Rust might seem like a newcomer on the scene but today it is heavily used in production. Combining memory safety and interoperability with top tier speed, Rust is a great choice for low-resource and green solutions. This talk takes a look at where and when Rust is the right tool for the job, navigating the landscape of big data bioinformatics and modern research pipelines while sharing the journey of building a community around an open source crate package.
Since its inception, the Turing Test - the seemingly audacious vision of machines that speak with such fluency that they blur the boundary between human and artificial minds - has served as the north star of language understanding and AI. In the era of large language models, this vision feels tantalizingly close, yet increasingly hollow, as AI systems now routinely surpass humans not only in language-related tasks but across a variety of complex domains.
As coding agents move from autocomplete to autonomous software engineering, they need an execution substrate designed for non-human developers. This talk will explore the systems challenges behind forkable, isolated environments, including filesystem state, networking and credentials, reproducibility, fast cloning, build/test execution, recovery, and observability. The focus will be on the underlying architecture, trade-offs, and open systems problems.
Recently a large community of puzzle designers has emerged, empowered by online tools for creating and sharing puzzles, and especially around the YouTube channel Cracking the Cryptic. Many of these puzzles are modern variants of Sudoku, with additional rules like thermometers, palindromes, knight's-moves, and pentominoes, and arithmetic as well as logical constraints. I’ll describe my experiences with designing puzzles in this genre, what I think makes a puzzle good, and how this differs from the usual notions of hardness in computational complexity.
Abstract:Multimodal models can reason over images, text, and sequences, but evaluating only their final answers reveals little about how those answers are produced.
Recently a large community of puzzle designers has emerged, empowered by online tools for creating and sharing puzzles, and especially around the YouTube channel Cracking the Cryptic. Many of these puzzles are modern variants of Sudoku, with additional rules like thermometers, palindromes, knight's-moves, and pentominoes, and arithmetic as well as logical constraints. I’ll describe my experiences with designing puzzles in this genre, what I think makes a puzzle good, and how this differs from the usual notions of hardness in computational complexity.
Professor Michael Elliott, University of Michigan School of Public Health
Longitudinal data has become a major part of the landscape for clinical and epidemiological research. While variance is typically understood as nuisance – the “noise” in “signal-to-noise” – there is increasing evidence that underlying variability in subject-level measures over time may also be important in predicting future health outcomes of interest.
Matthew McCormack - Software Developer, National Oceanography Centre
BODC (British Oceanographic Data Centre) receives data with a diverse range of sources, formats, and content. Previously this was a heavily manual process requiring data managers to invest significant time into each dataset to standardise the format, apply controlled vocabularies, and apply QC to the data. This also involved moving the data across multiple file systems and disparate software tools.
Christina Matteotti, Head of Universities, EMEA Partnerships, Google
The Beyond Academia: AI & Data Science Applications seminar series at the University of Cambridge explores the transformative role of data and artificial intelligence in the world. Designed for postgraduate researchers, early-career academics, and students, the series bridges the gap between high-level research and real-world deployment.
Dr Francesco Paolo Casale - Human Technopole, Milan, Italy; Helmholtz Munich, Munich, Germany; Faculty of Informatics, Technical University of Munich, Garching, Germany
Talk title: AI in Human Genetics: Linking Genetic Variation to Phenotypes Across Scales
The Beyond Academia: AI & Data Science Applications seminar series at the University of Cambridge explores the transformative role of data and artificial intelligence in the world. Designed for postgraduate researchers, early-career academics, and students, the series bridges the gap between high-level research and real-world deployment.
LLM agents are increasingly deployed on long-horizon tasks with tool use, irreversible actions, and unpredictable feedback. Yet we have few principled ways to tell, mid-episode, whether an agent is on track or quietly failing. Most uncertainty quantification (UQ) research still centers on single-turn QA, a poor match for interactive agents. In this talk, I'll present a general formulation of agent UQ and the challenges unique to agentic settings, from choosing uncertainty estimators to modeling how uncertainty evolves over an interaction.
