Deliberative alignment: reasoning enables safer language models
Deliberative alignment trains o1 models to directly reason over safety specifications rather than just pattern-match, representing a new alignment technique for reasoning models.
Excerpt
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.
Read at source: https://openai.com/index/deliberative-alignment