Neil Pagarkar Rathi

Hi. I am an undergraduate at Stanford advised by Dan Jurafsky and a safety fellow at Anthropic with Alec Radford.

I work on basic science for AI safety: why misalignments occur, and how we can prevent them during model training.

In a past life, I did computational psycholinguistics. I also consume media and do math outreach.

Selected Publications

• Humans overrely on overconfident language models, across languages. COLM. [pdf]
• Mechanistic evaluation of Transformers and state space models. arXiv. [pdf]
• Soft production preferences emerge from a bottleneck on memory. CogSci. Computational Modeling Prize. [pdf]
• TopoLM: brain-like spatio-functional organization in a topographic language model. ICLR. Oral. [pdf]

You can reach me at lastname [at] stanford [dot] edu.