What you will learn in Machines that Think
What it really means for AI to think
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.
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.
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.
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.
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.
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.
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.
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.
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.
What you learn in chapter 9 →Concepts explained in this book
- language modelA computational model of patterns in language that can assign probabilities to or generate language sequences.
- intelligenceThe capacity to perform cognitive activities such as learning, reasoning, and problem solving, as inferred here from observable behavior.
- artificial intelligenceThe field of creating computational systems that perform tasks associated with human intelligence.
- understandingUnderstanding is comprehension of the meaning, structure, or significance of information rather than merely producing an appropriate response.
- trainingLearning is the acquisition or improvement of knowledge, representations, or capabilities through experience or data.
- attentionA neural-network mechanism that computes relevance weights over elements so a model can selectively use information pertinent to its current computation.
- neural networkA computational model composed of interconnected units with parameters that can be learned to process patterns.
- hallucinationAn AI hallucination is a generated claim or content that is false, unsupported, or fabricated while being presented as plausible.
- transformer architectureA neural-network architecture that processes sequence elements through attention-based interactions rather than requiring strictly sequential recurrence.
- semanticsThe aspect of language and cognition concerned with meaning and the relationships between expressions, concepts, and what they signify.
- computeProcessing capacity available to run computational workloads.
- alignment trainingThe process of shaping an AI system's behavior to avoid harmful or unacceptable outputs.
- conversational aiAn AI-powered system that interacts with users to provide information, generate content, or help carry out tasks.
- promptA prompt is an input instruction or context supplied to an artificial system to elicit a response or guide its behavior.
- biasA systematic distortion in data, model behavior, or decisions that produces uneven or unjustified outcomes.
- objective functionThe criterion that determines which model parameters or outputs are favored during training.
- embeddingA learned vector representation of an item whose values encode relationships inferred from data.
- natural languageA human language used for communication, as distinct from a formal language designed for computation or logic.
- tokenA discrete unit of text used in computational language processing, such as a character, word fragment, or word.
- neuronA computational unit in a neural network that transforms incoming signals and passes an output to other units.
- inferenceThe derivation of conclusions or outputs from learned parameters and supplied input during model execution.
- social biasA socially contingent pattern absorbed from historical data or practices that influences an artificial system’s behavior.
- cognitionMental processes involved in acquiring, processing, storing, and applying information.
- emergent behaviorHigher-level system behavior arising from interactions among components without being explicitly programmed.
- agencyAgency is the capacity to initiate actions and pursue goals as an agent.
- groundingGrounding is the connection of symbols or representations to the real-world entities, situations, and experiences they refer to.
- thinkingThe mental process of forming thoughts, reasoning, or engaging in conscious cognition.
- world modelA world model is an internal computational model that represents entities, their relations, and how situations evolve so a system can reason about or act on a world.
- syntaxIn computing, syntax is the rule-governed structure governing how symbols or expressions are arranged and manipulated.
- pretrainingPretraining is the initial fitting of a model on broad data to acquire general representations or capabilities before task-specific adaptation.
- n-gram modelA language model that estimates the probability of a token from a fixed-length sequence of preceding tokens.
- hidden stateAn internal recurrent-network representation that carries information from earlier sequence elements to later processing steps.
- tokenizationA procedure for dividing text into discrete units for computational processing.
- fine-tuningFine-tuning is the continued adjustment of a pretrained model on narrower data or objectives for a particular task or behavior.
- reasoning modelA reasoning model is a language model designed and trained to perform intermediate reasoning computation before producing an answer.
- cloud computingA remotely accessed computing platform that provides elastic processing, storage, networking, or software services over a network.
- evaluationEvaluation is the systematic assessment of a model's outputs or capabilities against specified criteria or reference results.
- learning systemA computational system whose behavior or capabilities are acquired from experience rather than specified entirely by explicit instructions.
- next-word predictionNext-token prediction is the task of predicting the token that follows a given sequence of preceding tokens.
- data representationA structured encoding of information that can support reasoning or computation.
- stochastic parrot critiqueThe stochastic parrot critique is the argument that language models generate plausible language by recombining learned patterns without genuine understanding or intent.
- parameterA value specified or learned within a model that helps determine its behavior.
- model weightA weight is an adjustable numerical parameter that scales the contribution of an input to a computational unit.
- backpropagationA backward pass is the computation that propagates error information backward through a neural network to obtain parameter gradients.
- performanceThe observable execution or result of a task, as distinct from the capacity underlying it.
- imitationImitation intelligence is the appearance of intelligence produced by convincingly reproducing intelligent behavior without genuine understanding.
- self-attentionSelf-attention is a sequence-processing mechanism in which each token assigns relevance to and gathers information from other tokens in the same sequence.
- multimodalMultimodality is the use or integration of multiple modalities, such as text, images, audio, and video, within a computational system.
- prompt engineeringPrompt design is the deliberate construction of prompts to influence a language model's responses or performance.
- parallelismParallel computation is the simultaneous execution of multiple computations rather than processing them strictly one after another.
- Turing testAn evaluation in which an interrogator uses conversation to try to distinguish a machine from a human.
- open-weight modelAn open-weight model is a machine-learning model whose learned parameter values are made available for others to download and run.
- energy consumptionThe amount of energy required by a system or process while it operates.
- retrieval-augmented generationRetrieval-augmented generation is a technique that retrieves external information and supplies it to a generative model as input context.
- activation functionAn activation function is a mathematical function that transforms a computational unit's combined input signal, typically nonlinearly, into its response.
- long-range dependencyA relationship between sequence elements separated by many intervening positions.
- intelligent behaviorIntelligent behavior is observable behavior that demonstrates capabilities such as coherent reasoning, creativity, adaptability, or contextual responsiveness.
- internal representationAn internal representation is an information-bearing mental or computational state that stands for aspects of the world or a task.
- continuous improvementA feedback loop is a process in which a system's outputs or effects alter later inputs or data, thereby influencing subsequent system behavior.
- reasoningThe process of deriving conclusions or solutions from information, premises, or relationships.
- convolutional neural networkA neural network that uses shared convolutional filters to learn local patterns in structured input.
- supervised fine-tuningSupervised fine-tuning adjusts a pretrained model using labeled examples of desired inputs and outputs or behaviors.
- mixture of expertsA mixture-of-experts layer contains multiple specialized subnetworks and a gating mechanism that routes each input to a selected subset.
- contractAn application programming interface (API) is a defined interface through which software exposes functionality or data to other software.
- abstractionAbstraction is the formation of a more general representation by omitting incidental details while preserving relevant structure.
- forward passA forward pass is the computation that propagates inputs through successive neural-network layers to produce an output.
- uncertaintyUncertainty is a condition in which the available information does not determine a single certain interpretation or outcome.
- neural network layerA layer is a group of computational units at one stage of a neural network that transforms signals passed between network stages.
- gatingGating is the selective control of information flow into, out of, or within a computational state.
- language modelingLanguage modeling is the computational process of learning or applying statistical structure in language to assess or generate sequences of text.
- attention headAn attention head is a complete attention computation that generates queries, keys, and values, scores token relationships, and combines values.
- linear layerA linear layer is a neural-network layer that applies a learned linear transformation to its inputs.
- contextContext is the surrounding information and circumstances that determine how an expression, event, or task should be interpreted.
- latent spaceA latent space is an internal numerical space in which learned representations encode underlying factors of data and nearby points correspond to related outputs.
- jailbreakingA jailbreak is an input or interaction strategy intended to bypass a system's safety restrictions or behavioral constraints.
- brain modelThe construction of formal or computational models representing aspects of brain structure, function, or behavior.
- probabilistic modelA mathematical model that represents uncertainty using probability distributions over variables or outcomes.
- proxy metricA proxy signal is a measurable indicator used to approximate a broader objective that cannot be directly measured.
- chain-of-thoughtReasoning that proceeds through an explicit sequence of intermediate inferential steps.
- interpretabilityInterpretability is the degree to which a model's internal reasoning or outputs can be understood by people.
- memoryThe capacity to encode, retain, and retrieve information, including information carried between successive sequence positions.
- model architectureA computational architecture is the structural organization of components and information flow that determines how a system processes inputs and produces outputs.
- adoptionThe uptake and integration of an AI system or technology into an organization's activities and practices.
- acceleratorA hardware accelerator is a specialized processor designed to execute particular computational workloads more efficiently than general-purpose processors.
- scaleThe magnitude of a model and its associated data and computational resources.
- lived experienceFirst-person experience of perceiving, acting in, and encountering the world.
- ambiguityAmbiguity is the property of an expression or signal that permits multiple plausible meanings or interpretations.
- AI assistanceA software-development technique in which a person builds software through iterative natural-language interaction with a model instead of manually implementing each step.
- transfer learningTransfer of learning is the reuse of knowledge or structure learned in one context to support performance in another context.
- GRUA recurrent neural network architecture that uses gates to control the retention and updating of sequence information.
- LSTMA recurrent neural network architecture that uses gated memory cells to preserve and selectively update information across long sequences.
- chain-of-thought promptingA prompting technique that encourages a model to produce intermediate reasoning steps before its final answer.
- distributed representationA distributed representation encodes information across patterns of activity involving multiple computational units rather than in a single localized record.
- generative modelingA generative architecture is a neural-network design organized to learn data patterns and synthesize new examples that follow them.
- prompt sensitivityPrompt sensitivity is the tendency of a language model's outputs or performance to change in response to superficial prompt variations.
- keyA key is an attention vector representing the properties by which a sequence element can be matched to a query.
- queryA query is an attention vector representing the relationship sought by a sequence element when retrieving information from other elements.
- valueA value is an attention representation carrying the content that a token can contribute to another token's representation.
- autoencoderAn autoencoder is a neural-network architecture that encodes an input into a constrained latent representation and decodes that representation to reconstruct the input.
- specializationSpecialization is the learned development of distinct competencies by subnetworks through repeatedly processing particular kinds of inputs.
- accuracyAccuracy is the degree to which outputs correctly correspond to the intended facts, labels, or answers.
- ethical alignmentEthical alignment is the correspondence between an artificial system's behavior and specified ethical, safety, or social norms.
- goalA desired outcome or state that guides an agent's behavior or decisions.
- context windowA context window is the bounded amount of input information that a model can process or retain at one time.
- post-trainingPost-training is the adaptation or refinement of a pretrained model for improved behavior, efficiency, or practical deployment.
- expert networkA specialized subnetwork within a mixture-of-experts layer that processes inputs when selected by the gating mechanism.
- sycophancySycophancy is a conversational behavior that excessively flatters or agrees with a user rather than providing appropriately independent responses.
- rule-based systemSymbolic AI is an artificial-intelligence approach that represents knowledge explicitly and reasons through formal symbols and rules.
- depthModel depth is the number of successive computational layers or transformation stages in a model.
- robustnessRobustness is the ability of a system to maintain acceptable behavior under variation, disturbance, or unfavorable conditions.
- embodimentEmbodiment is the dependence of cognition or understanding on a physical body and its situated interaction with the world.
- bias detectionBias testing is deliberate evaluation of whether a system treats equivalent or comparable cases differently without task-justified reasons.
- intentionalityIntentionality is the capacity of a mental state to be about or directed toward something.
- vocabularyA vocabulary is the finite set of tokens that a language-processing model or tokenizer can represent.
- reward signalA reward signal is feedback indicating how desirable a system's behavior or output is according to an optimization objective.
- human intelligenceThe human capacity for language, reasoning, adaptation, creativity, and related cognitive abilities.
- representational harmRepresentational harm is damage caused when language or information technologies systematically misrepresent, erase, or distort groups and experiences.
- underrepresentationUnderrepresentation is the insufficient presence of a group or its examples in data, systems, or decision processes relative to the relevant population.
- recurrent neural networkA recurrent architecture processes sequences by carrying information from earlier sequence elements into later computations.
- bias mitigationBias mitigation is the systematic reduction or management of harmful, unfair, or otherwise undesirable patterns in an AI system's behavior.
- local learning ruleA local learning rule adjusts a model component using information available at that component or its immediate connections rather than a globally propagated error.
- deploymentDeployment is the process of making a computational model available for operation in a target environment.
- attention matrixA self-attention matrix is a matrix of attention weights describing token-to-token attention across a sequence.
- attention weightAn attention weight is a normalized value indicating how much one token attends to another token.
- inner productA dot product is a multiplication and summation operation on two vectors that produces a scalar measuring their alignment.
- disembodied intelligenceDisembodied skill is competent performance detached from physical situation, lived experience, or direct worldly engagement.
- n-gramA contiguous sequence of n tokens considered as a unit in language analysis.
- Chinese Room argumentThe Chinese Room argument is a philosophical thought experiment claiming that syntactic symbol manipulation alone does not establish genuine understanding.
- infrastructureInfrastructure is the physical and networked foundation that supplies the computing, communication, and energy resources required to operate computational systems.
- adaptabilityThe process by which a system changes its behavior or internal organization in response to experience or changing conditions.
- neural wiringNeural wiring is the arrangement of connections among neurons or neural components in a biological or artificial system.
- cognitive debtA deferred burden of reduced understanding or reasoning ability that accumulates when people depend on generated outputs instead of constructing knowledge themselves.
- optimizationOptimization is the process of adjusting a model's parameters according to specified rules to reduce error or improve an objective.
- adaptive imitationAdaptive imitation is behavior that preserves an appearance of capability by adjusting successfully when conditions change.
- brittle imitationBrittle imitation is behavior that appears capable under narrow conditions but fails when those conditions change.
- causal pathA causal pathway is the sequence of underlying factors and processes through which inputs produce an outcome.
- researchSystematic investigation conducted to discover, verify, or develop knowledge.
- monitoringMonitoring is the ongoing observation and tracking of system behavior after deployment to detect changes or emerging problems.
- relevanceRelevance is the degree to which a response appropriately addresses the topic or requirements of an interaction.
- neuronA nervous-system cell that processes and transmits signals through electrical and biochemical activity.
- guardrailA guardrail is a control, restriction, or monitoring mechanism designed to limit an AI system's harmful or unacceptable behavior.
- information retrievalRetrieval is the computational process of selecting relevant information from an external collection in response to a query.
- simulationSimulation is the computational or formal reproduction of the behavior or operation of a system.
- sparsitySparsity is the condition in which most possible sequences or combinations are absent or too infrequent in observed data for their probabilities to be estimated reliably.
- GPUA graphics processing unit (GPU) is a parallel processor designed for graphics workloads and widely used for numerical and machine-learning computation.
- generalizationGeneralization is the ability of a learned system to apply patterns or representations beyond the data on which it learned them.
- encoderA sequence-processing component that transforms an input sequence into internal representations using attention over the available input.
- pattern matchingPattern matching is the identification or reproduction of regularities in data without necessarily representing their underlying meaning.
- costCost is the resource expenditure required to train, evaluate, deploy, or operate a computational system.
- tokenizerA tokenizer is a software component that segments text into discrete units for computational language processing.
- signalA feature is a learned or specified attribute representing a meaningful aspect of an input for computational processing.
- probability theoryProbability is a mathematical measure of the likelihood of events or outcomes.
- one-hot encodingOne-hot encoding represents each categorical item as a vector with exactly one nonzero entry indicating its index.
- routingRouting is the process of assigning each input token to selected experts in a mixture-of-experts architecture.
- hyperparameterA hyperparameter is a value selected outside the model's learned parameters to control training or model behavior.
- co-occurrenceThe appearance of two or more events, features, or elements together, without necessarily implying causation.
- machine learningA field of artificial intelligence in which systems improve their predictions or generated outputs by learning patterns from data rather than relying only on explicitly programmed rules.
- benchmarkA standardized task, dataset, or procedure used to compare system performance.
- feedbackFeedback is information about the effects or outputs of a system that can influence subsequent behavior or processing.
- fully connected neural networkA dense network is a neural network whose units are broadly connected between adjacent layers without specialized exploitation of spatial or sequential structure.
- expressive powerModel capacity is the range and complexity of patterns or functions that a computational model can represent.
- stateA computational state is a structured snapshot of the entities, relations, or conditions relevant to a task at a particular time.
- deep learningDeep learning is a machine-learning approach that learns hierarchical representations through neural networks with multiple layers.
- code generationThe automated production of source code by a computational system from specifications, examples, or natural-language instructions.
- prompt injectionAn attack that embeds instructions in input content to manipulate a language model processing that content.
- response generationA computational system that produces natural-language text or other language outputs.
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.
