liveBookmanning.com

What you will learn in The AI Pocket Book

by Emmanuel Maggiori

Read it on liveBook Get the book at manning.com

Chapter by chapter

Concepts explained in this book

• hallucinationA model output that is confidently or plausibly produced despite being unsupported or wrong relative to the intended real-world relationship.
• language modelA neural language model trained on large text corpora to predict and generate natural language.
• LLM wrapperSoftware that sits around an LLM and modifies inputs and outputs.
• embeddingA vector representation used to encode the meaning of an image, text, or other content.
• self-supervised learningA training approach where a model learns from automatically generated targets derived from the input data itself.
• reinforcement learning with human feedbackA training method that uses human judgments as feedback to guide reinforcement learning and improve model outputs.
• supervised learningA machine-learning approach where the model learns from labeled examples showing the desired task behavior.
• machine learningA field that builds systems that learn patterns from data to perform tasks.
• overfittingA failure mode in which a model becomes too specialized to its training data and generalizes poorly to new cases.
• transformer architectureA neural network architecture that contextualizes each token using surrounding tokens rather than compressing the whole prompt into one summary vector.
• next-token predictionA language-model training objective that predicts the next token in a sequence from preceding context.
• clusteringAn unsupervised learning method that groups similar data points into clusters.
• machine learning paradigmA broad way of framing a machine learning task, such as supervised, self-supervised, reinforcement, or unsupervised learning.
• reinforcement learningA learning approach in which an agent improves by trying actions and using feedback from the results.
• unsupervised learningA machine learning approach that seeks patterns in unlabeled data without a target output.
• fine-tuningAdditional training applied to an existing model to adapt or improve its behavior using new examples.
• tokenA discrete text unit fed to a language model as input, which may correspond to a subword or word rather than an individual letter.
• prompt engineeringThe practice of designing prompts to steer a language model toward better or more reliable outputs.
• edge caseA rare or unusual situation that falls outside common patterns seen during training or testing.
• projectionA linear transformation that maps vectors into another space, used here as part of transformer attention computations.
• image recognitionThe task of assigning an image to one of several possible classes based on its visual content.
• foundational modelA preexisting model used as-is or with minimal adaptation instead of building one from scratch.
• convolutional neural networkA neural network architecture that applies convolutions and downsampling to learn spatial features from grid-like data such as images.
• trainingThe process of fitting a model using data and computation.
• memorizationA behavior where a model stores frequent patterns or solutions from training data rather than learning a general rule.
• parameterA learned model value optimized during training rather than fixed by hand.
• world modelAn internal representation an AI system builds about how the world works, including object properties and common situations, used to generalize beyond training examples.
• test-driven developmentA development practice where tests are written for small units of behavior before or alongside code implementation.
• trunk-based developmentA development workflow where changes are integrated directly into the main branch with supporting practices to avoid breaking production.
• vocabularyThe set of tokens a language model can recognize and use for tokenization and generation.
• retrieval-augmented generationA method of augmenting an LLM prompt with retrieved external documents so the model can answer using information not in its training data.
• open source AIA model released with accessible code or weights that can be inspected, modified, or deployed by users.
• excellence gapThe difference between work that is merely acceptable and work that is truly excellent in a given job, used here to explain AI exposure.
• misalignmentA mismatch between the task a model was trained to optimize and the task it is later used for.
• data scienceThe use of data analysis and modeling methods to extract insights or support decisions for business goals.
• operations researchA discipline that applies mathematical methods to make better decisions under constraints, often for business optimization.
• feature flagA configuration mechanism that controls which users see a given version of a feature.
• robustnessRobustness is the ability of a system or model to maintain acceptable behavior or performance despite variations, disturbances, faults, or challenging conditions.
• exploitationIn reinforcement learning, the practice of choosing actions believed to maximize current reward based on known information.
• data explorationIn reinforcement learning, the practice of trying random actions to gather information.
• generalizationThe ability of a model to perform well on new inputs not seen during training.
• SMOTEA data augmentation technique that generates additional minority-class examples by combining existing ones.
• synthetic dataArtificially generated training data used as a substitute for or augmentation to real data.
• underfittingA failure mode where a model is too simple to capture the training pattern and performs poorly.
• AI-first approachA product-development strategy that starts by seeking uses for AI before verifying a customer need.
• antipatternA commonly repeated but harmful approach that looks plausible yet tends to produce bad outcomes.
• customer-centric approachA product-development strategy that begins with customer needs and works backward to the best solution.
• evaluation datasetA held-out dataset used to tune or select among models during development, separate from the final evaluation data.
• competitive advantageA durable feature or condition that helps a company keep customers or market share away from rivals.
• recallA classification or retrieval metric measuring how many relevant or positive items are successfully found.
• entrepreneurship mindsetAn approach to selecting AI use cases by starting from the customer problem and choosing the most appropriate solution rather than assuming AI is needed.
• low-rank adaptationA fine-tuning method that adds small trainable layers or low-rank updates while leaving the original model largely intact.
• minimal productA minimal product version intended to learn about customers with the least necessary implementation.
• text-to-image modelA generative AI technique that produces images conditioned on textual descriptions.
• multiagent AIAn AI setup in which multiple model instances cooperate, critique, or verify one another’s outputs.
• product validationThe process of checking with prospective customers that a product idea actually addresses a real need.
• attention scoreA numerical value used by an attention mechanism to weight the relevance of one element relative to others.
• debuggingThe process of diagnosing and fixing software errors or incorrect behavior.
• mass unemploymentA large-scale rise in joblessness across an economy, discussed here as the feared macroeconomic effect of automation.
• carbon intensityThe amount of carbon dioxide emitted per unit of electricity consumed, used here to compare environmental impact.
• energy consumptionThe amount of electrical energy used by AI systems during training or inference.
• career future-proofingThe strategy of adapting one’s skills and roles so they remain valuable as technology and markets change.
• technology-business intersectionA work area where technical implementation and business goals are both central to the role.
• contextualized representationToken representations that encode surrounding context before being used to produce output predictions.
• attentionThe transformer computation that assigns attention scores and uses them to weigh contextual influence among tokens.
• feed-forward stepThe transformer update stage that further transforms contextualized embeddings after attention.
• multi-head attentionA transformer attention mechanism that splits representations into multiple heads to attend to different relationships in parallel.
• good bang for your promptA book-local phrase meaning a prompt yields useful results with relatively little refinement.
• logo designThe process of creating a visual brand mark, including research, originality, and fit with a market context.
• conditional diffusion modelA generative model that conditions denoising on auxiliary information such as a text caption.
• U-NetAn encoder-decoder neural network architecture that reconstructs a full-size output image from a compressed representation, often using intermediate features from the shrinking path.
• downsamplingA spatial reduction step that shrinks an image or feature map to a smaller resolution, usually discarding fine-grained location detail.
• revenue-maximizing priceThe product price that maximizes expected revenue by balancing sale probability against unit price.
• customer problemA genuine problem or requirement that a product can satisfy for its intended users.
• AstralA Python library mentioned as an off-the-shelf way to calculate the sun position from time and location.
• GPT-4An OpenAI large language model used as an example tokenizer and reference for token counts.
• documentationWritten reference material describing a model or system, including development details and supported information as required here.
• cookie-cutter taskA routine, standard task that is common enough that a prebuilt solution is likely to exist.
• sun position calculationThe computation of the sun's apparent position in the sky for a given time and location.

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.