We are excited to welcome Lynton Ardizzone, PhD in machine learning from Heidelberg University and independent ML/AI consultant, to our joint heidelberg.ai / NCT Data Science Seminar on 24th June at 5 PM.
In this in-person event, he will present a tour of uncertainty quantification for large language models, examining why classical UQ assumptions break down for these systems and what state-of-the-art approaches are doing about it.
We look forward to your participation in this seminar, where you will come away with a clearer sense of how far current methods get us, and what it would actually mean for a language model to know what it doesn't know.
Abstract
Large language models are known for their unwavering confidence — they recite true facts with the same fluency and authority as hallucinated nonsense. For anyone relying on these systems in domains where being wrong carries consequences, this is a core problem.
Uncertainty quantification has a long and principled history in science and engineering. We know what it means for a prediction to come with calibrated error bars, and we know how to verify that those bars are honest. But LLMs break most of the assumptions classical UQ relies on. Their outputs are discrete, high-dimensional, and compositional. How do you quantitatively measure doubt in such systems?
This talk introduces the core ideas of uncertainty quantification: what it is, how it is measured, and why it matters, grounding them in examples from scientific and biomedical applications. From there, we confront what makes LLMs challenging for this case and explore state-of-the-art approaches that attempt to construct faithful uncertainty estimates from these models. Rather than surveying individual methods, the goal is to build intuition for each family of approaches and the reasoning that motivates them.
A language model that knows what it doesn't know would be a fundamentally different kind of tool — one that can guide experimental design, flag its own limitations, and participate honestly in scientific reasoning. This talk takes stock of how far current approaches get us and where they fall short.
Biography
Dr Lynton Ardizzone holds a PhD in machine learning from Heidelberg University, where he developed deep generative models with built-in uncertainty quantification for inverse problems. His thesis on conditional invertible neural networks was awarded the Ruprecht-Karls-Prize. His methods have found applications in medical imaging, astrophysics, particle physics, and mechanical engineering. From 2022 to 2025, he was Head of Machine Learning at Copresence, an AI startup for 3D avatar reconstruction. He now works as an independent ML/AI consultant, helping startups and smaller teams assess feasibility, prototype, and make technical decisions around AI.
Event Info
Please help us plan ahead by registrating for the event at our
meetup event-site
.
| What? | Measuring Doubt in Systems That Have None: Uncertainty Quantification for LLMs |
|---|---|
| Who? | Lynton Ardizzone, Independent ML/AI Consultant |
| When? | June 24th 2026 @ 5pm |
| Where? | DKFZ Communication Center (K1+K2), Im Neuenheimer Feld 280 |
| Registration | meetup event-site |