In the last month, we saw two articles expressing opposite viewpoints about open source AI.

John McBride of the Linux Foundation wrote in Substack that “Unfortunately, there is no ‘open source’ AI and innovation will be stifled by a locked down ecosystem that only a few AI labs can access.” He exhorted developers to “Release the training data!” or risk stifling innovation.

Matt Asay of MongoDB (and long time open source pundit) wrote in InfoWorld that “It’s Linux all over again. One passionate start becomes a movement, then infrastructure, then a global standard…The key difference is that this time it’s happening in months, not decades.”

Who is right? Deepseek’s spark of AI innovation is undeniable, but McBride warned that lack of access to training data means there is no real transparency, and that will ultimately harm innovation.

Maybe both are right, in their own way. Their extreme difference of viewpoint comes from their context. Asay is an advocate for open source in business (COSS) and McBride is a developer advocate for free software. They are also using different effective definitions of open source AI: Deepseek does not even meet the OSI’s definition, and despite the innovations Deepseek has triggered, McBride says that definition doesn’t go far enough. Most people would probably think the pace of AI development is plenty fast already, as we are seeing a new and better model almost daily. But that’s not the only benefit of open source: transparency and reproducibility are goals that help build trust and equal access.