R research.google Achieving 10,000x training data reduction with high-fidelity labels 154 points badmonster 1 year ago 29 comments
