We are training Oversight Foundation Models. If you're excited to contribute, please apply to join our team.
November 6, 2026
Oversight models can predict how finetuning changes a subject model's behavior
August 20, 2026
Oracles improve with model size, data size, and data quality
August 19, 2026
Elicitation ability follows predictable power laws
July 28, 2026
A vision for training a foundation model that formalizes, tests, and answers questions about AI model behavior