Photograph of Nisan Stiennon

Nisan Stiennon

Mathematics and AI safety

Email · Google Scholar

Research interests

I work on agent foundations, the mathematical theory of knowing and choosing. The goal is a good future where artificial intelligences are aligned and cooperative with humans. My current work is in open-source game theory, applying tools from topology and domain theory.

Previously I helped work on alignment of large language models at OpenAI. Before that I did a PhD in algebraic topology.

Selected work

Learning to summarize from human feedback

Nisan Stiennon, Long Ouyang, Jeff Wu, Daniel M. Ziegler, Ryan Lowe, Chelsea Voss, Alec Radford, Dario Amodei, Paul Christiano · NeurIPS, 2020

"Step 0" of scalable oversight is applying RLHF to LLMs. See here for a discussion of whether this project was good or bad for the world.

An anytime proof of the Brouwer fixed point theorem

Notes from an expository talk about Brouwer's fixed point theorem.