Preface


Part I The Foundations of Generative AI


Chapter 1

Ten Breakthroughs That Made Generative AI Possible


Breakthrough 1: The Turing Machine


Breakthrough 2: The Artificial Neuron


Breakthrough 3: The Dartmouth Conference


Breakthrough 4: The Perceptron


The Rise of Symbolic Reasoning (1960s)


The First AI Winter (Early 1970s to Early 1980s)


Breakthrough 5: Neural Networks and Backpropagation


Breakthrough 6: Recurrent Neural Networks


The Second AI Winter (Late 1980s to Mid-1990s)


Breakthrough 7: Invention of the GPU


Breakthrough 8: Reinforcement Learning


Breakthrough 9: Language Modeling


Breakthrough 10: The Transformer


Summary


References

Chapter 2
The Machinery of Learning


Types of Learning


Supervised Learning


Unsupervised Learning


Reinforcement Learning


The Machine Learning Family Tree


What Is a Model?


How Models Are Trained


Training, Validation, and Test Datasets


Inference Models


How to Measure Model Accuracy


Hyperparameters


Summary

Chapter 3
Foundational Algorithms


Linear Regression: One Stroke to Represent the Data


Describing a Line


Loss Functions and Other Hyperparameters


Classification


Support Vector Machines


Discovering Structures in Data


K-Means, the Clustering King


DBSCAN and Growing Clusters


Summary

Chapter 4
An Introduction to Neural Networks


Neural Networks Key Concepts


ANNs: General Structure and Terminology


Training a Neural Network


Training Models and Overcoming Challenges


The Importance of Clean Data


Labeled Data: The Backbone of Supervised Learning


Avoiding the Pitfalls: Overfitting and Underfitting


Scaling Up Training


Summary

Chapter 5
Neural Network Architectures


Feedforward Neural Networks


Traditional FFNs


Convolutional Neural Networks (CNNs)


Traditional Generative Models


Generative Adversarial Networks (GANs)


Variational Autoencoders (VAEs)


Diffusion Models


Recurrent Models


Recurrent Neural Networks (RNNs)


Long Short-Term Memory Networks (LSTMs)


Summary

Chapter 6
Reinforcement Learning: Teaching Machines to Learn by Trial and Error


An AI That Learns Like Us


Key Concepts of Reinforcement Learning


The Markov Decision Process (MDP)


The Bellman Equation


Model-Based Versus Model-Free Systems


On-Policy Versus Off-Policy Learning: Two Paths to Learning


Monte Carlo Reinforcement Learning


Temporal Difference (TD) Learning


Q-Learning


Deep Reinforcement Learning


Summary


References


Part II The Generative AI Revolution


Chapter 7

Language Modeling: The Birth of LLMs


An Introduction to LLMs


Foundations of Language Modeling


Next-Word Prediction


From Words to Tokens


Word Embedding: Turning Tokens into Numbers


How Word Embeddings Are Learned


Semantic Relationships in the Embedding Space


The Semantics of Language


Summary


Reference

Chapter 8
Attention Is All You Need: The Foundation of Generative AI


A New Architecture Begins to Take Shape


Attention Is All You Need


From Sequential to Parallel Processing


Positional Encoding


The Self-Attention Mechanism


Summary


References

Chapter 9
Attention Isnt All You Need: Understanding the Transformer Architecture


The Encoder Block


The Multi-Head Attention Layer


The Add and Norm Layers and Residual Connections


The Feedforward Network (FFN) Layer


Layers Upon Layers of Encoder Blocks


How Encoders Are Trained


The Decoder Block


The Decoders Output Classifier


How Decoders Are Trained


What Type of Machine Learning Is Involved in Training LLMs?


Case Study: The GPT-3 Transformer


Future Directions


Summary


References


Part III Living with Generative AI


Chapter 10

Making Models Smarter: Prompt and Context Engineering


Prompt and Context Windows


Prompt Engineering Techniques


Shot-Based Approaches


Chain-Based Approaches


Self-Ask Approaches


Prompt Engineering Limitations


Context Engineering


Types of Contexts in LLM Workflows


Tools and Protocols


Context Design Techniques


Summary

Chapter 11
Retrieval-Augmented Generation


The Need for RAG


Common Applications of RAG


RAG Trends and Practices


The RAG Pipeline


Query Formulation


Retrieval Filtering


Working with Knowledge Databases


Loading Documents


Chunking: Splitting Documents


Embedding and Storing Segments


Retrieving Segments


Summary

Chapter 12
Fine-Tuning LLMs


The Need for Fine-Tuning


Comparing Fine-Tuning and RAG


Inference Hyperparameter Tuning for LLMs


Temperature


Top-K Sampling


Top-P (Nucleus) Sampling


Repetition Penalty


Principles of Fine-Tuning with New Data


Fine-Tuning for Model Types and Objectives


Supervised Fine-Tuning (SFT)


Transfer Learning


Parameter-Efficient Fine-Tuning (PEFT) Methods


Retrieval-Augmented Fine-Tuning (RAFT)


Reinforcement Learning from Human Feedback (RLHF)


Benchmarking Model Performance


Summary


References

Chapter 13
Securing LLMs from Attack


What Makes AI Security Different


The Emergence and Importance of AI Security Frameworks


NIST AI Risk Management Framework


The OWASP Top 10


MITRE ATLAS


The ISO/IEC Suite of AI Standards


A Comparison of the AI Security Frameworks


AI Vulnerabilities and Attack Vectors


Direct Prompt Injection Attacks


Prompt Injections with Jailbreaking


Indirect Prompt Injection Attacks


Extraction and Inversion Attacks


AI Supply Chain Threats


Defending Models from Attack


Extending the Guardrail System


Architectural Safeguards


Continuous Monitoring and Detection System


Generative Adversarial Defense Techniques


Summary


References

Chapter 14
AI Ethics and Bias: Building Responsible Systems


Bias and Ethical Risks in GenAI


The Biased Data That Shapes GenAI


The Difficulty of Stopping GenAI Bias


When Generative AI Goes Wrong: Unethical and Harmful Outputs


Hallucination and Misinformation


Synthetic Media, Deepfakes, and Disinformation


Ownership, Consent, and Copyright


Transparency, Explainability, and Trust


Building Responsible Generative AI


Alignment and AI Safety


Practical Responses to Ethical AI Challenges


Summary


References

Chapter 15
The Future of AI: From Generative to General Intelligence


Where We Stand: A Snapshot of Todays Capabilities


Current Limitations and Known Pain Points


The Emergence Question: Are We Seeing Sparks of AGI?


What Makes AGI Different?


What Is AGI?


What Is Intelligence Anyway?


Do Reasoning LLMs Really Reason?


Predicting Versus Understanding


Are We Already on the Path to AGI?


Paths to AGI


The Scaling Hypothesis


The Modular Hypothesis


The Embodied System Hypothesis


Hybrid Models


Is AGI the End of Humanity?


The Alignment Problem Revisited


Black Boxes and Loss of Interpretability


The Singularity and the Skynet Problem


Controlling Existential Risks


Is AGI Helping or Hurting Society?


Will AI Take Your Job?


Education in the Age of Generative AI


Societal Identity and Stability


The Evolving Voice of Generative AI


From Single-Goal Prompting to Multimodal Partnering


Responsive Interfaces


Redefining Creativity


The Future We Choose


Scenario A: The Co-creative Society


Scenario B: The Automated Present


Scenario C: The Disrupted Path


Summary


References

Appendix...