OpenAI Updates Its Model Spec

The author analyzes OpenAI's updated Model Spec document, highlighting changes to core values, risk taxonomies, side-effect controls, and decision-theory principles. While noting that the new spec remains philosophically similar to its predecessor, the author critiques several new examples as overdetermined and failing to clarify difficult trade-offs for models and developers. Open article →