AI Literacy

AI Terms, Explained in Plain English.

New to AI? Here are the words you will hear across our site and the wider industry, written for firm owners and their teams. Over 97 terms, grouped by topic, no math required.

You do not have to memorize any of this. That is our job. This page is here whenever a term stops you.

On this page

Start Here: The Big Picture Generative AI AI Agents Language and Search How AI Learns Model Architecture Vision and Images Measuring AI Running AI Advanced Concepts Safety and Ethics When AI Goes Wrong

Start Here: The Big Picture

Four words explain how the whole field fits together. Each one sits inside the one before it.

Artificial Intelligence (AI): Software built to do things that normally need human thinking, like reasoning, learning, and making decisions.
Machine Learning (ML): Software that gets better by learning from data instead of following fixed rules.
Deep Learning (DL): A kind of machine learning that uses many layers of a neural network.
Neural Network: A web of connected nodes that passes information along to find patterns.

Generative AI

The part of AI most people actually touch: tools that create things from a prompt.

Generative AI: AI that creates new text, images, audio, or code.
Prompt: The instruction you give the model.
Token: The smallest chunk of text a model reads or writes.
Context Window: How much text a model can keep in mind at once.
Inference: The moment a model produces an answer from your prompt.
Temperature: A setting that controls how predictable or creative the output is.
Diffusion Model: A model that makes images by slowly cleaning up random noise.
Multimodal Generation: Producing more than one type of content, like text and images, at once.

AI Agents

AI that does not just answer, but takes action toward a goal.

AI Agent: Software that senses a situation, reasons, and takes steps toward a goal.
Agentic AI: AI that plans several steps and carries them out on its own.
Tool Use: An agent calling outside software or services to get something done.
Chain-of-Thought: A model working through a problem step by step.
RAG (Retrieval-Augmented Generation): Pulling in real, current information before the model answers.
MCP (Model Context Protocol): A shared standard that connects AI to outside tools and data.
Multi-Agent System: Several agents working together on one job.
Orchestrator: The coordinator that hands out tasks across agents.
Subagent: A focused agent that handles one part of a larger task.
Human-in-the-Loop: A person who reviews or approves what the AI does.
Autonomous Workflow: A chain of tasks the AI finishes without step-by-step input.

Language and Search

How machines read, understand, and find meaning in words.

NLP (Natural Language Processing): Getting machines to understand human language.
Embedding: A list of numbers that captures the meaning of text.
Tokenization: Breaking text into small pieces a model can process.
Attention: How a model decides which words matter most.
Semantic Search: Finding results by meaning, not just matching keywords.
Named Entity Recognition: Spotting names, places, and dates inside text.

How AI Learns

The main ways a model is trained.

Supervised Learning: Training on examples that already have the right answers labeled.
Unsupervised Learning: Finding patterns in data that has no labels.
Reinforcement Learning: Learning by trial and error, guided by rewards.
Self-Supervised Learning: A model that creates its own labels from raw data.
Transfer Learning: Reusing what a model learned on one task for another.
Fine-Tuning: Adjusting a pre-trained model for a specific job.
Few-Shot Learning: Learning a task from just a handful of examples.
Zero-Shot Learning: Doing a task with no examples at all.

Model Architecture

The building blocks under the hood.

Transformer: The design behind most modern large language models.
LLM (Large Language Model): A large model trained on huge amounts of text.
Foundation Model: One general model that can be adapted to many tasks.
Parameters: The learned settings that shape how a model behaves.
Training Data: The dataset a model learns from.
Batch Size: How many examples a model processes at one time.
Attention Head: One of many parts of a model that each focus on different details.
Encoder-Decoder: A design that reads the input, then writes the output.
Mixture of Experts: A model that routes each task to specialized sub-models.
Small Language Model (SLM): A lighter, faster model built for a narrower job.
Quantization: Shrinking a model size while keeping most of its accuracy.

Vision and Images

AI that works with pictures and video.

Computer Vision: Software that interprets images and video.
Multimodal AI: AI that handles text, images, and audio together.
OCR (Optical Character Recognition): Pulling text out of an image.
Object Detection: Finding and locating objects inside an image.
Image Segmentation: Splitting an image into meaningful regions.

Measuring AI

How teams check whether a model is any good.

Benchmark: A standard test used to compare models.
Accuracy: How often the model is right overall.
Precision: Of the answers it marked positive, how many were actually right.
Recall: Of all the real positives, how many the model caught.
F1 Score: One number that balances precision and recall.
Perplexity: A measure of how well a model predicts text.
Human Evaluation: People, not just metrics, judging the output.

Running AI

The plumbing that makes AI work in the real world.

API: The connection other software uses to talk to a model.
Latency: The time between a request and the response.
Throughput: How much work a system can handle in a given time.
GPU: Specialized hardware that speeds up training and answering.
Inference Endpoint: The address where a live model receives requests.
Vector Database: Storage that finds similar items fast using embeddings.
Edge Deployment: Running a model on a local device instead of the cloud.
Serverless Inference: Running a model without managing your own servers.
Model Serving: Making a trained model available for real-time use.
Caching: Saving past results to cut cost and speed things up.

Advanced Concepts

Ideas you will hear in the headlines.

AGI (Artificial General Intelligence): A hypothetical AI with human-level ability across tasks.
ASI (Artificial Superintelligence): A theoretical AI beyond human ability.
Emergent Abilities: Skills that only show up once a model gets large enough.
Scaling Laws: The pattern where more data and compute steadily improve performance.
In-Context Learning: Learning a task from examples inside the prompt itself.
Sparse Model: A model that uses only part of itself for each task.
Reasoning Model: A model tuned to think through problems, not just answer fast.
Test-Time Compute: Spending extra processing while answering to get a better result.
Synthetic Data: Training data that is generated artificially.
Model Distillation: Training a smaller model to copy a larger one.

Safety and Ethics

The guardrails that matter, especially for regulated firms.

Explainable AI (XAI): Making how a model reached its answer easy to understand.
Red Teaming: Deliberately stress-testing a model to find weaknesses.
Model Card: A short document describing what a model can and cannot do.
Guardrails: Rules that filter unsafe input and output.
Data Provenance: Tracking where training data came from.
Watermarking: Marking content so people can tell it was AI-generated.
Dual-Use Risk: A capability that can help or harm depending on how it is used.
Responsible AI: Building and running AI with accountability in mind.

When AI Goes Wrong

The failure modes worth knowing before you trust an answer.

Hallucination: A confident answer that is actually wrong.
Overfitting: When a model memorizes its training data and cannot handle new cases.
Underfitting: When a model is too simple to learn the pattern.
Bias: Skewed results that come from skewed data.
Alignment: Getting an AI to behave the way people actually intend.
Drift: A model getting worse over time as the real world changes.
Catastrophic Forgetting: Losing old knowledge while learning something new.
Prompt Injection: A sneaky input designed to hijack a model behavior.
Jailbreak: Tricking a model into ignoring its own safety rules.

You do not need to be an expert. We are.

If you want AI working for your firm without the jargon, start with a plain read on where you stand, or see how we make firms findable when people ask an AI engine who to hire.

Take the free AI Readiness Assessment See AI Visibility →

FAQ

AI Basics, Frequently Asked Questions

What is artificial intelligence (AI)?

Artificial intelligence is software built to do things that normally need human thinking, like reasoning, learning, and making decisions. In everyday use today, most AI is machine learning, where the software improves by learning from data instead of following fixed rules.

What is the difference between AI, machine learning, and deep learning?

They nest inside each other. AI is the broad goal of human-like software. Machine learning is the main way we get there, by learning from data. Deep learning is a type of machine learning that uses many-layered neural networks, and it powers most modern AI you hear about.

What is a large language model (LLM)?

A large language model is a large AI model trained on huge amounts of text so it can understand and generate language. ChatGPT, Claude, and Gemini are built on LLMs. They predict likely text, which is why they are powerful but can still be wrong.

What is generative AI?

Generative AI is AI that creates new content, such as text, images, audio, or code, from an instruction called a prompt. It is the kind of AI most people interact with directly.

What is an AI agent?

An AI agent is software that senses a situation, reasons about it, and takes steps toward a goal, often by calling other tools. Agentic AI goes further and plans several steps, then carries them out on its own.

What is RAG (retrieval-augmented generation)?

RAG means pulling in real, current information before the model answers, instead of relying only on what it memorized during training. It makes answers more accurate and up to date, and it reduces made-up answers.

What is an AI hallucination?

A hallucination is when AI gives a confident answer that is actually wrong. It happens because models predict likely text rather than look up facts, which is why a human review still matters for anything important.

What is a prompt?

A prompt is simply the instruction you give an AI model. Clearer, more specific prompts usually produce better results.

What is fine-tuning?

Fine-tuning is adjusting a general pre-trained model so it does a specific job better, using your own examples. It is one way to make a model sound and act like it belongs to your firm.

Why should a business owner understand these AI terms?

You do not need to build AI, but knowing the words helps you make good decisions, ask better questions, and avoid being oversold. It also helps you see where AI can safely help your firm and where a human still needs to stay in the loop. If you want a plain read on your own firm, take the free AI Readiness Assessment at /assessment.