Meta releases LLaMA-AI language models for scientific research

Meta has released a collection of AI language models to help researchers investigate language models. The LLaMA language models are not yet publicly available. Meta has been working on AI language models for years, just like Google and OpenAI, for example.

Large Language Model Meta AI, or LLaMA, consists of four AI language models of different sizes, all of which are relatively small. Therefore they require according to meta less computational power than other AI language models, which makes them better suited for research purposes.

“Over the past year, language models have revealed new possibilities for creating creative texts, solving mathematical theorems, predicting protein structures, and answering reading comprehension questions,” writes Meta. By this, the company refers, among other things, to the chatbots of, for example, Google and Microsoft. However, the Facebook parent company says it’s difficult to do scientific research with these models because they are so large that it takes a lot of computing power to understand how and why they work.

Meta says this barrier makes it more difficult to improve robustness and solve “known problems such as bias, toxicity and the potential spread of disinformation.” The company hopes that with the smaller language models better research can be done into this. The LLaMA language models are only provided after review to researchers involved in government agencies, civil society organizations or scientific institutions. Last May, Meta released another language model for scientific research.