Be the first open model builder to demonstrate alignment with open source communities

AI models are built on the labor of open source communities in two distinct ways: as a dependency of the technology (the frameworks, libraries, and tooling models and their infrastructure run on) and as the content they are trained on (the code, documentation, and discussion absorbed into the models). Yet those communities have no way to feed back into the models the way paid annotators and synthetic data do. This places maintainers and communities in a largely reactive state.

In the CHAOSS AI Alignment working group we've drafted a metric model (a collection of metrics) to close that loop: Community-Aligned Training (Feedback). This metrics model is an articulation of criteria needed for alignment between model builders and open communities they depend on.

We are a group of experts in open source, open education, open science, software ethics, engineering, data science and more. What we need is someone actively building open models or open weights to share your experiences, expertise - to collaborate, and in return we will be your bridge to at least one OSS community to pilot a first alignment effort.

This is our draft. This is how to connect (Slack, GitHub issue) or you can also just reach out directly to me: emma.irwin at gmail.com with questions. Excited to meet you!

We will also be at FOSSY - come find us!