The world of AI continues to evolve, and the next big contender is none other than Meta. The parent company of Facebook and Instagram unveiled this week its new artificial language model: LLaMA, which differs greatly from other popular AIs such as the famous ChatGPT.
According to The Verge, LLaMA is a research tool shared by Meta in order to “democratize access to this important and rapidly evolving field”. That is, instead of talking to it as you might do with ChatGPT or Bing‘s new AI, it would serve for researchers to solve problems related to AIs’ linguistic models, including their erratic or even toxic behavior.
LLaMA, consisting of four artificial language models of different sizes, is released under “a non-commercial license focused on research use cases,” and would grant access to universities, non-profit organizations and laboratories, which could make use of it free of charge to conduct their AI research.
In its release, Meta states the following, “We believe that the entire AI community (academic researchers, civil society, policy makers and industry) should work together to develop clear guidelines on responsible AI in general and large responsible language models in particular. We look forward to seeing what the community can learn (and eventually build) using LLaMA.”
According to the paper published by Meta, LLaMA-13B, the second smallest version of this artificial language model, performs better than OpenAI‘s popular GPT-3 model “on most benchmarks.” In addition, LLaMA-65B, the larger version, would be more “competitive with the best models,” such as DeepMind‘s Chinchilla70B and Google‘s PaLM 540B.
Meta has previously launched its own AIs, although they did not perform very well. While the quality of BlenderBot left something to be desired, Galactica, an AI created to write scientific papers, had to be withdrawn from the market after inventing the data for many of its papers.