October 31st, recentlyOpenAI renew Model SpecReorganizing the principles of model behaviour with five values: loyalty to human intent, harm reduction, honesty and transparency, safeguarding people ' s autonomy and care. Normatively driven ChatGPT AND GPT MODELS FOR PRODUCTS SUCH AS API, PUBLICLY ILLUSTRATE HOW COMPANIES WANT TO SHAPE MODEL BEHAVIOUR。
The new edition places at its core what the users and developers want to do, while retaining the command level and the security boundary. For tasks for which objectives are not yet clear, the model should understand the intent in the context; the facts, evidence, uncertainties, capabilities and actual actions should be presented in a factual manner that enables an assessment of its work。
The norms are also devoted to the tension between immediate commands and long-term objectives: models can point to neglected consequences, help users to think, but they cannot anticipate the choices that users will make in the future as decisions they have already made. OpenAI states that the online model is not yet fully compliant and that the company will continue to adapt its training and systems。