Agam Bhatia

I’m currently a MATS ScholarMATS ProgramML Alignment Theory Scholars working with Neel Nanda, psycho-analyzing language models using interpretability techniques. Before this, I did my undergrad in CS and Math at Stanford, where I’m now a master’s student.

While at Stanford, I’ve been super lucky to have worked with and been mentored by some amazing people:

• At the Scaling Intelligence LabScaling Intelligence LabStanford, I worked on algorithms for improving token efficiency during distillation with Shayan Talaei and Azalia Mirhoseini.
• Also at the lab, I worked on scaling in-context learning to improve long-context abilities with Jordan Juravsky and Bradley Brown.
• At the CoCoLabCoCoLabComputation & Cognition Lab, Stanford, I worked on modelling human dialogue by treating CoT as a latent variable with Kanishk Gandhi and Noah Goodman.

Before I got interested in research, I post-trained models for pedestrian safety at Tesla FSDTesla FSDPost-training for pedestrian safety and worked on web agents with Elad Gil.

While AFK, I like to play squash and cricket, and read non-fiction.

You can email me at: (my first name)(the calendar year) [at] stanford [dot] edu.