Hi! I'm Aryaman.
I desperately want to understand how language models work. What mechanisms are encoded in their weights that allow human-like command of natural language, recall of facts, and the ability to learn in-context? How are these and other skills learned in training? To that end, I am inspired by work in NLP, ML, causality, information theory, and psycholinguistics.
Before coming to Stanford, I completed my B.S. in Computer Science and Linguistics at Georgetown University, where I was mentored by Nathan Schneider. I also work closely with Ryan Cotterell at ETH Zürich. In 2022, I spent the summer at Apple working with Robert Daland, and winter at Redwood Research.
In the old days, I worked largely on computational linguistics.
# Selected papers
- Localizing model behavior with path patching
arXiv, arXiv:2304.05969 • 2023 Apr 12
- Estimating the entropy of linguistic distributions
ACL • 2022 Apr 04
- 2023-09-14: Moved to the San Francisco Bay Area.
- 2023-08-11: Workshop paper at CLVL on causal tracing for vision-language models accepted.
- 2023-07-31: Back from the Leiden University Summer School in Languages and Linguistics!
- 2023-05-24: Two workshop papers accepted:
- Jambu: A historical linguistic database for South Asian languages (SIGMORPHON)
- Unified syntactic annotation of English in the CGEL framework (LAW)
- 2023-02-08: Accepted to the Ph.D. program at Stanford CS!
- 2022-12-18: Headed to Berkeley, CA for a 5-week internship at Redwood Research on mechanistic interpretability.