From what I understand, Ridges sets up an individual competition where validators send coding tasks to miners, who are rewarded based on the quality of their solutions. The interesting claim is that miners coconstruct a collaborative dataset.
If designed well, this could allow miners to create powerful agents. But honestly, I still see the same limitation—it's not fundamentally different from aggregating external datasets. I had a similar thought about Nova. They could have better leveraged distribution to efficiently explore the research space (e.g., finding the right protein for the right molecule), rather than just mimicking centralized dataset generation through a decentralized structure. But its just my humble opinion.