🔮 Free Reference · No Signup Needed

AI Glossary, in Plain English

72+ AI terms across 9 categories, explained without the jargon. From LLMs and prompt engineering to AI agents and automation. Built for business owners, not data scientists.

Free · Searchable · Updated regularly · No signup

Browsing 72 terms

◇

Foundational AI Concepts

10

AI Model

A computer program that has learned to do specific tasks using data. It's the core of any AI tool, like chatbots or decision-making systems.

Machine Learning (ML)

AI that learns from data without being directly told what to do. Instead of fixed rules, ML systems find patterns and make predictions based on examples they've seen.

Neural Network

A computer system designed to work like the human brain. It uses connected layers to process information and learn patterns by repeatedly looking at data.

Deep Learning

A part of machine learning that uses many layers of neural networks to learn very complex patterns. This is what makes advanced AI possible, like recognising images, understanding language, and self-driving cars.

Foundation Model

A very large, all-purpose AI model (like GPT) that can be adapted for many different tasks. These models are a starting point for specialised AI tools.

Generative AI

AI that creates new content, text, images, audio or code, based on patterns it learned during training. ChatGPT, Midjourney and Claude are all examples of generative AI.

Multimodal AI

AI that can work with multiple types of input at once, such as text, images, audio and video, giving it a more human-like understanding of the world.

Supervised Learning

A training method where the AI learns from labelled examples, data that already has the correct answers. Like learning by studying an answer key.

Unsupervised Learning

Training AI on data without any labels or correct answers. The AI discovers its own patterns and groupings, useful for finding hidden insights in large datasets.

Reinforcement Learning

An AI learns by trial and error, receiving rewards for good actions and penalties for bad ones. The same principle used to train AI to play chess or video games.

🔮

Language & Reasoning

9

LLM (Large Language Model)

An AI model trained on huge amounts of text to understand and create human language. Models like GPT-4 can write, answer questions, summarise text, and even generate code.

NLP (Natural Language Processing)

AI's ability to understand and use human language. NLP helps computers read, listen, interpret meaning, and respond in a natural way.

Prompt Engineering

The skill of writing good instructions to get better results from AI. It means creating clear, specific commands to help AI models give accurate and useful answers.

System Prompt

Behind-the-scenes instructions given to an AI before a conversation starts. This shapes the AI's personality, role and rules, invisible to the end user but critical to how it behaves.

CoT (Chain of Thought)

AI shows its step-by-step thinking. This helps models solve hard problems by breaking them down into logical steps, similar to how people solve problems.

RAG (Retrieval-Augmented Generation)

Combines looking up real-time information with AI-generated answers. RAG systems use outside knowledge to give more accurate and current information than what the AI already knows.

Reasoning Model

An AI built to do structured, logical tasks. These models are good at solving problems that need step-by-step thinking, math, or formal logic.

Transformer

The core architecture behind most modern AI language models. Transformers use an "attention mechanism" to understand the relationship between words in a sentence, the technology that made ChatGPT possible.

Context Window

The total amount of text an AI can see and remember during a single conversation. Larger context windows allow longer, more complex conversations without the AI "forgetting" earlier messages.

⚛️

AI in Action

9

AI Agents

Smart programs that can act and make decisions on their own, like scheduling meetings, browsing the web, or managing projects, without needing a person to supervise every step.

Multi-Agent System

Multiple AI agents working together, each with a specific role. One agent might research, another write, another review, like a team of AI specialists working in parallel.

Tool Use

The ability for an AI to use external tools, like web search, calculators or APIs, to complete tasks beyond what it knows from training alone.

Chatbot

An AI that converses with you, commonly used in customer service. They range from simple rule-based systems to advanced AI assistants that understand nuance and context.

AI Wrapper

An easy-to-use interface built on top of an AI model. Wrappers make powerful AI accessible to everyday users without requiring technical expertise.

Inference

The moment an AI model uses what it learned to generate a result. Training is the learning phase; inference is the doing phase.

Vibe Coding

Writing or fixing code using everyday language. You describe what you want in plain English and the AI writes the correct code, no syntax knowledge required.

Human-in-the-Loop

A system where humans review or approve AI decisions at key points. Keeps humans in control while still benefiting from AI speed and scale.

AI Pipeline

A series of connected AI steps that process data in sequence. The output of one stage becomes the input for the next, like an assembly line for AI tasks.

🦾

Technical Terms

8

Parameters

The internal settings an AI model adjusts as it learns, like knobs and dials being fine-tuned. Large AI models can have hundreds of billions of these settings.

Weights

Numbers inside an AI model that determine how important each piece of information is. Constantly adjusted during training to improve accuracy.

Tokenization

How AI breaks text into smaller pieces called "tokens" it can process. Tokens can be whole words, partial words, or single characters.

Token Limit

The maximum amount of text an AI can process at one time, its "short-term memory." Exceed it and the AI loses track of earlier parts of the conversation.

Embedding

How AI converts words, images or data into numerical codes it can understand, capturing meaning mathematically so it can find similarities and relationships.

Compute

The raw processing power AI needs to train and run. Advanced AI requires enormous compute, typically measured in GPU hours or flops.

API (Application Programming Interface)

A connection that lets software talk to other software. AI APIs let developers plug powerful AI capabilities into their own products without building the AI themselves.

GPU (Graphics Processing Unit)

A type of processor originally built for graphics, now essential for training AI. GPUs process many calculations simultaneously, critical for the parallel computing AI requires.

⚙️

Training & Tuning

8

Training

Teaching an AI model using large datasets. The model sees thousands of examples and adjusts its internal parameters until it learns to recognise patterns.

Fine-Tuning

Taking an existing AI model and adjusting it for a specific job using a smaller, targeted dataset. Much faster and cheaper than training from scratch.

Deployment

Making a trained AI model available for real-world use. Once deployed, it starts processing new inputs and generating outputs in production.

Few-Shot Learning

Teaching an AI to perform a new task using only a small number of examples, sometimes just 2 or 3. Highly efficient for specialised use cases.

Zero-Shot Learning

When an AI handles tasks it was never specifically trained for, using general knowledge to reason through novel problems.

LoRA (Low-Rank Adaptation)

An efficient technique for fine-tuning large AI models that dramatically reduces the computing power and data required, while maintaining performance.

RLHF (Reinforcement Learning from Human Feedback)

A training method where humans rate AI outputs and those ratings are used to improve the model. Used extensively to make AI assistants more helpful and safe.

Synthetic Data

Artificially generated data used to train AI models when real data is scarce, expensive, or sensitive. Increasingly used to accelerate AI development.

⚠️

Risks & Guardrails

8

Hallucination

When AI confidently states something false or fabricated. The AI isn't lying, it's pattern-matching in a way that produces plausible-sounding but incorrect information.

Jailbreaking

Techniques used to bypass AI safety restrictions and get the model to produce content it was designed to refuse.

Guardrails

Safety rules and filters built into AI systems to prevent harmful, biased or inappropriate outputs. The first line of defence in responsible AI.

AI Alignment

The challenge of ensuring AI systems act in ways consistent with human values and intentions, especially important as AI becomes more powerful and autonomous.

Red Teaming

Deliberately trying to break or misuse an AI system to find vulnerabilities before they can be exploited in the real world.

Differential Privacy

A mathematical approach to training AI that protects individual user data, ensuring patterns can be learned without exposing personal information.

Automation Bias

The tendency for people to over-rely on AI decisions and accept them without critical evaluation, even when the AI is wrong.

Shadow AI

Employees using unauthorised AI tools at work without IT or management awareness. A growing governance and security risk for organisations.

🪐

Data & Decision Intelligence

7

Ground Truth

Verified, labelled data used as the definitive reference for training or evaluating an AI model. The gold standard against which AI performance is measured.

Vector Database

A specialised database that stores information as mathematical codes, enabling ultra-fast similarity searches. Essential infrastructure for RAG and semantic search.

Retrieval System

A system that fetches relevant information to augment AI responses. Combines the AI's intelligence with access to specific, up-to-date knowledge.

Explainability

The ability to understand and articulate why an AI made a specific decision. Critical for trust, compliance and debugging AI systems.

Ontology

A structured framework that defines concepts and their relationships in a specific domain. Helps AI systems understand context and connections between ideas.

Data Augmentation

Artificially expanding a training dataset by creating modified versions of existing data, for example, rotating or cropping images. Improves model robustness without collecting new data.

Feature Engineering

The process of selecting and transforming raw data into the most useful inputs for a machine learning model. A critical skill that significantly impacts model performance.

🛰️

Operational Efficiency

6

Latency

The delay between sending a request to an AI and receiving its response. Critical for real-time applications where speed directly affects user experience.

Temperature

A dial that controls how random or creative an AI's outputs are. Low temperature = predictable and precise. High temperature = more varied and imaginative.

Top-k / Top-p Sampling

Settings that control how an AI selects its next word. They balance between staying on-topic (predictable) and being creative (varied).

Cost per Token

The pricing unit for AI API usage. Understanding token costs helps you optimise prompts for both quality and budget efficiency.

Throughput

How many AI requests a system can process per second or minute. High throughput is essential for business applications that handle many simultaneous users.

Caching

Storing frequently requested AI responses to serve them instantly without re-running the model. Significantly reduces cost and latency for repeated queries.

💎

AI for Business

7

AI Strategy

A deliberate plan for how an organisation will adopt, deploy and govern AI to achieve its goals. Distinguishes companies that use AI strategically from those that just experiment.

AI ROI

The measurable return from AI investment, time saved, revenue generated, costs reduced. The business case that justifies AI adoption.

Automation Workflow

A series of connected automated tasks triggered by an event. AI-powered workflows can handle complex multi-step processes that used to require human intervention.

Prompt Template

A reusable AI prompt with placeholder variables. Standardises AI interactions across a team, ensuring consistent quality and saving time on repetitive tasks.

AI Governance

The policies, processes and oversight structures that ensure AI is used responsibly within an organisation, covering ethics, compliance, and risk management.

Digital Transformation

The integration of digital technology, including AI, across all areas of a business, fundamentally changing how it operates and delivers value to customers.

No-Code AI

AI tools that let non-technical users build, customise and deploy AI applications through visual interfaces, without writing any code.

Want to actually use this knowledge?

From understanding AI to using it in your business

The BrightMind Academy and our 20 free + premium tools take you from terminology to practical, applied AI for your business, at your pace.

🔮 Explore the Academy See the Toolkit →

Investment Expectations

Build smarter.
Talk to us.

Ready to transform your brand with AI? Let's build something extraordinary together.

BrightMind Studio
Got a question? Ask me 💠