Research overviews, technical explainers, and longer-form updates from the group.
In-Context Operator Networks acquire and apply knowledge from numerical context. A frozen model infers the predictive relation expressed by contextual examples and applies it to new queries, adapting across scientific and social systems without retraining.
Sharing our observations on how AI is reshaping the software development paradigm, along with our reflections and practices within this wave.
Starting late and compute-constrained, we placed third out of 256 participants in IGP24 by developing mathematical constructions and eventually adopting an autonomously coordinated agent ensemble. The campaign offers an early look at research organizations that can scale beyond an individual worker—and asks what greater mathematical capacity should be for as the production of mathematics becomes cheap and abundant.
AI can do more and more computation for us, but computation is not the same as information. The value of humans will increasingly concentrate on collecting information that AI cannot easily obtain, and compressing that information into problems that AI can use.
A short account of the research path leading to the current work, from uncertainty quantification and Bayesian inference for PDEs, to optimal transport, particle dynamics, and eventually ICON.