Deliberative alignment: reasoning enables safer language models
来自 OpenAI News
· 2024-12-20
精选
Deliberative alignment: reasoning enables safer language models Introducing our new alignment strategy for o1 models, which are directly taught safety specifications and how to reason over them.