Deliberative alignment: reasoning enables safer language models

OpenAI introduces a new alignment strategy for o1 models, which are directly taught safety specifications and how to reason over them.

Summary written by Kernelia from the original article by OpenAI News. The story and its rights belong to its author.