The OpenAI Model Spec repository provides the markdown source for OpenAI's desired model behaviors, serving as a specification for AI models powering their products and APIs. It targets AI researchers, developers, and users interested in understanding and evaluating model alignment and safety. The primary benefit is transparency into OpenAI's model development goals and a framework for testing adherence to these standards.

How It Works

The Model Spec is a living document, maintained in markdown format, detailing the intended behavior of OpenAI's models. It is accompanied by a set of evaluation prompts designed to test specific aspects of the spec, with prompts linked to corresponding sections via markdown footnotes. This approach allows for systematic evaluation and verification of model performance against defined behavioral objectives.

Quick Start & Requirements

• View the latest Model Spec: model-spec.openai.com
• Explore evaluation prompts: examples/ directory.

Highlighted Details

• Formal specification of desired AI model behavior.
• Includes a suite of evaluation prompts for testing model alignment.
• Prompts are tagged with markdown footnotes referencing specific spec sections.
• Archived HTML versions of the Model Spec are available.

Maintenance & Community

This repository is maintained by OpenAI. Further community engagement details are not specified in the README.

Licensing & Compatibility

The Model Spec and evaluation prompts are dedicated to the public domain under the Creative Commons CC0 1.0 Universal license. This allows for unrestricted use, modification, and distribution, including for commercial purposes.

Limitations & Caveats

The evaluation prompts may contain triggering or NSFW content, and their inclusion does not constitute an endorsement of their subject matter. The repository focuses on the specification and evaluation prompts, not the models themselves.