Projects & Publications
Research
My work lies at the intersection of understanding models and applying them. I am interested in investigating internal model representations and in using generative models to make scientific inference more efficient.
2026Interpretability · Geometry & FoundationsCount Me If You Can: Geometric Failure Modes in Language Model Counting
Nicholas Bai, Ayushi Mehrotra
Traces language-model counting errors to the geometry of their internal number representations, and improves high-count accuracy by up to 20%.
ICML 2026 WorkshopRead more
2026AI for Science · Generative ModelsDiffusion Guided Particle Sampling for Bayesian Inverse PDE Problems
Nicholas Bai
Leveraging diffusion priors to recover the full posterior for inverse PDE problems, instead of a single MAP estimate.
OngoingRead more
2025InterpretabilityInterpreting Neurons in Deep Vision Networks with Language Models
Nicholas Bai, et al.
A language-model-based framework that produces natural-language descriptions of individual neurons in vision networks.
2025Geometry & FoundationsCAT: Curvature-Adaptive Transformers for Geometry-Aware Learning
R. Y. Lin, S. Ojha, Nicholas Bai
Transformers that route tokens across three geometric attention branches with a mixture-of-experts gate.
2025Generative Models · InterpretabilityFast Steering of Diffusion Models using Inversion-Anchored Concept Activation Vectors
Nicholas Bai, Akshay Kulkarni, Tsui-Wei Weng
Inversion-anchored concept activation vectors that steer diffusion models without retraining.
PreprintRead more
5 projects shown.