Hacking LLMs - An Introduction to Mechanistic Interpretability

Explore the inner workings of Large Language Models through this 30-minute conference talk that introduces the cutting-edge field of Mechanistic Interpretability. Discover how researchers dissect neural networks to decode the black box of LLMs into understandable, human-readable components and uncover the mechanisms behind their behavior. Learn foundational concepts of what Mechanistic Interpretability is and why it matters for ensuring models behave safely and ethically, optimizing performance, and fostering trust in AI systems. Gain practical insights into tools and techniques using Python libraries like PyTorch, Transformers, and interpretability-specific tools. Understand how to start applying these concepts in your own work, whether you're a researcher, developer, or curious enthusiast interested in one of the most exciting frontiers in AI. The presentation assumes familiarity with AI fundamentals but introduces advanced concepts with approachable explanations, requiring no specialized hardware or prerequisites beyond curiosity about understanding how LLMs "think" and process information.