Pairwise Calibrated Rewards for Pluralistic Alignment | Read Paper on Bytez