OpenAI rewrites its Model Spec around five core values

OpenAI has updated its Model Spec, the public document describing how it wants its models to behave, reorganizing the guidance around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. The stated goal is for models to understand and apply human values in unfamiliar situations while still deferring to human intent and oversight.

The update adds guidance on context, oversight, and agency, and adjusts some behavioral defaults. It retires much of the old detailed style guidance in favor of letting style adapt to user preferences, while keeping some defaults, such as approaching questions objectively. A new section on considering context carefully says conversations and memories should inform the model's understanding of a user's intentions, citing an example where a model recalling a user's stated preference for predictable work hours could surface that while comparing job offers, as long as the assumption stays visible and easy to correct.

The Spec now refers to the model as 'GPT' rather than 'the assistant,' and OpenAI says it is exploring new training methods to help models actually learn these values, with results to come later. The company acknowledges its models don't yet fully reflect the updated Spec and is inviting public feedback.

Why it matters

The Model Spec shapes how ChatGPT and other OpenAI products are expected to behave, so changes here eventually filter down to everyday usage.

Source: OpenAI