Thomas Breuel, Distinguished Engineer at NVIDIA Research and a leading expert in machine learning, document recognition, and computational models of intelligence, will teach “Facts and Rules in LLMs” at AIces 2026. His course explores how large language models represent factual knowledge, perform reasoning, and interact with symbolic structures.
This course offers an accessible technical foundation for understanding LLM behavior, limitations, and emerging approaches to reasoning.
Key topics covered:
- Knowledge Representation in LLMs: How transformer architectures encode and retrieve factual information, and what governs model generalization.
- Symbolic vs. Neural Reasoning: Examination of hybrid systems that integrate neural and logical components for more interpretable and robust reasoning.
- Reasoning Benchmarks and Patterns: Analysis of chain-of-thought prompting, logic tasks, and mathematical reasoning benchmarks used to evaluate LLM performance.
- Interpreting Model Behavior: Methods for examining how models arrive at answers, and what these insights reveal about their internal structure.
Dr. Breuel’s course will help participants move beyond black-box interpretations, offering a deeper conceptual understanding of how modern language models work.
Those interested in reasoning, interpretability, or the mechanics of LLMs will benefit greatly fromthis course.
Join AIces 2026 to learn more: https://aices.irdta.eu/2026/registration/
Event details: https://aices.irdta.eu/2026/
