In software development, “model code” or “example code” is a short snippet of real software code that can demonstrates a particular concept. In our AI Model Code, most examples will be written in the Python programming language. Below, you can see specific technical concepts and examples that are linked to the statutes. If you’d like to dive into the model statute, you can find it here.

Evaluating predictive models for disparate performance

This section covers how to evaluate a model for disparities across different groups, which is a requirement of Section 2.1 of the model statute.

Evaluating language models for accuracy and bias

This section covers techniques for evaluating language models, which require a different set of tools because they are generative rather than discriminative. These techniques can be useful for the requirements in Section 2.4.

Tools for model documentation

This section covers techniques for documenting models, which are useful for several sections, including Sections 2.1, 2.2, and 3.

Updated: