These are my notes as an attendee at the ‘Workshop on Uncertainty in Machine Learning (WUML2026), Tartu’, held from 2 to 4 February 2026.
A three-day international workshop in Tartu focusing on uncertainty in AI systems and machine learning.
What I took away
Hallucinations and uncertainty are not simply random noise. They can be better understood, measured and accounted for through checks in a workflow.
How I use this in my work
I use these insights in training when explaining when to trust AI output, when to check it and why confidence does not always mean correctness.
Event information: Workshop on Uncertainty in Machine Learning (WUML2026), Tartu.
The ChatGPT and Gemini image-generator comparison provides another example of why output needs checking. Editing an initially convincing image revealed artefacts that can be easy to miss in a small preview.