As part of the EMAI programme, Prof. Stjepan Picek visited the Department of Computer, Control and Management Engineering (DIAG) at Sapienza University of Rome to conduct research and teaching activities. During his visit, he delivered a seminar titled “Safety Neurons in Large Language Models.”

The talk explored recent advances in understanding how specific neurons and features within large language models may contribute to safety-aligned behaviour. It addressed methods for identifying these interpretable components, evaluating their role in model safety, and assessing the robustness of modern AI systems, including Mixture-of-Experts architectures.