMA enable others in the research community who don’t have access to large amounts of infrastructure to study these models, further democratizing access in this important, fast-changing field. Training smaller foundation models like LLaMA is desirable in the large language model space because it requires far less computing power and resources to test new approaches, validate others’ work, and explore new use cases,” Meta explained in its announcement. “We believe that the entire AI community — academic researchers, civil society, policymakers, and industry — must work together to develop clear guidelines around responsible AI in general and responsible large language models in particular. We look forward to seeing what the community can learn — and eventually build — using LLaMA.”
LLaMA’s release doesn’t include any option for the kind of generative AI chatbots that have exploded in popularity of late. Instead, LLaMA is to “address the risks of bias, toxic comments, and hallucinations.” That also means Meta won’t face the need to immediately shut it down in a few days like the Galactica Large Language Model. Shared in November to demonstrate using generative text AI to write academic papers, Galactica quickly attracted a ton of online ridicule over nonsensical and problematic writing, leading to its removal less than a week later. LLaMA doesn’t even include the limited demo collection shared when Meta showed off its Make-A-Video synthetic media engine.