The progress of AI in the past few months has been so rapid that it has surprised me a bit: the on-chain credit scoring thing has not yet been truly realized by the ensemble model. The challenges of DeFi undercollateralized credit in the past were not just about the model, but also included high training thresholds for large models, too few available labels, difficulty in binding on-chain identities, low witch costs, and the weak ability to pursue debts in the real world. But now these conditions are actually in place: models and computing power are available to everyone, DeFi data is extremely vast and completely public, and with the advancement of regulations and KYC/offchain attestations, pursuing claims through legal means is also starting to become possible. I increasingly believe that Machine Learning will definitely make on-chain undercollateralized credit happen; it's just a matter of time. Has any big shot already been seriously working in this direction?