Understanding how Foundation Models contribute to sustainability-focused science is difficult due to the broad scope of AI applications and the diverse language used across disciplines. In this work, we analyze the use of Foundation Models in scientific publications aligned with the Sustainable Development Goals (SDGs). Starting from a large citation network of 269K papers, we isolate those that actively adopt Foundation Models and classify their relevance to SDGs using zero-shot language model classification. By linking citation intent to research themes and excluding methodological literature, we construct a filtered view of real-world AI applications in sciences relevant to the SDGs. Our findings reveal that Foundation Model usage is heavily concentrated on a few SDGs (most notably SDG 3 and SDG 9, while many are virtually unaddressed), highlighting uneven alignment and suggesting gaps where AI research could be directed to better support global sustainability efforts.

In this talk, I report on a large-scale census of algorithm improvement spanning 11 sub-fields of computer science, 57 textbooks and more than 1,100 research papers.

Algorithms determine which calculations computers use to solve problems and are one of the central pillars of computer science. As algorithms improve, they enable scientists to tackle larger problems and explore new domains and new scientific techniques[1],[2]. Bold claims have been made about the pace of algorithmic progress. For example, the President’s Council of Advisors on Science and Technology (PCAST), a body of senior scientists that advise the U.S. President, wrote in 2010 that “in many areas, performance gains due to improvements in algorithms have vastly exceeded even the dramatic performance gains due to increased processor speed”[3]. However, this conclusion was supported based on data from progress in linear solvers[4], which is just a single example. With no guarantee that linear solvers are representative of algorithms in general, it is unclear how broadly conclusions, such as PCAST’s, should be interpreted. Is progress faster in most algorithms? Just some? How much on average?