AI RESEARCH Consolidating Rewarded Perturbations for LLM Post-Training arXiv CS.LG • June 01, 2026 Post- Read Full Article