OpenAI Rewrites Model Spec Around Five Values and Calls the Model GPT
Summary
OpenAI updates its Model Spec, reorganizing guidance around five values: faithfulness to human intent, harm minimization, honesty and transparency, human agency, and care. The company says the revision adds guidance on context, oversight and agency, retires much detailed style guidance, and now calls the model GPT rather than the assistant.
Key Points
- Most Spec guidance takes the form of defaults that users and developers can override, apart from firm safety boundaries.
- OpenAI says it is exploring new training methods to teach models why the Spec's principles matter and will report results in future updates.
- A new subsection, consider context carefully, says conversations, memories and other contextual evidence should shape how models read user intent.