The Model Spec Feedback
On May 8th, we announced the Model Spec, a document that specifies our approach to shaping desired model behavior and how we evaluate tradeoffs when conflicts arise. The Model Spec (opens in a new window)includes three types of principles:
Objectives: Broad, general principles that provide a directional sense of the desired behavior
Rules: Instructions that address complexity and help ensure safety and legality
Default behaviors: Guidelines that are consistent with objectives and rules, providing a template for handling conflicts and demonstrating how to prioritize and balance objectives
The Model Spec is not exhaustive, and we expect it to change over time. As a first step, we invite you to share your initial reactions to the objectives, rules, and default behaviors through the form below.
This form will be open until May 22nd, and your early thoughts will help us develop a robust process for gathering and incorporating feedback across a wide range of stakeholders—including users, developers, policymakers, trusted institutions, domain experts, and more.
Over the next year, we will share updates about changes to the Model Spec, our response to feedback, and how our research in shaping model behavior is progressing.