Teaching language models to reason over explicit safety policies before answering simultaneously improves resistance to jailbreaks and reduces overrefusal in OpenAI's o1 series.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CL 1years
2024 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Deliberative Alignment: Reasoning Enables Safer Language Models
Teaching language models to reason over explicit safety policies before answering simultaneously improves resistance to jailbreaks and reduces overrefusal in OpenAI's o1 series.