OpenAI Sets
The Decoder is right to pair the update with a skeptical headline: a spec is a statement of intent, not a control mechanism, and past versions have shown gaps between written rules and actual model behavior. The Model Spec has quietly become OpenAI's most consequential public document — it defines who the model serves in developer-versus-user conflicts, and the new emphasis on inferring 'what users actually want' nudges models toward more paternalistic defaults.… more
In our view that cuts both ways: guessing intent helps in ambiguous chats, but it also licenses a model to override what was literally asked. The upside of formalizing values like correctability is that researchers and auditors finally get a published yardstick — when a model misbehaves, they can cite the rule it broke. Whether training and system prompts can enforce all this remains the genuinely open question.
OpenAI sets new behavioral guidelines for AI models that have already proven hard to control
OpenAI has updated its Model Spec, restructuring its AI behavior guidelines around five core values, including faithfulness to human intent and correctability. GPT models should now draw on conversation history to figure out what users actually want and di…
Source:
The Decoder
Discussion (0)