top of page
  • Koosmik

What does ‘open source AI’ mean?

The struggle between open source and proprietary software is an old one, but this conflict is now creeping into the realm of artificial intelligence (AI), creating new controversies. A recent New York Times article praised Meta CEO Mark Zuckerberg for his support of “open source AI,” though it pointed out that Meta's AI models, branded under "Llama," aren’t truly open source.

The debate about what constitutes “open source AI” is getting more intense. The Open Source Initiative (OSI), led by executive director Stefano Maffulli, is attempting to address this issue. The OSI has been defining “open source” for over 25 years, but applying software licensing terms to AI is challenging. Some experts, like Joseph Jacks of OSS Capital, argue there's "no such thing as open-source AI" because AI’s neural network weights differ fundamentally from software source code.

Meta’s Llama models are promoted as open source, but they come with restrictions. For example, app developers with over 700 million users must get special permission from Meta. This inconsistency has led to debates about the true openness of Meta’s AI models.

The OSI is working on a formal “Open Source AI Definition.” This definition aims to allow people to use, study, modify, and share AI systems without restrictions but faces challenges, particularly around access to training data. Sometimes, the data used to train AI models cannot be shared due to confidential or legal constraints.

The OSI plans to finalize its definition by the end of October, intending to ensure it evolves with technology. The goal is to make AI systems easy to replicate and transparent, aligning them as closely as possible with open source principles.

Source TechCrunch


bottom of page