Every AI Term Explained
in Plain English.

AI systems that can take multi-step actions autonomously to complete a goal — browsing the web, writing code, sending emails, or booking meetings without step-by-step human instruction. Goes beyond just answering questions.

A set of step-by-step instructions a computer follows to solve a problem or complete a task. AI algorithms learn patterns from data rather than following fixed rules.

The challenge of ensuring AI systems behave in accordance with human values and intentions. A major research area at organisations like Anthropic, OpenAI, and DeepMind. Central to AI safety.

Hypothetical AI that can perform any intellectual task that a human can, with the ability to learn and reason across domains. Does not yet exist. Current AI systems are "narrow AI" — very good at specific tasks.

The tendency for humans to over-rely on automated systems and AI recommendations without applying critical judgment. A significant risk in healthcare, legal, and financial AI deployments.

When an AI system produces systematically unfair or skewed outputs — often reflecting biases present in its training data. For example, hiring AI trained on historical data may discriminate against certain groups.

AI systems whose decision-making process is opaque — even to their creators. You see the input and output but cannot easily explain the reasoning. Deep learning models are often "black boxes." Contrast with "explainable AI."

AI's ability to interpret and understand visual information from images and video. Used in medical imaging (spotting tumours in scans), autonomous vehicles, manufacturing quality control, and facial recognition.

The amount of text an AI language model can "see" and consider at once when generating a response. Measured in tokens. A larger context window allows the AI to work with longer documents.

A type of machine learning that uses neural networks with many layers to learn complex patterns. The technology behind most modern AI breakthroughs — image recognition, voice assistants, language models.

The type of AI model that generates images (used by Midjourney, DALL-E, Stable Diffusion). Works by learning to "denoise" random noise into coherent images. Responsible for the AI image generation explosion of 2022–2025.

A way of converting text, images, or other data into lists of numbers (vectors) that capture their meaning, allowing AI systems to find semantic similarities between concepts.

AI systems designed to explain their reasoning in human-understandable terms. Particularly important in healthcare, finance, and legal AI where decisions must be justifiable and auditable.

Training a pre-existing AI model further on specialised data to make it better at specific tasks. For example, fine-tuning GPT-4 on medical literature to create a medical AI assistant. Less expensive than training from scratch.

A large AI model trained on vast amounts of data that can be adapted for many different tasks. GPT-4, Claude, and Gemini are foundation models. They form the "foundation" that other specialised AI tools are built on top of.

AI that creates new content — text, images, audio, video, code — rather than just analysing existing content. ChatGPT, Midjourney, and GitHub Copilot are all generative AI tools. This is the major wave reshaping most professions.

Specialised computer chips originally designed for gaming graphics but now the primary hardware for training AI models. NVIDIA dominates this market. The global shortage of GPUs is a major bottleneck in AI development.

When an AI confidently states something that is factually incorrect or fabricated. A major limitation of current language models. The AI isn't lying — it is generating plausible-sounding text that happens to be wrong.

AI systems designed to include human review or approval at key decision points. Ensures AI outputs are validated before consequential actions are taken. The standard for high-stakes AI in healthcare, legal, and finance.

The process of using a trained AI model to make predictions or generate outputs. When you type a message to ChatGPT and it responds, that is inference. Contrast with "training" (teaching the model).

The use of AI to enhance human intelligence and capability rather than replace it. The framing most aligned with how the most successful AI-era professionals are approaching the technology.

AI models trained on vast amounts of text to understand and generate human language. ChatGPT, Claude, Gemini, and LLaMA are all LLMs. The backbone of most AI tools professionals interact with daily.

The operational practices for deploying, monitoring, and managing large language models in production systems. A fast-growing specialisation within ML engineering.

A branch of AI where systems learn from data to improve performance without being explicitly programmed for each task. All modern practical AI — from Netflix recommendations to fraud detection — is built on machine learning.

In AI, a "model" is a trained system that takes inputs and produces outputs. When people say "the model said X," they mean the AI system produced that output. GPT-4, Claude Sonnet, and Gemini Pro are all examples of models.

AI systems that can process and generate multiple types of content — text, images, audio, and video — in a single model. GPT-4o and Claude 3 are multimodal. Enables richer, more natural interactions with AI tools.

AI's ability to understand, interpret, and generate human language. The field behind chatbots, translation tools, search engines, and language models. LLMs represent the current peak of NLP capability.

A type of machine learning system loosely inspired by the human brain — made of interconnected nodes (neurons) that process data in layers. Deep neural networks with many layers are the foundation of deep learning and modern AI.

The numerical values inside an AI model that are adjusted during training. More parameters generally (but not always) means a more capable model. GPT-4 has approximately 1 trillion parameters. Often used as a rough proxy for model capability.

The input you give to an AI — the text, question, or instruction you type. The quality of your prompt directly impacts the quality of the AI's response. "Prompting" is now a genuine professional skill.

The skill of crafting effective prompts to get the best outputs from AI systems. Involves techniques like chain-of-thought prompting, few-shot examples, role-setting, and output formatting instructions.

The initial training phase where an AI model learns from massive datasets of text, images, or other data. After pre-training, models are typically fine-tuned for specific applications. The most computationally expensive part of creating an AI model.

A technique that improves AI responses by allowing the model to search a specific knowledge base before generating an answer. Helps AI provide accurate, up-to-date information rather than relying solely on training data. Used in enterprise AI deployments to ground AI in company-specific knowledge.

A training technique where human evaluators rate AI outputs, and the model learns to produce outputs humans prefer. Used to make AI models safer, more helpful, and better aligned with human values. ChatGPT was one of the first widely deployed models trained with RLHF.

Instructions given to an AI at the start of a conversation that define its role, behaviour, and constraints. Invisible to end users in most products but critical to how enterprise AI tools behave. Prompt Engineers and AI Product Managers work extensively with system prompts.

Artificially generated data used to train AI models when real data is scarce, private, or expensive to collect. Particularly important in healthcare AI (where patient data has strict privacy protections) and for testing AI systems.

The basic unit of text that AI language models process. Roughly equivalent to 0.75 words in English. Models charge by tokens and have token limits per context window. "1000 tokens ≈ 750 words" is a useful approximation.

Using a model trained on one task as the starting point for training on a different (related) task. Dramatically reduces the data and computing power needed. The basis for how foundation models are adapted into specialised AI products.

The neural network architecture underlying virtually all modern large language models. Introduced in the landmark 2017 paper "Attention Is All You Need" (Vaswani et al., Google Brain — published open access via arXiv). The "T" in GPT stands for Transformer. Revolutionised AI capability and enabled the current wave of generative AI.