Constitutional AI

GPTKB entity

Statements (23)
Predicate Object
gptkbp:instanceOf AI alignment technique
gptkbp:aimsTo align AI behavior with human values
gptkbp:alternativeTo gptkb:RLHF
gptkbp:appliesTo language models
gptkbp:citation https://arxiv.org/abs/2212.08073
gptkbp:describedInPaper gptkb:Constitutional_AI:_Harmlessness_from_AI_Feedback
gptkbp:developedBy gptkb:Anthropic
gptkbp:firstDescribed 2022
gptkbp:focusesOn helpfulness
honesty
harmlessness
gptkbp:goal reduce reliance on human feedback
https://www.w3.org/2000/01/rdf-schema#label Constitutional AI
gptkbp:influencedBy constitutional law concepts
gptkbp:involves revisions based on principles
self-critiquing
gptkbp:relatedTo AI safety
AI alignment
gptkbp:usedBy gptkb:Claude_AI_models
gptkbp:uses AI-generated feedback
set of principles or constitution
gptkbp:bfsParent gptkb:Anthropic
gptkbp:bfsLayer 5