Developing the Mathematical Foundations of Explainability and Using Them to Catch an LLM in a Lie

Explore the mathematical foundations of explainability in machine learning and discover how these principles can be applied to detect deception in large language models in this 28-minute conference talk from the Computational Genomics Summer Institute. Delve into cutting-edge research that bridges theoretical explainability concepts with practical applications in AI systems, examining how mathematical frameworks can provide insights into model behavior and trustworthiness. Learn about innovative approaches to understanding and interpreting complex AI models, with particular focus on identifying when language models may be providing misleading or false information. Gain insights from related research spanning transformer-based AI applications in single-cell clinical data analysis, AI-enhanced electrocardiography for diabetes assessment, deep learning architectures with equivariant properties, and machine learning applications in magnetic resonance imaging for disease identification.