OpenAI Rebuilds Model Spec on Five Values, Renames Model 'GPT'

OpenAI reorganized its Model Spec around five values: faithfulness to human intent, minimising harm, honesty and transparency, human agency, and care. In the same pass, the document renames "the assistant" to "GPT."

Most of the detailed style guidance from earlier versions is gone. In its place is a lighter architecture in which most rules act as defaults a user or developer can modify, with firm restrictions kept only where OpenAI says they are necessary to protect people. "As our models get smarter and take more consequential actions in the world, it becomes increasingly important that they go beyond just interpreting the specific instructions they receive," the company writes in the Spec. On the first value, OpenAI says a model should "grasp what people actually want rather than follow instructions word for word," according to The Next Web.

The company also concedes a gap between the document and the shipping products. OpenAI says its models "do not yet fully reflect" the new Spec and that it is testing novel training methods to instill the values. It is inviting public feedback on where the balance between defaults and hard limits should sit.

The rewrite is the second refresh this year, after an August version that focused on teenage interactions. It lands amid a stretch of OpenAI policy moves our tracker has been following, including the opening of an election operations center for the November 3 midterms.