Do Models Read What They Write? Causal Registers in Scratchpad Reasoning
B. Shih, J. Winnicki, and E. Darve
Benjamin Shih
I am a graduate student at the Stanford Institute for Computational and Mathematical Engineering. My current research is in mechanistic interpretability, reverse-engineering the internal computations of neural networks to understand how they represent information and produce their behavior. In Fall 2026, I will join Jump Trading as a Quantitative Researcher in New York.
Previously, I received an Sc.B. with Honors in Applied Mathematics–Computer Science and an A.B. in Mathematics from Brown University.
Research interests
- Scientific ML
- Mech interp
- Theoretical ML
Selected research
Transformers as Neural Operators for Solutions of Differential Equations with Finite Regularity
B. Shih, A. Peyvan, Z. Zhang, and G. E. Karniadakis
Research overview
At Stanford, I work in the DASH Lab with Eric Darve on mechanistic interpretability, across several projects on how neural networks compute internally. Previously at Brown, I worked in the CRUNCH group with Zhongqiang Zhang and George Em Karniadakis on neural operators for differential equations.
CV
Complete CV available upon request. Please contact me at benjamin.shih@stanford.edu.