New York, Feb. 24. Meta joined the race for big tech companies in the field of artificial intelligence (AI) on Friday with a tool intended only for researchers with prior authorization, dubbed LLaMA (Large Language Model Meta AI).
It is not a “chatbot” like those launched by Microsoft or Google, but a program for those who develop this type of technology, which hopes to help “limit or eliminate” the current problems of bias, “toxicity” or “misinformation” that weigh on these first results.
LLaMA is a “large next-generation foundational language model designed to help researchers advance their work in this subfield of AI” and seeks to “democratize access” for those without the infrastructure to study it, Meta said. .
“Even with all of these recent advances in large language models, full access to their research is still limited by the resources needed to train and operate” this technology, hampering understanding and solving its problems, he said. explain.
LLaMA is able to generate text and is trained in the 20 most spoken languages, especially those with Latin and Cyrillic alphabets, the note says.
Meta noted that it releases this tool under a research-based non-commercial license to “prevent its misuse” and will only authorize its use after reviewing one-by-one requests from associated researchers, educational institutions , governments and civil society. ECE
si/fjo/bdp