This project presents a first-principles reconstruction of the GEN-1 model, grounded in publicly available evidence from Generalist AI—including technical blogs, demo analyses, and insights shared by Pete Florence and collaborators.
Our objective is to systematically reverse-engineer and extend the ideas behind GEN-1 toward a unified physical foundation model—one that natively integrates perception, action, and interaction with the real world.
We are actively developing:
- research write-ups that distill core architectural and training insights
- prototype models and training pipelines
- small-scale experimental validations, with a roadmap toward larger-scale training
This is an open and evolving effort. We aim to make both the reasoning and the systems transparent, and to iterate quickly based on empirical results.
If you’re interested in contributing or collaborating, feel free to reach out:
- WeChat: jamieworkingonai
- Email: jiaming@comp.nus.edu.sg