The CDS Monthly Research Feature |
Cutting-edge climate models produce useful data but currently run on supercomputers. The NYU Earth Systems Institute, which CDS Keller Professor of Applied Mathematics and Data Science Laure Zanna leads, is built on the idea that projections of a changing planet can be produced faster and on ordinary hardware, then handed to anyone with a question to ask. NYU launched the Institute on June 8 as part of its science and technology initiative, drawing together researchers who had been working on environmental problems in separate corners of the university. Its two co-deputy directors come from different schools: Donald F. Othmer Associate Professor of Chemical Engineering Miguel Modestino, who directs Tandon’s Sustainable Engineering Initiative, and Chair of the Department of Environmental Studies and CDS-affiliated Associate Professor of Environmental Studies Sonali McDermid.
|
|
|
Have you ever talked to an AI model and found the response so long and complex that you were unable to remember any of it? The model has no idea, because it remembers everything it says perfectly, and that perfect recall limits its ability to imagine how hard it might be for humans to remember certain things. In “Simulating Human Memory with Language Models,” CDS researchers show that intentionally limiting a language model’s memory creates more human-like, and therefore helpful, user simulators.
|
When a political figure posts something market-moving, every trader feels it in the same instant, and the models researchers use to imitate crowds like that mostly pretend the shared jolt never happens. CDS PhD student Grégoire Lambrecht and his adviser, Mathieu Laurière, Assistant Professor of Mathematics and Data Science at NYU Shanghai, built the first framework for learning to copy a large population’s behavior when a shock hits all of its members at once.
|
|
|
Copyright (C) 2026 Center for Data Science. All rights reserved.
|
|
|
|
