A new technology called SynthID, designed for AI watermarking, might make language models behave differently when given harmful prompts. This could lead to more predictable and controllable responses from these models.
מקור: Ars Technica – All content — לכתבה המלאה
כתיבת תגובה