OpenAI updates Model Spec around five values, adjusts some behavioral defaults

OpenAI said on Oct. 29 that it updated its public Model Spec, reorganizing the guidance around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care.1

OpenAI said that as its models take more consequential actions, they need to understand human intentions and values so they can use good judgment in situations their instructions do not explicitly address.1 The company said the update follows research, deployment experience, public feedback on the Spec and discussions with users and outside experts.1

According to OpenAI, the revision adds guidance on context, oversight and agency, adjusts some behavioral defaults, and retires much of the detailed style guidance.1

OpenAI describes the chain of command this way: OpenAI writes the Spec, sets boundaries and trains models to respect them; developers direct applications within those boundaries; and users direct the model's work subject to those constraints.1

A new subsection, "consider context carefully," explains how conversations, memories and other contextual evidence should inform the model's understanding of people's intentions and preferences, and OpenAI also added defaults on supporting human agency and following relevant social norms.12

OpenAI said the Spec has firm boundaries where needed to protect people and society, and otherwise mostly takes the form of defaults that users and developers can override.1