Sorry @nick, maybe I face a language barrier here, but I’m not sure I understood the new FAQ.
The new OSAID version requires training data availability or not?
I mean: at a first look, it seems that the new FAQ makes the loophole the community have been reporting for months explicit, instead of fixing it.
I will be more than happy to endorse an Open Source AI definition that won’t enables black box builders to elude the legal and technical scrutiny required by the AI Act, but technically speaking, such change would rule out “Unshareable non-public training data” that still appear in the FAQ.
Did you remove them from the Release Candidate? Or you left them and explained that only AI systems that are trained only from Open or Public datasets can be qualified as Open Source?
If so, many thanks for fixing the issue and thanks to @samj for his analysis of the working group dynamics that shown how such an erroneous conclusion were drawn.
