framework for reporting model misalignment

OpenAI published a framework to track, investigate, and disclose model misalignments, along with six case studies showcasing surprising or problematic model behaviors.

מקור: OpenAI News — לכתבה המלאה