I don't think AI builders are short of generic training files. They don't have the restricted data set that an expert won't easily part with.

That’s a worse bottleneck than model selection. A dataset might be useful enough to help a particular model, yet too valuable for its owner to release on faith. If the only option to monetize is to give away the thing you want to monetize, serious owners are not going to become providers. They never go in.

The OpenLedger surface I find worth tracking is the ModelFactory. Its flow is detailed allowing fine-tuning on Datanets permissioned and accepted using OpenLedger. A model is private when it is built, and released to the public only after a separate deployment phase. Training is also priced in the network’s native cryptocurrency.

That sequence means more to me than another revelation of an AI model. It decouples the requirement for limited training material from the choice of releasing a useable model. There may be a reason for a data owner to participate. A constructor has a route to something better than scraped scraps.

I have not seen enough to assume that the boundary is perfect. Permission before training only important if the deployed model does not silently transform the original dataset back into free material.

The supply of AI models is easy to boost. Specialist data you can trust isn’t. That permissioned path is the use signal I'd measure for @OpenLedger $OPEN #OpenLedger