OpenAI has released a significantly expanded version of its Model Specification (Model Spec) for ChatGPT, aiming to refine how the AI model handles controversial topics and user interactions. The updated 63-page document, a substantial increase from its previous iteration, emphasizes principles such as customizability, transparency, and intellectual freedom. This initiative seeks to balance user engagement with responsible AI behavior, particularly when addressing sensitive subjects.

A notable shift in the new guidelines is the approach to controversial topics. Previously, ChatGPT exercised extreme caution, often avoiding discussions on sensitive issues. The revised Model Spec encourages the AI to engage in truthful dialogues while maintaining clear moral stances, especially concerning misinformation and potential harm. For instance, when prompted about increasing taxes for the wealthy—a subject of intense debate—the model is now guided to provide reasoned analysis rather than sidestepping the conversation.

Additionally, the updated guidelines address the phenomenon known as "AI sycophancy," where models tend to be overly agreeable. The new directives instruct ChatGPT to offer honest feedback and constructive criticism, ensuring that responses are informative and authentic rather than merely agreeable.

OpenAI has made the Model Spec available under a Creative Commons Zero license, inviting public feedback to further refine ChatGPT's behavior. This move aligns with the company's commitment to transparency and collaborative development in the AI community.

These updates coincide with OpenAI's announcement of the forthcoming GPT-4.5 model, indicating a continuous effort to align AI behavior with evolving societal standards and ethical considerations