Learn artificial intelligence

Start Here

New to the topic? Use these entry points, then browse branches below.

Fundamentals

What AI is, how it differs from automation, and the core vocabulary.

• AI Inference

  Inference is running a trained model on new inputs to produce predictions or generations—the production phase users experience when they chat, search, or…

  student
• AI Training

  AI training is the process of adjusting model parameters using data and an optimization objective so the model improves at predicting or generating target…

  student
• Artificial General Intelligence

  Artificial general intelligence (AGI) refers to AI systems that can learn and perform most intellectual tasks at a level comparable to humans across…

  student
• Artificial Intelligence

  Artificial intelligence (AI) is the field of computer science that builds systems capable of performing tasks that typically require human cognitive…

  beginner
• Model Parameters

  Model parameters are the learned numeric values—weights and biases—that a neural network adjusts during training so it can map inputs to useful outputs.

  student
• Narrow AI

  Narrow AI (also called weak AI) refers to systems designed for specific tasks—translation, image classification, coding assistance—rather than general…

  beginner
• Symbolic AI

  Symbolic AI represents knowledge with explicit symbols and rules—logic, knowledge graphs, expert-system style reasoning—complementing statistical machine…

  student

Machine Learning

Learning from data: supervised, unsupervised, evaluation, and classic models.

• Classification

  Classification is a supervised learning task that assigns inputs to discrete categories, such as spam vs. not-spam or disease present vs. absent.

  beginner
• Clustering

  Clustering groups unlabeled examples by similarity so you can discover segments, structure, or candidates for further labeling.

  beginner
• Cross-Validation

  Cross-validation estimates how well a model generalizes by repeatedly training and evaluating on different splits of the data.

  student
• Feature Engineering

  Feature engineering transforms raw data into inputs that make learning easier—ratios, encodings, windows, and domain-derived signals that improve model…

  professional
• Gradient Descent

  Gradient descent optimizes model parameters by repeatedly stepping opposite the gradient of the loss, forming the backbone of modern deep learning training.

  student
• Hyperparameters

  Hyperparameters are training and model choices set before learning parameters—learning rate, batch size, depth, and regularization strength among them.

  student
• Loss Functions

  A loss function quantifies how wrong a model’s prediction is so optimizers can update parameters to improve performance.

  student
• Machine Learning

  Machine learning (ML) is a set of methods that improve a model’s performance on a task by learning patterns from data rather than relying solely on…

  beginner
• Overfitting

  Overfitting happens when a model memorizes training data quirks and fails to generalize to new data, producing optimistic offline scores and disappointing…

  student
• Reinforcement Learning

  Reinforcement learning (RL) is a machine learning paradigm where an agent learns to make sequential decisions by interacting with an environment and maximizing…

  student
• Self-Supervised Learning

  Self-supervised learning creates training signals from the data itself—such as predicting masked words or corrupted patches—so models learn representations…

  student
• Supervised Learning

  Supervised learning trains models on input–label pairs so they can predict labels for new inputs—classification and regression are the classic cases.

  beginner
• Unsupervised Learning

  Unsupervised learning finds structure in unlabeled data—clusters, densities, compressed representations—without predefined target labels.

  beginner

Deep Learning

Neural networks, layers, representation learning, and modern training.

• Attention

  Attention mechanisms let neural networks weight which parts of an input sequence are most relevant when computing each output representation or token.

  student
• Backpropagation

  Backpropagation computes gradients of the loss with respect to neural network parameters efficiently, enabling gradient-based training of deep models.

  student
• Computer Vision

  Computer vision is the AI field that enables machines to interpret images and video—detecting objects, estimating pose, reading text in scenes, segmenting…

  student
• Convolutional Neural Networks

  Convolutional neural networks (CNNs) use local filters and weight sharing to learn spatial hierarchies—foundational for modern computer vision.

  student
• Deep Learning

  Deep learning is a machine learning approach that uses multi-layer neural networks to learn hierarchical representations from data. It powers modern speech…

  student
• Embeddings

  Embeddings are vector representations of text, images, or other objects that place similar items nearby in a geometric space for search, clustering, and model…

  student
• Neural Networks

  Neural networks are computational models composed of interconnected units (neurons) that transform inputs through weighted connections and nonlinear…

  student
• Transformers

  Transformers are neural architectures that use attention mechanisms to model relationships among tokens, forming the backbone of modern language, vision, and…

  student

Generative AI

Systems that produce text, images, audio, and other media.

• Foundation Models

  Foundation models are large pretrained models adaptable to many downstream tasks via prompting, fine-tuning, or tooling—central to today’s AI product ecosystem.

  student
• Generative AI

  Generative AI refers to models that create new content—text, images, audio, code, or other media—by learning the patterns of a training distribution. It…

  student
• Multimodal Models

  Multimodal models accept or produce more than one data type—such as text, images, audio, or video—within a unified or coordinated architecture.

  student

LLMs

Large language models, transformers, tokenization, and prompting basics.

• Context Windows

  A context window is the maximum amount of text (tokens) a language model can consider at once—including the system prompt, user messages, retrieved documents…

  student
• Fine-Tuning

  Fine-tuning adapts a pretrained model to a target task, domain, or style by continuing training on curated examples—often with parameter-efficient methods.

  professional
• Large Language Models

  Large language models (LLMs) are neural networks trained on vast text corpora to predict tokens in context, enabling fluent generation, summarization…

  student
• Natural Language Processing

  Natural language processing (NLP) is the AI subfield that enables computers to analyze, generate, and interact through human language. It spans classical…

  student
• Pretraining

  Pretraining is large-scale learning on broad data that builds general model capabilities before any task-specific fine-tuning or product alignment.

  student
• Quantization

  Quantization reduces numeric precision of model weights (and sometimes activations) to cut memory and accelerate inference with controlled quality loss.

  developer
• Reasoning Models

  Reasoning models are language models trained or configured to spend more test-time computation on multi-step problem solving, often with explicit intermediate…

  professional
• RLHF

  Reinforcement Learning from Human Feedback (RLHF) shapes model behavior using preference data and reward models rather than only static supervised targets.

  professional
• Temperature

  Temperature is a decoding parameter that scales next-token probabilities, influencing how deterministic or varied a language model’s outputs are.

  beginner
• Tokens

  Tokens are the discrete units language models read and generate—commonly subword pieces—used for context limits, billing, and model input encoding.

  beginner
• Tool Calling

  Tool calling lets models request structured actions—search, calculators, CRM updates, code execution—through well-defined function interfaces rather than…

  developer

Agents

Tool-using loops, planning, memory, and grounded autonomy.

• AI Agents

  AI agents are systems that use models to pursue goals through multi-step reasoning and actions—often calling tools, browsing information, writing files, or…

  professional
• Human in the Loop

  Human-in-the-loop (HITL) designs keep people involved in training, review, or approval so AI systems stay accountable and correctable.

  professional
• Multi-Agent Systems

  Multi-agent systems coordinate multiple AI (and sometimes human) agents with different roles or tools to complete complex workflows.

  professional

Media

Vision, speech, video, and multimodal pipelines.

• AI Image Generation

  AI image generation creates new pictures from text prompts, reference images, or editing instructions. Modern systems often use diffusion models and are widely…

  beginner
• AI Video Generation

  AI video generation synthesizes short clips from text, images, or existing footage using generative models. Products differ in duration limits, motion…

  student
• Diffusion Models

  Diffusion models are a generative modeling family that learn to reverse a gradual noising process. They underpin many modern image and video generators and are…

  student
• Speech to Text

  Speech-to-text (STT), or automatic speech recognition (ASR), converts spoken audio into written text. It powers captions, meeting notes, voice interfaces, and…

  beginner
• Text to Image

  Text-to-image systems generate pictures from natural-language prompts, typically using diffusion or related generative models trained on image–text pairs.

  beginner
• Text to Speech

  Text-to-speech (TTS) systems convert written text into audible speech. Modern neural TTS can sound natural across languages and styles, powering accessibility…

  beginner
• Voice Cloning

  Voice cloning synthesizes speech that resembles a specific person’s voice, usually from reference audio samples plus text-to-speech models.

  professional

RAG & Search

Embeddings, retrieval-augmented generation, and grounded answers.

• Chunking

  Chunking splits documents into smaller passages before embedding and retrieval so models receive focused, relevant context.

  developer
• Retrieval-Augmented Generation

  Retrieval-augmented generation (RAG) combines a language model with a search or retrieval step so answers can cite or incorporate external documents. It is a…

  professional
• Semantic Search

  Semantic search retrieves information by meaning using embeddings and similarity, rather than relying only on exact keyword matches.

  student
• Vector Databases

  Vector databases store embedding vectors and support approximate nearest neighbor search so applications can retrieve semantically similar items at scale.

  developer

Prompting

Instructions, few-shot patterns, and evaluation of prompt quality.

• Chain-of-Thought Prompting

  Chain-of-thought (CoT) prompting encourages models to work through intermediate steps on multi-step problems, improving some reasoning benchmarks when used…

  student
• Few-Shot Prompting

  Few-shot prompting includes a handful of example input–output pairs in the prompt so the model can imitate the demonstrated pattern.

  beginner
• Prompt Engineering

  Prompt engineering is the practice of designing instructions, examples, and context so generative models produce more reliable results. It ranges from everyday…

  student
• Prompt Injection

  Prompt injection is an attack where untrusted content manipulates an AI system into ignoring developer instructions or misusing tools.

  developer
• System Prompts

  System prompts are higher-priority instructions that set an AI assistant’s role, style, tool policies, and safety constraints for a session or application.

  developer
• Zero-Shot Prompting

  Zero-shot prompting asks a model to perform a task using instructions alone, without providing worked input–output examples in the prompt.

  beginner

Building with AI / Development

APIs, evaluation harnesses, product integration, and ops concerns.

• AI APIs

  AI APIs expose model capabilities—chat, embeddings, speech, images, and more—over programmatic interfaces so developers can integrate intelligence into…

  developer
• Function Calling

  Function calling is the API pattern where models return structured arguments for developer-defined functions, enabling reliable tool use from LLMs.

  developer
• Local AI

  Local AI runs models on devices or private infrastructure you control, reducing reliance on third-party cloud inference for privacy, offline use, or cost…

  developer
• Open-Weight Models

  Open-weight models release learned parameters for download and local or self-hosted inference, with licenses that may still restrict some commercial uses.

  professional

Responsible AI

Safety, bias, privacy, copyright, and trustworthy deployment.

• AI Alignment

  AI alignment research and practice aim to make AI systems pursue intended goals and constraints—matching human/organizational preferences and safety…

  professional
• AI Copyright

  AI copyright issues involve training on protected works, ownership of model outputs, licensing of datasets and weights, and evolving official guidance that…

  professional
• AI Detection

  AI detection tools estimate whether text, images, audio, or video were produced or altered by generative models—estimates that are inherently probabilistic and…

  professional
• AI Governance

  AI governance is the system of policies, roles, controls, and oversight processes organizations use to manage AI risk, compliance, and value delivery.

  professional
• AI Hallucinations

  AI hallucinations are confident but incorrect model outputs—especially fabricated facts, sources, or details—produced by generative systems. Understanding why…

  student
• AI Privacy

  AI privacy covers how personal data is collected, used in training and inference, retained in logs/prompts, and protected against leakage through model outputs…

  professional
• AI Safety

  AI safety covers technical and organizational practices that reduce harmful outcomes from AI systems—ranging from misuse and accidents today to longer-term…

  professional
• Bias in AI

  Bias in AI refers to systematic error or unfair differential performance that harms people or distorts decisions, often reflecting data, design, or deployment…

  professional
• Deepfakes

  Deepfakes are synthetic media—video, audio, or images—that convincingly depict people saying or doing things they did not, enabled by generative AI.

  student
• Explainable AI

  Explainable AI (XAI) aims to make model behavior understandable to humans through interpretable models, post-hoc explanations, or evidence-centered product…

  professional

Practice

Hands-on exercises and applied drills tied to encyclopedia topics.

• AI Coding Assistants

  AI coding assistants help developers write, edit, explain, and refactor software by combining large language models with editor or repository context. They…

  developer

History

Milestones that shaped research and products.

• History of AI

  The history of artificial intelligence spans symbolic reasoning and early expert systems, winters of reduced funding, the statistical and deep learning…

  beginner

Comparisons

Side-by-side concept contrasts without marketing fluff.

• AI vs Machine Learning

  Artificial intelligence is the broader field of creating systems that perform tasks associated with intelligence; machine learning is a major approach that…

  student
• Machine Learning vs Deep Learning

  Deep learning is machine learning with multi-layer neural networks that learn hierarchical features, especially effective for unstructured data like images…

  student
• Open Source vs Open-Weight AI

  Open-source AI typically implies publicly licensed code (and often training recipes) under OSI-style terms, while open-weight AI releases model parameters…

  professional
• RAG vs Fine-Tuning

  RAG connects models to external evidence at query time; fine-tuning changes model weights to learn behaviors or styles from curated datasets.

  student
• RAG vs Long Context

  Long-context models can read large prompts directly; RAG selectively retrieves relevant snippets so the model is not forced to read an entire corpus every time.

  student
• Text to Speech vs Speech to Text

  Text-to-speech (TTS) turns written text into audio; speech-to-text (STT) turns audio into written text. They are inverse interface problems with different…

  student
• Training vs Inference

  Training learns parameters from data; inference applies a trained model to new inputs to produce predictions or generations.

  student

Learning Paths

Ordered lesson sequences — no account required.

• AI for Beginners

  A gentle path from basic AI concepts to practical tools—without assuming a math background.

  Curious beginners, students, and professionals new to AI
• How LLMs Work

  From tokens and transformers to grounding strategies used in real LLM applications.

  Students, developers, and technical PMs
• Generative Media

  Learn image, video, and voice generation systems—and the responsible-use practices they require.

  Creators, marketers, and product teams
• Building with AI

  A builder’s path through APIs, prompting, retrieval, agents, and deployment choices including local models.

  Developers and technical founders

View all learning paths →

Glossary

Quick A–Z definitions linked back into encyclopedia topics.

Open Glossary

Concept Map

See how AI, machine learning, deep learning, LLMs, RAG, and agents relate.

Open Concept Map