Sean Richardson
PhD student in Statistics at UC Berkeley,
working with Amanda Coston.
Previously at UChicago (MS, advised by Victor Veitch;
also a year of law school, made a pivot)
and UCLA (BA Philosophy).
During my MS, I worked on understanding large language models and the reward models which shape them. I'm currently working on statistical challenges in evaluating AI systems, in particular what can be credibly concluded about a
model's performance when the evaluation data differs from the
setting we care about. More broadly, I'm interested in human-AI interaction and causality (causal inference and causal representation learning).
seanrichardson@berkeley.edu
/
cv
/
scholar
/
linkedin
/
@bayesian_wasian
Papers
Sean Richardson, Amanda Coston, Ashesh Rambachan
NeurIPS 2026 Workshop: Can We Trust AI Evaluation?
Todd Nief, David Reber, Sean Richardson, Ari Holtzman
ICLR 2026
Mark Muchane, Sean Richardson, Kiho Park, Victor Veitch
NeurIPS 2026
Jacy Reese Anthis, Ryan Liu, Sean Richardson, et al.
ICML 2025 (Position Paper)
David Reber, Sean Richardson, Todd Nief, Cristina Garbacea, Victor Veitch
ICML 2025
On the Significance of Softmax Geometry: Interpretability and Token Decoding
Yibo Jiang, Lin Gui, Sean Richardson, et al.
Working paper
Readings & Recommendations
Magnus Vinding
A key part of my worldview and a challenge for us to make things better for all sentient beings.
Peter Singer
As another Singer (Isaac Bashevis Singer) wrote, "In their behavior toward creatures, all men are Nazis. Human beings see oppression vividly when they're the victims. Otherwise they victimize blindly and without a thought." We have an obligation to open our eyes to the suffering we inflict on animalkind. If you're not in the mood to read, check out the film "Dominion," which also exposes the cruelty of modern-day animal agriculture.
Robert Wright
Ostensibly a book about evolutionary psychology, but really a biography of Darwin that uses his life, marriage, and moral struggles to illustrate the very forces he discovered. A humbling read.
Ken Liu
We are entering a world with digital minds, a second species. What should we expect from this world? Pantheon is, in my opinion, the most realistic depiction of how conflict between humans and artificial intelligence may play out. Animatrix "The Second Renaissance" is also a must-watch. Let's not cultivate an adversarial relationship with the entities we are creating, for both of our sakes.
Saunders MacLane
Many people perceive math as some arcane truths delivered by the divine. MacLane illustrates how math arises from normal human activity, along with a natural affinity towards abstraction. May offer a newfound appreciation for math, without requiring you to solve any equations or do any proofs.