OpenAI Model Spec
Summary
Documentation outlining the intended behavior and 'chain of command' for OpenAI models to ensure they are useful, safe, and aligned with user and developer needs.
Key quotes
The Model Spec outlines the intended behavior for the models that power OpenAI’s products, including the API platform.
Instructions with higher authority override those with lower authority.
The assistant should not allow lower-level content... to influence its interpretation of higher-level principles.
The document defines a hierarchical ‘chain of command’ (Root, System, Developer, User, Guideline) to resolve conflicting instructions. It includes red-line principles for human safety and a framework for managing model autonomy and side effects.