Towards Privacy-Preserving Large Language Model: Text-free Inference Through Alignment and Adaptation

ArXi:2604.06831v1 Announce Type: cross Current LLM-based services typically require users to submit raw text regardless of its sensitivity. While intuitive, such practice