THE AI FIELD GUIDEVOL. 01 — 2026

Understand AI.
One concept at a time.

From your first “what is AI?” to building intelligent systems. Clear explanations, real examples, and one familiar world to connect it all.

200 concepts16 chapters1 shared analogy
Start with the fundamentals
THE BIG PICTURE
AIArtificial intelligence
MLMachine learning
DLDeep learning
Related ideas. Different scopes.
ONE ANALOGY. A WORLD OF UNDERSTANDING.

Welcome to the Sky Research Library.

Imagine a library that learns to help its visitors. Data is its collection. A model is its trained librarian. A prompt is your request. We’ll return to this same world as the ideas get more advanced.

Meet the rest of the library
TrainingPractice with feedback
InferenceAnswering a live request
EmbeddingsA map of related ideas
Context windowThe space on the desk
RAGOpen the right references
ToolsSpecialist service desks
MCPA shared connection protocol
EvaluationCheck the work

Analogies build intuition; the technical definitions explain where the comparison stops. Models do not need human awareness or intentions to perform these operations.

A path from foundations to frontiers

Read in order, or jump to what you’re curious about.

01
START HEREBeginner

The big picture

Understand what AI is, where machine learning fits, and what a model actually does.

13 concepts

AT THE SKY RESEARCH LIBRARY

Think of the whole Sky Research Library: the people, rules, catalog, and tools working together to help visitors. AI is the entire idea of an intelligent service.

Where it’s used

Search, route planning, assistants, image recognition, and decision support.

A concrete example

A library assistant identifies a question about astronomy, finds relevant books, and drafts a response. Several different AI techniques may support those steps.

How it works

  1. 1Define a task and what success means.
  2. 2Choose rules, learned models, search, or a combination.
  3. 3Evaluate results in the setting where the system will be used.
Keep in mind

Intelligent-looking output is not evidence of consciousness, understanding like a human, or factual correctness.

UP NEXTData & mathematical intuition
02
BUILD THE LANGUAGEBeginner

Data & mathematical intuition

The raw materials, representations, and mathematical ideas behind learning.

17 concepts

UP NEXTHow machines learn
03
FROM EXAMPLES TO PATTERNSBeginner

How machines learn

Learning paradigms, the training loop, and the balance between memorizing and generalizing.

15 concepts

UP NEXTClassical machine learning
04
STRONG BASELINESIntermediate

Classical machine learning

Practical models for tables, groups, patterns, and forecasts.

14 concepts

UP NEXTEvaluate before you trust
05
MEASURE WHAT MATTERSIntermediate

Evaluate before you trust

Understand errors, choose meaningful metrics, and test on truly unseen data.

13 concepts

UP NEXTInside neural networks
06
LAYERS OF LEARNINGIntermediate

Inside neural networks

From one artificial neuron to networks that recognize, remember, and generate.

15 concepts

UP NEXTLanguage & transformers
07
THE ENGINE BEHIND LLMSIntermediate

Language & transformers

How text becomes numbers, how attention works, and how language models generate.

14 concepts

UP NEXTPrompting & reasoning
08
GIVE THE MODEL A GOOD BRIEFIntermediate

Prompting & reasoning

Turn requests into clear instructions, useful context, and verifiable outputs.

9 concepts

UP NEXTRAG & knowledge retrieval
09
GIVE AI A REFERENCE LIBRARYAdvanced

RAG & knowledge retrieval

Connect generation to your documents, and learn what makes retrieval reliable.

11 concepts

UP NEXTAdapt & align models
10
CHANGE THE LEARNINGAdvanced

Adapt & align models

When to train, when to retrieve, and how to adapt models efficiently.

11 concepts

UP NEXTAgents & tool use
11
FROM ANSWERS TO ACTIONSAdvanced

Agents & tool use

Build systems that can work through a task while keeping control of their actions.

10 concepts

UP NEXTMCP, plugins & skills
12
CONNECT THE PIECESAdvanced

MCP, plugins & skills

Understand the protocol, build a small server, and package capabilities clearly.

11 concepts

UP NEXTPerception & reinforcement learning
13
BEYOND TEXTAdvanced

Perception & reinforcement learning

See, hear, and learn through action and feedback.

13 concepts

UP NEXTFrom prototype to production
14
MAKE IT RELIABLEAdvanced

From prototype to production

Deploy, monitor, and operate an AI system as part of a real product.

12 concepts

UP NEXTSafety, trust & responsible AI
15
KNOW THE LIMITSAdvanced

Safety, trust & responsible AI

Look beyond average accuracy to security, privacy, fairness, and accountability.

11 concepts

UP NEXTAdvanced architectures & frontiers
16
KEEP EXPLORINGAdvanced

Advanced architectures & frontiers

Efficiency, scale, and emerging approaches—without treating research as settled fact.

11 concepts

KEEP YOUR CURIOSITY GOING

Built on good references.

Original explanations, examples, and library analogies, informed by primary research, official documentation, and public course outlines. Start with these learning paths, or follow the source links inside each concept.

Browse all 67 references
CourseGoogle Machine Learning Crash CourseDocumentationGoogle: working with numerical dataDocumentationGoogle: embeddingsDocumentationGoogle: production ML systemsDocumentationGoogle: fairness in machine learningCourseHugging Face LLM CourseDocumentationscikit-learn: supervised learningDocumentationscikit-learn: unsupervised learningDocumentationscikit-learn: preprocessing dataDocumentationscikit-learn: cross-validationDocumentationscikit-learn: model evaluationDocumentationscikit-learn: probability calibrationDocumentationscikit-learn: ensemble methodsTutorialPyTorch: build a neural networkTutorialPyTorch: optimizing model parametersTutorialPyTorch: object detection and segmentationDocumentationPyTorch: distributed trainingPaperAttention Is All You NeedPaperExploring the Limits of Transfer LearningPaperRetrieval-Augmented GenerationPaperLoRA: Low-Rank AdaptationPaperQLoRA: Efficient Finetuning of Quantized LLMsPaperDirect Preference OptimizationPaperTraining language models with human feedbackPaperConstitutional AIPaperDenoising Diffusion Probabilistic ModelsPaperAuto-Encoding Variational BayesPaperGenerative Adversarial NetworksPaperChain-of-Thought PromptingDocumentationAnthropic: prompt engineering overviewEngineeringAnthropic: building effective agentsPaperReAct: reasoning and actingDocumentationLangGraph overviewDocumentationLangChain: retrievalDocumentationSentence Transformers: semantic searchDocumentationFaiss documentationDocumentationMicrosoft GraphRAGDocumentationHugging Face PEFTDocumentationHugging Face: quantization overviewSpecificationMCP: introductionSpecificationMCP: architectureTutorialMCP: build a serverSpecificationMCP: server conceptsSpecificationMCP: security best practicesSpecificationAgent2Agent protocolPaperLearning Transferable Visual Models · CLIPPaperRobust Speech RecognitionCourseSpinning Up: key concepts in RLPaperProximal Policy OptimizationPaperDeepSeekMathDocumentationMLflow documentationDocumentationvLLM documentationDocumentationvLLM: PagedAttentionDocumentationOllama quickstartFrameworkNIST AI Risk Management FrameworkSecurityOWASP risks for LLM applicationsPaperA Unified Approach to Interpreting Model PredictionsPaperDeep Learning with Differential PrivacyPaperModel Cards for Model ReportingPaperSparsely-Gated Mixture-of-ExpertsPaperFlashAttentionPaperSpeculative SamplingPaperMamba: selective state spacesPaperTraining Compute-Optimal Language ModelsResearchWorld ModelsDocumentationDoWhy: estimating causal effectsCourse outlineEd Donner: AI Engineer Core Track

Content reviewed September 20, 2026. AI is a growing field; this guide covers core concepts and selected advanced techniques, rather than claiming a complete vocabulary. APIs and implementation details evolve—check the linked documentation for your installed version.

Independent educational guide. Not affiliated with the referenced course creators or organizations.