From Zero to Synthesis

What happens between the moment you type a question and the moment a fluent answer appears? How can the same system write a poem and then invent a court case that never existed? From Zero to Synthesis explains generative AI from the ground up, with real stories, original figures and no technical prerequisites.

This is not a book of tricks for a particular app. Apps change every few months. The ideas underneath them change far more slowly, and once you understand them you can judge any new tool on your own.

Who it is for

Anyone who uses AI and wants to understand how it works. No technical background is required. Instructors and students. The book is designed to serve as a course textbook. Technical readers. Mathematics and code wait in skippable boxes for those who want them.

What is inside

Part I, Foundations: the language of generative AI, inside the transformer, diffusion models, GANs, VAEs and diffusion compared, open and closed models, and the hardware underneath. Part II, How Models Learn, Measure and Expand: pretraining, fine-tuning and RLHF; benchmarks and their limits; multimodal AI; large and small language models; explainability. Part III, The Ecosystem in the Real World: the economics of AI, agents, deployment and inference, data, and the limits of AI. Part IV, Society, Mind and the Future: deepfakes and the post-truth age, a career map, AI and human cognition, AGI and its risks, and where to build your expertise. Part V, The Field Manual: seven ways to stay in control of AI in daily work, with a ready-to-use master prompt template.

What makes it different

Every chapter opens with a real, sourced story: the eight authors of the transformer paper, the AI-made portrait sold at Christies, the lawyers who cited court cases that did not exist, and the deepfake executives on a video call that led to a multimillion-dollar transfer. 21 case studies and 150 original figures, some of them showing real output from small models trained for this book. 41 Under the Hood boxes hold the mathematics and Python code, all of which has been run. You can skip any of them without losing the thread. Every chapter ends with a summary, key concepts, review and discussion questions and an activity. The 251 quiz questions have explained answers at the back of the book. A 275-term glossary and a timeline of 59 milestones in generative AI. Every factual claim rests on a source that was opened and read. Fast-changing facts (models, prices, laws) are dated "as of late 2026".

For instructors: the book includes a suggested fifteen-week course plan, and an Instructors Companion with model answers and additional exam items is available to instructors who adopt it.