Yet another inevitable issue is the compatibility or boundary issue between open source definition (OSD) and open source AI definition (OSAID).
Should we consider OSAID a super set of OSD, where OSAID adds additional standards on dataset, pre-trained nets, etc. Or should we consider OSAID an independent and orthogonal definition w.r.t. OSD, where OSAID covers the parts (such as dataset, pre-trained weights) that the OSD does not cover very well without ambiguity?
I think the answer to this question also depends on how much components we want to cover from the AI system. And we may need to consider possible cases where OSD conflicts the OSAID through some sort of loophole.
In my fuzzy memory I raised the question in the private mailing list before. Will update this post if I find some missing points.
Source of this topic: Training data access - #56 by Aspie96