Reverse Constitutional AI: A Framework for Controllable Toxic Data Generation via Probability-Clamped RLAIF
Yuan Fang, Yiming Luo, Aimin Zhou, Fei Tan
Preprint: not peer reviewed. Results may change in later versions.
The abstract is not reproduced here because its license does not permit redistribution. Read it at the original source.
Record
- Archive ID
- arxiv:2604.17769
- Published
- Apr 20, 2026
- arXiv
- 2604.17769
- Categories
- cs.CL, cs.AI
- Provenance
- arxiv