liveBook · Manning

What you will learn in Machines that Think

What it really means for AI to think

by Rubén Castillo Sánchez · Manning Early Access Program (MEAP)

Chapter by chapter

Chapter 1: Old questions, new machines

This introduction frames large language models as a practical and philosophical challenge: their fluent performance can look like understanding, so readers need a framework for judging their capabilities and limits.

1.1 The question of intelligence returns · 1.1.1 Intelligence at the crossroads · 1.1.2 When machines began to speak · 1.1.3 The illusion of understanding · 1.1.4 Why this question matters · 1.2 Is it really about the machines? · 1.2.1 A question older than bytes · 1.2.2 Believing before building · 1.2.3 From dreams to designs · 1.3 The first circuits of thought · 1.3.1 The recipe for modern machines · 1.3.2 The architecture of generality · 1.3.3 From computation to cognition · 1.3.4 The principle of imitation · 1.3.5 Learning, surprise, and unpredictability · 1.3.6 Turing’s test and its evolving meaning · 1.4 Learning to adapt · 1.4.1 When rules failed · 1.4.2 Living with uncertainty · 1.4.3 The spark of scale · 1.4.4 The price of complexity · 1.4.5 Progress in cycles · 1.5 The measure of thought · 1.5.1 What we mistake for intelligence · 1.5.2 Mind in the world · 1.5.3 The return of experience · 1.5.4 Inside the imitation · 1.5.5 New walls for the room · 1.5.6 Syntax, semantics, and the limits of imitation · 1.5.7 Complex boundaries · 1.5.8 Builders and philosophers · 1.6 The shifting meaning of intelligence · 1.6.1 Create to redefine · 1.6.2 The ever-receding horizon · 1.7 Summary

What you learn in chapter 1 →

Chapter 2: Representing language

This chapter introduction frames language as a difficult computational problem and previews the progression from rules and statistical units to numerical representations of meaning.

2.1 Language equals intelligence? · 2.1.1 When words stand in for minds · 2.1.2 What machine language does reveal · 2.1.3 Thought beyond language · 2.1.4 Why language still matters · 2.2 Machines’ first words · 2.2.1 When language was logic · 2.2.2 Patterns without understanding · 2.2.3 From rules to evidence · 2.3 Counting words · 2.3.1 From intuition to probability · 2.3.2 What is an n-gram? · 2.3.3 Building language from statistics · 2.3.4 The problem of sparsity · 2.3.5 Balancing context and complexity · 2.4 A token full of meaning · 2.4.1 From raw text to tokens · 2.4.2 Words, subwords, and characters · 2.4.3 Compressing language · 2.4.4 From BPE to modern tokenizers · 2.4.5 Handling the unknown · 2.4.6 Tokens as information boundaries · 2.4.7 Why tokens matter · 2.5 Embedding reality · 2.5.1 From indexes to vectors · 2.5.2 Learning embeddings · 2.5.3 Embeddings go global · 2.5.4 The geometry of meaning · 2.5.5 Meaning in context · 2.5.6 Why embeddings matter · 2.6 The last frontier · 2.6.1 One reality, infinite expressions · 2.6.2 Meaning stretched across distance · 2.6.3 What language leaves unsaid · 2.6.4 The gaps between the words · 2.6.5 Diversity across languages · 2.6.6 Measuring success · 2.7 Summary

What you learn in chapter 2 →

Chapter 3: Artificial neural networks

This chapter introduction frames neural networks as multi-stage systems that build structured representations from complex data and can be architected for different kinds of patterns.

3.1 What is a neural network? · 3.1.1 Neurons as a web of mathematics · 3.1.2 Layers that shape information · 3.1.3 Weighing what is important · 3.1.4 Shaping the neuron’s response · 3.1.5 Tracing the journey · 3.1.6 Becoming the dominant paradigm · 3.2 How do machines learn? · 3.2.1 Learning from mistakes · 3.2.2 Tracing the path of error · 3.2.3 Nudging in the right direction · 3.2.4 How rules emerge from examples · 3.2.5 What the network learns · 3.3 Why learning is not memorizing · 3.3.1 When training becomes memorization · 3.3.2 Learning under constraint · 3.3.3 Knowing when to stop · 3.3.4 Augmenting the data · 3.3.5 Intelligence in abstraction · 3.4 One concept, many architectures · 3.4.1 The limits of dense networks · 3.4.2 Convolutional networks: structure in space · 3.4.3 Recurrent networks: structure in sequences · 3.4.4 LSTMs and GRUs: unlocking long-term memory · 3.4.5 Autoencoders: compressing the world · 3.4.6 Generative architectures: learning to create · 3.4.7 Architecture as a form of intelligence · 3.5 Specific approaches to language · 3.5.1 Why language challenges neural networks · 3.5.2 Lessons from recurrence · 3.5.3 Pushing the memory barrier · 3.5.4 A word is worth a thousand pictures · 3.5.5 Learning where to look · 3.5.6 Learn first, specialize later · 3.5.7 From models to true language · 3.6 Summary

What you learn in chapter 3 →

Chapter 4: Large language models

This introduction frames large language models as scaled statistical systems whose architecture, capacity, multimodal designs, and costs shape both their apparent intelligence and who can build them.

4.1 The transformers architecture · 4.1.1 Attention was all we needed · 4.1.2 Providing a sense of order · 4.1.3 Listening to the self · 4.1.4 Multiple heads are better than one · 4.1.5 Making depth trainable · 4.1.6 One block, many behaviors · 4.1.7 Connecting representations · 4.1.8 From blocks to outputs · 4.2 What makes a language model “large” · 4.2.1 Where the learning fits · 4.2.2 Large models through broad exposure · 4.2.3 Scale comes with demands · 4.3 From large to usable · 4.3.1 Learning before specialization · 4.3.2 Adapting from within · 4.3.3 Transferring scale · 4.3.4 Compressing computation · 4.4 Somebody call the expert · 4.4.1 Experts needed · 4.4.2 Learning to specialize · 4.4.3 Global scale, selective growth · 4.5 Beyond text: multimodality · 4.5.1 Worlds beyond words · 4.5.2 A shared space · 4.5.3 Architectures for multimodal fusion · 4.5.4 Grounding without causation · 4.6 The cost of intelligence · 4.6.1 Training at scale · 4.6.2 The daily cost of intelligence · 4.6.3 Powering intelligence · 4.6.4 Cost as a research constraint · 4.7 When scale becomes a gatekeeper · 4.7.1 The narrowing field · 4.7.2 Infrastructure as power · 4.7.3 Centralization and dependency · 4.7.4 Who gets to know · 4.7.5 An uneven field · 4.8 Summary

What you learn in chapter 4 →

Chapter 5: Back to humans

This chapter frames human cognition as an essential but limited reference for evaluating AI, setting up an examination of neural inspiration, brain modelling, and the differences between artificial networks and human thought.

5.1 Why it comes back to humans · 5.1.1 Using ourselves as the reference · 5.1.2 The language of minds · 5.1.3 Why the metaphor endured · 5.2 A source of inspiration · 5.2.1 From observation to abstraction · 5.2.2 Designing for trainability · 5.2.3 Learning as weighted influence · 5.2.4 Keeping representation diffuse · 5.3 Modelling the human brain · 5.3.1 Why model the brain at all? · 5.3.2 Levels of biological fidelity · 5.3.3 Focusing on the wiring · 5.3.4 Efficiency in human hardware · 5.3.5 What brain modelling offers · 5.4 Differences with artificial neural networks · 5.4.1 Learning without consequence · 5.4.2 Intelligence without agency · 5.4.3 Memory as structure, not state · 5.4.4 Uniform time and frozen memory · 5.4.5 Representation without understanding · 5.4.6 Why surface similarity misleads · 5.5 Summary

What you learn in chapter 5 →

Chapter 6: The reliability problem

This introduction frames LLM reliability as a problem of fluent probabilistic text generation that can create misplaced trust, and previews hallucinations, safeguards, jailbreaks, and context limits.

6.1 Sorry, are you hallucinating? · 6.1.1 When fluency outruns grounding · 6.1.2 Optimization without truth · 6.1.3 Extrapolating beyond experience · 6.1.4 When fabrication becomes operational risk · 6.2 Only facts, please · 6.2.1 Constraining generation · 6.2.2 Monitoring confidence · 6.2.3 Reliability without guarantees · 6.3 Breaking the rules · 6.3.1 Rules without rule-keeping · 6.3.2 The impact of small details · 6.3.3 The difficulty of improving prompts · 6.3.4 Tilting the balance · 6.3.5 Containing instability · 6.4 A budget for memories · 6.4.1 Where the answer comes from · 6.4.2 A window, not a memory · 6.4.3 Present but not decisive · 6.4.4 Borrowed memory · 6.4.5 Retrieval is not grounding · 6.5 Summary

What you learn in chapter 6 →

Chapter 7: Alignment and bias

This chapter introduction frames alignment and bias as learned, context-sensitive tendencies in language models rather than genuine moral understanding, and previews how their failures and harms can be examined and reduced.

7.1 Training the moral compass · 7.1.1 Building on top · 7.1.2 Engineering ethical behavior · 7.1.3 Alignment without commitment · 7.2 Deep learning, loose ethics · 7.2.1 When observation misleads · 7.2.2 Alignment as strategy · 7.2.3 Instrumental misconduct · 7.2.4 From statistics to consequences · 7.3 Inherited biases · 7.3.1 A chip off the old bad block · 7.3.2 The world as written · 7.3.3 The weight of repetition · 7.3.4 Blind spots and silence · 7.3.5 Outputs and outcomes · 7.4 Towards fairer models · 7.4.1 Inside the black box · 7.4.2 Why did you say that? · 7.4.3 Exposing the bias · 7.4.4 Mitigation and its limits · 7.5 Summary

What you learn in chapter 7 →

Chapter 8: The impact of AI

The chapter introduces how language models are spreading despite their limitations and surveys their effects on work, education, science, and human interaction.

8.1 Accelerated adoption · 8.1.1 A frictionless interface · 8.1.2 The ecosystem · 8.1.3 The ripple effect · 8.1.4 Go as you pay · 8.1.5 From tool to habit · 8.2 Transforming how we work · 8.2.1 Work the model · 8.2.2 The great transformation · 8.2.3 Beyond software · 8.2.4 The exception proves the rule · 8.2.5 Adapting to the new interface · 8.3 Rethinking education · 8.3.1 How did you know? · 8.3.2 The cost of assistance · 8.3.3 Learning as interaction · 8.4 Science at scale · 8.4.1 The faster scientists · 8.4.2 Computers do science · 8.4.3 When trust must catch up · 8.4.4 Better machines through machines · 8.5 Reframing human relations · 8.5.1 Doing, asking, expressing · 8.5.2 Winning Turing’s game · 8.5.3 The risk of sounding right · 8.5.4 The illusion of being heard · 8.6 Summary

What you learn in chapter 8 →

Chapter 9: When machines seem to think

The chapter introduces a balanced way to judge whether current AI can count as intelligent without treating either its failures or its impressive performance as decisive.

9.1 When it walks like a duck · 9.1.1 A convincing performance · 9.1.2 A tricky definition · 9.1.3 Intelligence in others · 9.1.4 What performance is worth · 9.2 Thought behind chains · 9.2.1 Thinking token after token · 9.2.2 A new member of the family · 9.2.3 Lessons from reasoning · 9.3 More than the sum · 9.3.1 When ability appears · 9.3.2 The recipe for emergence · 9.3.3 From emergence to intelligence · 9.4 The illusion hypothesis · 9.4.1 The illusion of thinking · 9.4.2 It’s just a parrot · 9.4.3 A ghost without a shell · 9.4.4 Different kinds of learning · 9.5 The case for machine intelligence · 9.5.1 Intelligence without perfection · 9.5.2 Parroting into structure · 9.5.3 Think outside the body · 9.5.4 Own will hunting · 9.5.5 Different, not worse · 9.5.6 Thinking, but not like us · 9.6 Summary

What you learn in chapter 9 →

Concepts explained in this book

These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.