LLMs, Made Intuitive

A Clear Mental Model of Transformers, Attention, Tokenization, Training, Inference, and Reasoning

von D. K. Harlan

444 Seiten, Taschenbuch

Kurzbeschreibung des Verlags

What actually happens inside a large language model after you press Enter?An LLM can explain an idea, write code, translate languages, follow examples, generate arguments and solve unfamiliar problems. Yet explanations of how these systems work often collapse into one of two claims: "It is only autocomplete."or"It thinks like us."Both miss the machinery that matters.LLMs, Made Intuitive builds a clear mental model of modern language models from the ground up, without requiring a machine-learning background.Beginning with next-token prediction, D. K. Harlan explains how text becomes tokens, how tokens become numerical representations, how attention moves information through a Transformer, how training turns prediction error into capability, and what happens during inference when a model generates an answer one token at a time.The book then goes further into the questions that matter when LLMs leave the laboratory and enter real work: Why do models hallucinate?What changes during post-training?Why can additional inference-time computation improve reasoning?How do retrieval and tools alter what a model can do?Why does prompting work without retraining the model?What does an LLM actually "know"?How should we decide when its output can be trusted?The result is a connected explanation of tokenization, embeddings, attention, Transformer architecture, training, post-training, inference, reasoning, context, hallucination, retrieval, tools, agents, prompting, verification and trust.The mathematics appears where it adds precision, but intuition comes first. The aim is not to replace technical ideas with metaphors. It is to make the real mechanisms understandable.Written for managers, researchers, students, software professionals, executives, writers and technically curious readers who want to understand modern AI without either hype or unnecessary jargon.Understand the machine behind the conversation.This description tracks the actual structure of the book, which moves from prediction and tokenization through attention, training, inference and reasoning, then into failure, systems, prompting and trust. The manuscript explicitly targets readers without a machine-learning background, including managers, researchers, students, software professionals, executives and technically curious readers.

• Reihe: Independently published
• ISBN: 9798175607667
• Sprache: Englisch
• Erscheinungsdatum: 19.09.2026
• Größe: 216 x 140 mm
• Verlag: Independently published