Safe dialogue control for AI models without retraining

Original: SafeCtrl-RL: Inference-Time Adaptive Behaviour Control for LLM Dialogue via RL-Driven Prompt Optimisation

Source: arxiv.org ↗