Research & Projects
Projects
FLOP Around And Find Out: LLM Training Workload Size
A post on estimating the size of an LLM training workload from system-level signals on a dual-V100 node.
Estimating LLM Training FLOPs on the NVIDIA Jetson Orin Nano
A report on estimating FLOPs used in LLM training runs on the Jetson Orin Nano edge device, using only system-level metrics and without access to model code or logs. My work at UChicago XLab currently involves extending this to a dual-V100 GPU setup.
AI Incident Tracker: June 2026 Update
A blog post covering progress on the AI Incident Tracker, a tool that uses LLMs to classify news reports of AI incidents.
Policymakers Should Understand the Future of Decentralized Training
A report written with Konstantin Pilz at Pivotal Research on the technical feasibility and progress of decentralized training for frontier AI models, translating the bandwidth and engineering constraints for a policy audience and proposing metrics policymakers should track.
Publications
-
Encoding Consistency: Optimizing Self-Driving Reliability With Real-Time Speed Data
Proceedings of the 4th Workshop on Flexible Resource and Application Management on the Edge, 2024 -
AutoLearn: Learning in the Edge to Cloud Continuum
SC-W '23: Proceedings of the SC '23 Workshops of The International Conference on High Performance Computing, Network, Storage, and Analysis, 2023
Before getting involved in AI Safety, I did ML research with the Chameleon Cloud team at the University of Chicago.