tl;dr Many communities benefit from a world with many AI models. Let’s do research, write specifications, and build software in order to make that world more likely.
Many models describes a world with many pre-trained AI models, including the large private models extant today as well as open-source models, public models, foundation models, narrow models, large language models, text-to-image models, and models running on edge or decentralized infrastructure.
Many models is also an open research program intended to explore, specify, and build infrastructure for a world of many models.
In this repo
This repository organizes a research program, including research questions and a (soon-to-come) regular research seminar.
Currently, our research questions are organized into three “layers”:
- a model layer, covering interactions between many models
- a data layer, covering sharing, transmission, integration, and management of aggregate data sets
- a world layer, covering applications, processes, and institutions through which data is generated and then published on the internet
Contributing
This research project is intended to coordinate a small group of researchers in AI, machine learning, computer science, and related fields. Please use GitHub pull requests to add or edit content; this will also be how we assign credit for contributions. You can also directly contact the maintainers.
Norms
Many models is a technical research program intended to improve our understanding of many models. Some of these improvements may entail or suggest overt changes to the political economy of certain industries and of the internet, especially proposals in the data and world layers. While we acknowledge these politics, we believe that progress in both the political mission and the technical research program will be faster if the two efforts are clearly distinguished and organized separately.
License
The work in this repository is licensed under the Creative Commons Attribution-ShareAlike 4.0 International license.