Ought

San Francisco, CA

winding-downFounded 2016

Ought was a 501(c)(3) nonprofit research lab founded in 2016 by Andreas Stuhlmuller to develop mechanisms for delegating high-quality reasoning to advanced machine learning systems. The organization pioneered research on factored cognition, which decomposes complex cognitive tasks into simpler, verifiable subtasks, an approach closely connected to Paul Christiano's Iterated Distillation and Amplification framework for AI alignment. Ought's flagship product was Elicit, an AI-powered research assistant that used language models to automate scientific literature review and scale up good reasoning. In September 2023, Ought spun off Elicit as an independent public benefit corporation, transferring its IP and most staff to the new entity, and has since operated with minimal activity.

Funding Details

Annual Budget: -
Monthly Burn Rate: -
Current Runway: -
Funding Goal: -
Funding Raised to Date: -
Fiscal Sponsor: -

Theory of Change

Ought believed that by developing factored cognition, which decomposes complex reasoning into smaller verifiable steps, and process supervision, which supervises how AI systems reason rather than just evaluating their outputs, they could create scalable approaches to AI alignment. The core idea was that as AI systems become more capable, humans need reliable methods to oversee and verify their reasoning processes, reducing risks from opaque optimization and goal misspecification. Through building practical tools like Elicit that demonstrated these principles in scientific research, Ought aimed to shift the AI development equilibrium from outcome-based to process-based systems, which they argued would be inherently safer and more aligned with human interests.