Part VII: Advanced and Emerging Topics

"The important thing is not to stop questioning." — Albert Einstein

AI engineering is a vast field, and not every important topic fits neatly into the narrative arc from foundations through transformers to systems. Part VII covers four advanced topics that every serious AI engineer should understand, even if they do not work with them daily.

Reinforcement learning, which teaches agents to make sequential decisions through trial and error, has become increasingly important as a technique for training and aligning language models (as we saw with RLHF in Chapter 25). Graph neural networks extend deep learning to non-Euclidean data structures — social networks, molecular graphs, knowledge bases — that cannot be represented as sequences or grids. Interpretability and mechanistic understanding address the critical question of why neural networks make the predictions they do, a question that grows more urgent as AI systems are deployed in high-stakes domains. AI safety, ethics, and governance tackle the societal implications of the technology we are building.

These chapters can be read independently based on your interests, though each builds on the foundations from Parts I–IV.

Chapters in This Part

What You Will Be Able to Do After Part VII

• Implement core RL algorithms (Q-learning, policy gradients, PPO)
• Build and train graph neural networks for node and graph classification
• Apply interpretability techniques to understand model behavior
• Evaluate AI systems for bias, fairness, and safety concerns

Prerequisites

• Part III (deep learning foundations)
• Part IV (transformer architecture, for Chapters 38–39)
• Specific chapters may reference Part VI concepts

Chapters in This Part

• Chapter 36: Reinforcement Learning for AI Engineers
• Chapter 37: Graph Neural Networks and Structured Data
• Chapter 38: Interpretability, Explainability, and Mechanistic Understanding
• Chapter 39: AI Safety, Ethics, and Governance