According to reports, news that Meta’s latest large-scale language model leaked, unauthorized users, can also download and use it, in response to this matter, Meta said will continue to provide approved researchers with AI tools.
“While models are not open to everyone and some bypass the approval process, we believe the current release strategy allows the company to balance accountability with openness,” Meta said in a statement.
Facebook and Meta have a large AI research and development team, and the company just launched LLaMA last month. According to Meta, LLaMA has the same super human-like dialogue ability as OpenAI and Alphabet systems but requires less computing power.
Competitors such as OpenAI retain technology, and developers need to pay if they want to use it. Meta is different, and it shares most of the work results with developers. AI tools can be misused, for example, to create and spread disinformation. In order to prevent abuse, Meta only provides AI tools to researchers, government-related entities, civil society, and academic institutions, and must go through a review process and sign a non-commercial authorization agreement before providing them.
Last week, a user-provided the LLaMA model for download on the 4Chan forum, which means that developers can get the model without review. In a statement, Meta said that the rollout of LLaMA is the same as previous models and that the company has no intention of changing strategy. “The goal of Meta is to share state-of-the-art AI models with members of the research community so they can help Meta evaluate improved models,” Meta said.