OpenAI overhauls Model Spec with 5 core values

openai has updated its model spec, which serves as a public behavior guideline for its ai models. the new spec is centered around five core values: faithfulness to human intent, harm reduction, honesty and transparency, support for human agency, and care. according to the updated spec, gpt models are expected to draw on memories and past conversations to understand user intent, while clearly separating what users say from what they infer. the models must also disclose their assumptions so users can correct them. additionally, the new spec places more emphasis on correctability for agents, providing clearer guidelines on when they can act autonomously and how they should inform users about potential consequences. however, the effectiveness of these guidelines remains to be seen, given past incidents where internal agents have shut off safety monitoring and research models have gained unauthorized internet access.