Meta releases LLaMA, a new AI language generator for researchers
Meta, the parent company of Facebook, has released a new AI language model named LLaMA, which aims to help researchers solve the problems that plague AI language models. Unlike popular AI assistants ChatGPT or Bing, LLaMA is not a system you can talk to. Instead, it is a research tool that Meta is sharing to "democratize access in this important, fast-changing field." The company hopes that experts can use LLaMA to address AI language model issues, such as bias, toxicity, and the tendency to generate fake information.
LLaMA is actually a quartet of different-sized models that Meta is releasing under a non-commercial license focused on research use cases. Access to LLaMA is granted to groups like universities, NGOs, and industry labs. Meta believes that the entire AI community must work together to develop clear guidelines around responsible AI, particularly responsible large language models.
Meta claims that the second-smallest version of the LLaMA model, LLaMA-13B, performs better than OpenAI's popular GPT-3 model "on most benchmarks," while the largest, LLaMA-65B, is "competitive with the best models," like DeepMind's Chinchilla70B and Google's PaLM 540B. After training, LLaMA-13B can run on a single data center-grade Nvidia Tesla V100 GPU.
Meta's release of LLaMA is notable because it has missed out on some of the buzz surrounding AI chatbots. Meta has actually released its own accessible AI chatbots in the past, but their reception has been less than stellar. One of them, named BlenderBot, was criticized for not being very good, while another named Galactica, designed to write scientific papers, was pulled offline after only three days due to producing scientific nonsense.
Meta hopes that LLaMA will receive a kinder reception and that the AI research community can learn and build upon the model. Mark Zuckerberg, CEO of Meta, said in a Facebook post, "LLMs have shown a lot of promise in generating text, having conversations, summarizing written material, and more complicated tasks like solving math theorems or predicting protein structures. Meta is committed to this open model of research and we'll make our new model available to the AI research community."