Last week, Meta unveiled LLaMA, the largest artificial intelligence model 🤖 with 65 billion parameters, and the highlight is that Meta provided pre-trained models for users to use. However, the usage is limited to educational research purposes only, and researchers must apply to use it individually. Recently, someone obtained the model and distributed it on torrent sites 🦾
Meta released the LLaMA model as open source under the GPLv3 license, allowing it to be used for commercial purposes, but the source code must be disclosed, similar to Linux. However, the data sets and parameters of the trained AI model were not disclosed, and it is difficult for the general public to collect them. Moreover, training these models can cost hundreds of thousands or even millions of dollars, making it impossible for small businesses or individuals to train these models themselves.
When distributing the LLaMA parameters, Meta created URLs for individual licensed users. However, when developers compared the downloaded files, they found that the data was the same and there was no watermark in the parameters. In this case, the person who distributed the torrent file mistakenly included their own URL in the torrent, allowing Meta to trace the source of the leaked files.
LLaMA can be further trained with human-like capabilities, similar to ChatGPT. The leak of this model may shed light on further development of LLaMA. However, the licensing agreement of the model is not open source, and organizations that need to be cautious about copyright issues may not be able to use it.
Let’s wait and see who will do something with this model 🔥