Definition
RLCD is TypeSafe’s training method optimizing for calibrated probabilities — when a model reports 70% confidence it should be correct ~70% of the time — distinct from RLHF (preference) or RLVR (verifiable correctness).
RLCD is TypeSafe’s training method optimizing for calibrated probabilities — when a model reports 70% confidence it should be correct ~70% of the time — distinct from RLHF (preference) or RLVR (verifiable correctness).