Understanding and Shaping the Behaviour of Generative AI for Reliable Deployment by Dr Francesco Quinzan
Abstract
Generative AI systems can write code, summarise legal documents, assist doctors, and support scientific discovery, but they can also fail in subtle and unpredictable ways, especially after deployment. This talk argues that many key safety challenges arise not during initial training, but when models are aligned, customised, and exposed to changing environments. Drawing on machine learning, decision theory, and interpretability, the speaker discusses emerging approaches to ensure reliability, alignment, and transparency in generative AI systems, with the goal of building models that can be genuinely trusted.
About the Speaker
Francesco Quinzan is a Senior Research Associate in the Machine Learning Research Group at the University of Oxford and a member of the Oxford Martin School. He is also an Affiliated Senior Research Associate at the Institute for Technology and Humanity at the University of Cambridge. He received his PhD in Computer Science from the Hasso Plattner Institute, and has held research and visiting positions at institutions including KTH Royal Institute of Technology, the Max Planck Institute for Intelligent Systems, and ETH Zurich. His research is supported by UKRI, COST, and other funding bodies.