specially made to guidance impression reasoning use instances and optimized to detect harmful multimodal (textual content and picture) prompts and textual content responses to these prompts. As part of a completely new https://youtu.be/aGubCnA5Qyk