This week I got to hear a presentation by Nick Vidal, OSI’s head of community, about the definition of open source AI. The fact that this definition requires developers to share all data and all source code that was used to create and train the model. While this makes sense from the perspective of the original open source definition, the marketing and competition implications are confusing to me. The market for AI models is characterized by intense competition, and the emergence of a new model can cause a stir in the industry. However, after I asked him this question, Nick explained that many (non-FOSS) AI companies already share their training data and code, including large ones like DeepSeek.

The sharing of source code is interesting, though, because that means that all the components of the training process must be open source. For example, if I create a tool that uses an external AI model that is not open source, my entire tool does not fall under the open source definition. Nonetheless, I do agree that we should treat open source AI like any other open source project, which means holding its creators to the same expectations of sharing all of the backend.

Aside from the guest lecture, we’re still in the early stages of project work. Will report back later.