In collaboration with researchers from Tsinghua University, DeepSeek developed a technique that combines methods referred to as generative reward modelling (GRM) and self-principled critique ...