Tim Dettmers

2024 Early Career Fellow

Tim Dettmers is an Assistant Professor at Carnegie Mellon University and a Research Scientist at the Allen Institute for AI. His work focuses on making coding agents, such as Claude Code, accessible to researchers and practitioners by reducing their resource requirements. His main focus is to develop open-source agent systems that are competitive with closed-weight agents and can be run on consumer hardware, such as laptops. His research won oral, spotlight, and best paper awards at conferences such as ICLR and NeurIPS and was awarded the Block Award, Madrona Prize, and the Google ML and Systems Junior Faculty Award. He created the bitsandbytes open-source library for efficient foundation models, which is growing at 4.8 million installations per month, and for which he received Google Open Source and PyTorch Foundation awards.

AI2050 Project

AI models like ChatGPT work well for general use but fail in specialized expert domains, such as the medical sciences. To make AI models work in expert domains, one must adapt them, which is costly and requires significant AI expertise. This project overcomes these cost and expertise barriers through two new approaches: (1) use AI models themselves to perform the AI model adaptation process automatically; (2) make the adaptation process cheap so it can be run on regular consumer hardware. With this solution, non-AI scientists can adapt AI models to expert domains without needing AI expertise or large computational resources.

Affiliation

Assistant Professor, Carnegie Mellon University

Hard Problem

Capabilities