It is true that open source models can be abused, like any other technology, but this view tends to overemphasize the risks of open source AI and underestimate its benefits. The information needed to build a biological weapon is already available on the Internet, and as Mark Zuckerberg says, a well-designed open source AI should favor defenders rather than attackers. Furthermore, Chinese models are now, in many cases, comparable to those of Meta.
Meanwhile, the benefits of open source software are clear. It underpins the entire technology industry and powers the devices billions of people use every day. The core software of the web, for example, is open source, with standards released into the public domain by CERN’s Tim Berners-Lee. The same is true of the Ogg Vorbis special lead compression algorithm, which Spotify uses to stream music to millions of people.
developers to harden their code, test its reliability, leverage the volunteer work of many, and in some cases, make money by offering technical support. Openness should also be the basis for innovation in artificial intelligence. If the technology truly has the potential its proponents claim, then it is a way to ensure that power does not remain concentrated in the hands of a few Californian companies.
Closed models will continue to be needed for sensitive applications or cutting-edge tasks. However, open, or partially open, models will play a crucial role. The Open Source Initiative defines a model as open source if it is freely downloadable and usable, with a description of the underlying training data. However, none of the open models from large labs, such as Alibaba and Meta, fully meet these criteria. By offering partially open platforms, however, these labs provide information about their models, allowing others to learn and, sometimes, further develop their techniques.