News

The LLaMa Language AI Model Created by Facebook Was Published on 4chan

Daniel ZimmermannMarch 10, 2023

623 1 minute read

Last week, 4chan user llamanon leaked the LLaMA (Large Language Model Meta AI) AI language model, which had previously been accessed by select researchers and scientists.

So far, Meta management has taken no action to remove this torrent file, which is still available for download.

language AI model LLaMA

Vice Motherboard notes that this is the first time that information about the AI model of a large company has become public. To date, major players such as Google, Microsoft, and OpenAI have kept their latest models secret, ostensibly for fear of abuse and misuse.

Let me remind you that we also wrote that Microsoft to Limit Chatbot Bing to 50 Messages a Day, and also that Amateur Hackers Use ChatGPT to Create Malware.

And security experts also reported that Bing Chatbot Could Be a Convincing Scammer, Researchers Say.

Meanwhile, 4chan users claim to be successfully running LLaMa on their computers.

Interestingly, in a media statement, Meta representatives did not deny the LLaMa leak, but did not confirm it either.

Meta’s goal is to share state-of-the-art AI models with members of the research community so that they can help us evaluate and improve these models. LLaMA was made available to [specialists] for research purposes, in line with the way we used to provide previous large language models. Although LLaMA was not available to everyone, and some tried to bypass the approval process, we believe that the current strategy allows us to balance accountability and openness.the company writes.

It was previously reported that LLaMa will be released under a non-commercial license focused on research use cases in order to “maintain integrity and prevent misuse” of the language model.

In February of this year, it was reported that access to LLaMa would be provided to select researchers in academia, individuals associated with government organizations, NGOs and academia, as well as industry research laboratories around the world.

At the same time, it was noted that Meta seems to be submitting requests to remove the language model from the public domain in order to control its distribution. For example, the head of artificial intelligence company Hugging Face, Clement Delangue, said that Meta requested the removal of a previously published model on GitHub, stating that it was “unauthorized distribution of Meta property that constitutes copyright infringement or inappropriate / unauthorized use.”

As noted by AnalyticsIndiaMag, a person who distributes LLaMa via a torrent and publishes a link to 4chan is unlikely to remain anonymous. The fact is that he did not remove the unique identifier from the code, which is used specifically to track leaks.

language AI model LLaMA