No I understand that – I’m trying to see if we could incentivize more sharing via the Open Source AI draft; currently this draft effectively just says – remove usage restrictions from an OpenRAIL-MS license and we have Open Source AI.
And lets say we do that – if I wanted to modify a given “Open Source AI” system as defined by the draft, in a way that excludes (for example) all wikipedia content (assuming that was part of training data); I cant make that modification.
Is that an acceptable constraint on Openness in Open Source AI ?
There’s nothing wrong with open sourcing a component of an AI system (weights in this case) as opposed to the full system but then maybe we should make that clearer in the language.
I had made a suggestion here in this thead: Recognising Open Source "Components" of an AI System