Reverse Constitutional AI: A Framework for Controllable Toxic Data Generation via Probability-Clamped RLAIF.

Yuan Fang, Yiming Luo, Aimin Zhou, Fei Tan

Browse the full ACL paper archive.