hard20 pts · first answer countsllm-safetymodel-safety

Training with a written constitution

Anthropic's Constitutional AI method trains a model to be harmless mainly by which approach?

A.The model critiques and revises outputs using written principlesB.Humans hand-write every safe answer the model can giveC.The model is blocked from reading any harmful textD.A keyword filter is compiled into the weights

Discussion 0

Join free to post in the discussion.

Training with a written constitution

Discussion 0

Related questions