autoregressive-diffusion

Worked on autoregressive diffusion for video generation @ Decart.

open-source-rlhf

Automated RLHF pipelines to align LLMs w/ live human feedback @ MIT CSAIL. Advised by Prof. Jacob Andreas and Prof. Leshem Choshen. In collaboration with Cohere and Hugging Face.

reward-hacking

Studied how CoT monitoring awareness effects reward hacking behavior in coding agents.

adversarial-feedback

Researched how adversarial preference signals shift model alignment + behavior; stress-testing LLMs under adversarial feedback.

jenbenarye.com

You're here.