Reviews, studies, or reports assessing AI behavior changes post-implementation

OpenAI Releases Model Spec for API, ChatGPT to Guide AI Behavior, Benefit Humanity

May 8, 2024, 10:03 PM

OpenAI has publicly released the Model Spec, a comprehensive framework designed to guide the behavior of its AI models within the API and ChatGPT. This document, which is still in draft form and open for public feedback, outlines the core objectives, rules, and default behaviors that OpenAI's AI systems, including GPT-5 and future models, should follow. The Model Spec aims to ensure safe, lawful interactions with users, improve decision-making and instruction-following capabilities of the AI, and benefit humanity. OpenAI has described the Model Spec as a set of principles that help shape desired model behavior and is intended to start a public conversation about potential improvements.