We recently reported on the launch of LLaMA, the new artificial language model created by Meta to “democratize access to this important and rapidly evolving field” for universities, non-profit organizations and laboratories to conduct their AI research.
Although Meta made this AI available to these groups completely free of charge, its access was not public, a circumstance that would have changed a few days ago, although not by Meta’s will.
A few days ago, several LLaMA models were leaked and are now publicly available via Torrent. The author of the original leak is a 4chan user named “llamanon”, who posted on the tech board Torrent links to LLaMA models 7B and 65B. The post was made on AI Chatbot General, a mega-thread where the operation and capabilities of the latest AI chatbots are compiled.
Later, the Torrent link to LLaMA’s IA models was added as a pull request on the Meta project’s GitHub page by user ChristopherKing42, under the title “‘save bandwidth by using a torrent to distribute more efficiently”.
The pull request was posted along with the Google Forms link that Meta was using to provide access to the bot (apparently a hint to the LLaMA request process). Although the identifier code of the user who leaked the template was included in error, making his identity known, Meta has so far taken no action and the Torrent link is still available.
LLaMA is not the first AI created by Meta. A year ago, the company launched Make-A-Scene and Make-A-Video, two artificial intelligences capable of creating images and videos, respectively, from text. It also created BlenderBot and Galactica, although the results of both were rather poor, and even the latter was withdrawn from the market after it was shown that it made up data on many occasions.
