permissiveContentTransformations
Guardrails that allow for permissively transforming text input, including potentially unsafe content, to text responses.
Declaration
static let permissiveContentTransformations: SystemLanguageModel.GuardrailsMentioned in
Discussion
The permissiveContentTransform guardrail model lets the model handle potentially unsafe content, such as summarizing a news article. In this mode, requests you make to the model that generate a String will not throw LanguageModelError.guardrailViolation(_:) errors. However, the model may still sometimes refuse to respond to a sensitive prompt, in which case it generates a String refusal message.
When you generate responses other than String, this mode behaves the same way as default mode and throws LanguageModelError.guardrailViolation(_:) errors.