I work on understanding neural models (language models, vision models, game-playing models), both their internals (more so during my PhD; i.e., interpretability) and their behavior (more so in my professional role; i.e., evaluation). I completed my PhD at Brown University in 2023 and thereafter have worked as a research scientist at Kensho Technologies.
Research
Brief breakdowns of some of my research.
On Finding Inconsistencies in Documents
Language Models Struggle With Numeric Calibration
Training Priors Predict Text-To-Image Model Performance
Evaluation Beyond Task Performance
Predicting Inductive Biases of Pre-Trained Models
Towards Interpretable Reinforcement Learning
Listicles
Artifacts
Visualizations and interactive notebooks.
Exposition
I found it educational to walk through some papers/concepts. Or rather, mostly, I liked trying to make the figures nice.
Notes
Code tricks that I found useful during my PhD. Nowadays, though, Claude has got you covered!
Haiku Merge Params
Auto-Format Python on Save
3-D Indexing with PyTorch
Pretty-Print Pandas in Notebooks
Build a DefaultDict from a Dict
Fix "Too Many Values" for Seaborn
Fix "Too Many Files" for Bash
Git on a Server
Readable File/Folder Sizes
Number of Batches
Compile LaTeX with a Bibliography