AI Glossary129 terms
Every important AI term, defined clearly and explained at three levels. From tokens and transformers to alignment and agents.
A
- Agentic Workflow A way of using AI where the model plans and carries out a multi-step task with some autonomy, deciding its own next actions, rather than answering in one shot.
- AI Adoption The process of getting AI actually used and delivering value in an organisation; usually a people-and-process challenge more than a technical one.
- AI Agent An AI system that can take actions to reach a goal, using tools, taking steps and reacting to results, not just producing a single answer.
- AI Alignment The challenge of making AI systems reliably pursue what people actually intend, and behave safely and honestly while doing it.
- AI Ethics The study and practice of building and using AI in ways that are fair, transparent, accountable and respectful of people's rights.
- AI Literacy The practical understanding of what AI can and cannot do, and the skill of using it well and responsibly.
- AI Safety The field working to make AI systems reliable, controllable and beneficial, and to prevent harm from both today's models and more capable future ones.
- AI Watermarking Hidden signals or labels embedded in AI-generated content so it can later be identified as AI-made.
- Algorithm A step-by-step set of instructions for solving a problem; the basic building block of all software, including AI.
- Anthropic The AI safety company behind the Claude family of models and the Model Context Protocol; a leading frontier lab with a strong safety focus.
- API (for AI)API A standard way for software to talk to an AI model over the internet, so developers can build the model into their own apps.
- Artificial General IntelligenceAGI A hypothetical AI that can match or exceed human ability across the full range of intellectual tasks, not just narrow ones. It does not exist today.
- Artificial IntelligenceAI The field of building computer systems that perform tasks we normally associate with human intelligence, such as understanding language, recognising images or making decisions.
- Attention The mechanism that lets a model weigh how relevant each part of the input is to every other part; the core idea inside the transformer.
- Augmentation vs Automation Two ways to apply AI: augmentation makes people better at their work; automation replaces the human task entirely. Which you choose shapes the outcome.
B
- Backpropagation The algorithm that works out how much each of a network's many settings contributed to its error, so they can all be adjusted efficiently.
- Base vs Instruct Model A "base" model just continues text; an "instruct" (or chat) model has been further trained to follow instructions and behave like an assistant.
- Batch Inference Running many AI requests together in one go to use the hardware efficiently, which lowers cost per request but can add a little delay.
- Benchmark A standard test used to measure and compare how well AI models perform on specific tasks.
- Bias (in AI) When an AI system produces systematically unfair or skewed results, usually because it learned patterns from biased data.
C
- Catastrophic Forgetting When training a model on something new makes it forget things it previously knew; a key challenge in updating AI models.
- Centaur vs Cyborg Two styles of working with AI, "centaur" divides tasks cleanly between human and AI, "cyborg" blends them tightly, switching back and forth moment to moment.
- Chain-of-ThoughtCoT A prompting technique where the model is encouraged to work through a problem step by step, which improves accuracy on harder tasks.
- Co-Intelligence The idea of working *with* AI as a collaborator and thinking partner, rather than treating it as either a magic oracle or a mere tool.
- Code Generation AI that writes, completes, explains or fixes computer code from natural-language instructions or partial code.
- Compute The raw amount of processing power used to train or run an AI model, one of the main factors deciding how capable it can be.
- Computer Use An AI that can operate a computer the way a person does, looking at the screen and clicking, typing and navigating apps to complete tasks.
- Computer VisionCV The field of AI that lets computers interpret and understand images and video.
- Context Engineering The broader discipline of deciding what information to put into a model's limited context window, and how to organise it, to get the best results.
- Context Window The maximum amount of text, measured in tokens, that a model can take into account at one time.
- Copyright and AI The unsettled legal questions about using copyrighted material to train AI, and about who owns what AI produces.
- Cost per Token How AI services usually charge, by the number of tokens processed, both what you send in and what the model generates.
D
- Data Privacy (in AI) Protecting personal and confidential information when it is used with AI, both what goes into models and what they might reveal.
- Deep Learning The branch of machine learning that uses many-layered neural networks to learn complex patterns directly from raw data.
- Deepfake A fake but realistic image, video or audio clip made by AI, typically showing a real person saying or doing something they never did.
- Diffusion Model The kind of generative model behind most modern AI image generators; it creates images by starting from noise and gradually removing it.
- Digital Labor The idea of AI agents doing real work as a kind of "digital workforce" alongside people, and the economic shift that implies.
- Direct Preference OptimizationDPO A simpler alternative to RLHF that tunes a model directly on "this answer is better than that one" comparisons, without a separate reward model or reinforcement learning.
E
- Embedding A list of numbers that captures the meaning of a piece of text (or an image), so that similar meanings end up close together.
- Encoder & Decoder Two building blocks of transformer models; roughly, an encoder "reads and understands" input, a decoder "writes" output. Most chatbots are decoder-only.
- Epoch One complete pass of the training process through the entire training dataset.
- EU AI Act The European Union's comprehensive AI law, which regulates AI systems based on how risky they are.
- Existential Risk (from AI)x-risk The debated worry that extremely advanced AI could, in the long run, pose catastrophic or even civilisation-threatening dangers.
- ExplainabilityXAI The degree to which we can understand why an AI produced a particular result; hard for modern models, which are largely "black boxes".
F
- Few-shot Learning Giving a model a handful of worked examples inside the prompt so it grasps the task, without any extra training.
- Fine-tuning Further training of an already-trained model on a smaller, specific dataset so it specialises in a particular task, style or domain.
- FLOPsFLOPs A measure of raw computation, the number of arithmetic operations, used to quantify how much compute training or running a model takes.
- Foundation Model A large model trained on broad data that serves as a versatile base, adaptable to many different downstream tasks.
- Frontier Model The most capable, cutting-edge AI models at any given time, typically the largest and most expensive, built by a handful of leading labs.
G
- Generative Adversarial NetworkGAN A generative model made of two networks competing, a generator that fakes data and a critic that tries to catch it, which together learn to produce realistic output.
- Generative AIGenAI AI that creates new content, text, images, audio, video or code, rather than only analysing or classifying existing data.
- Google DeepMind Google's AI research division, behind the Gemini models and landmark systems like AlphaGo and AlphaFold.
- GPUGPU A chip originally built for graphics that turns out to be ideal for AI, because it can do enormous numbers of calculations in parallel.
- Gradient Descent The core optimisation method that trains neural networks by repeatedly nudging their settings in the direction that reduces error.
- Guardrails The safety checks and limits placed around an AI system to keep its inputs and outputs within acceptable bounds.
H
I
- In-context LearningICL A model's ability to pick up a new task from information in the prompt alone, without any change to its underlying training.
- Inference Actually using a trained model to produce an answer, as opposed to training it; every time you send a prompt, you are running inference.
- Instruction Tuning Fine-tuning a raw model on examples of instructions and good responses, so it learns to follow requests and behave like a helpful assistant.
K
- Knowledge Distillation Training a smaller, cheaper "student" model to imitate a larger "teacher" model, keeping much of the quality at a fraction of the cost.
- KV CacheKV cache A memory trick that makes AI text generation faster by storing intermediate results so they don't have to be recomputed for every new word.
L
- Large Language ModelLLM An AI system trained on huge amounts of text that generates language by predicting the next piece of text, one step at a time.
- Latency How long an AI takes to respond; a key factor in how usable and pleasant an AI product feels.
- LoRALoRA An efficient fine-tuning method that adapts a model by training a small set of extra weights while leaving the huge original model frozen.
- Loss Function The formula that scores how wrong a model's prediction is; training works by making this number as small as possible.
M
- Machine LearningML A way of building software that learns patterns from examples instead of being programmed with explicit rules.
- Machine TranslationMT AI that translates text (or speech) from one language to another; one of the oldest goals of AI and now remarkably fluent.
- Memory (in AI systems) How an AI application remembers things beyond a single request, since the model itself forgets everything once the conversation ends.
- Meta AI The AI division of Meta (Facebook), best known for the Llama family of open-weight models that power much of the open AI ecosystem.
- Mistral AI A European (French) AI company known for strong, efficient open-weight models; a leading non-US frontier player.
- Mixture of ExpertsMoE A model design that contains many specialised sub-networks ("experts") but uses only a few for each input, giving large capacity at lower running cost.
- Model Collapse The degradation that can happen when AI models are repeatedly trained on AI-generated data, gradually losing quality and diversity.
- Model Context ProtocolMCP An open standard that lets AI applications plug into external tools and data through one common interface, instead of a custom integration for each.
- Multi-Agent Systems Several AI agents working together, often with different roles, to tackle a task that is too big or varied for one agent alone.
- Multimodal AI AI that can work with more than one type of content at once, for example understanding text, images, audio and video together.
N
- Narrow AI AI that is built for specific tasks and has no general understanding; this describes essentially all AI in use today.
- Natural Language ProcessingNLP The field of AI focused on getting computers to understand, interpret and produce human language.
- Neural Network A computing system loosely inspired by the brain, made of layers of simple connected units that learn by adjusting the strengths between them.
O
- Open Source AI AI that is released openly, so anyone can download, run, study and build on it, as opposed to closed models you can only use through a company's service.
- OpenAI The AI lab behind ChatGPT and the GPT family of models; one of the leading frontier AI companies.
- Orchestration The coordination layer that decides which models, tools and steps run, in what order, to complete an AI task reliably.
- Overfitting When a model memorises its training data instead of learning general patterns, so it does great on what it has seen but poorly on anything new.
P
- Parameter One of the many internal, adjustable numbers a model tunes during training; their count (e.g. "70 billion parameters") roughly indicates a model's size.
- Pretraining The first, largest training phase where a model learns general knowledge and skills from a huge amount of data, before any task-specific tuning.
- Prompt The text you give an AI model to tell it what you want; its input.
- Prompt Engineering The craft of writing and refining the instructions you give an AI model so it produces better, more reliable results.
- Prompt Injection A security attack where hidden or malicious instructions in the input trick an AI into ignoring its real instructions and doing something it shouldn't.
- Prompt Template A reusable prompt with blanks to fill in, so the same reliable instruction can be applied to many different inputs.
R
- Reasoning (in AI) An AI's ability to work through multi-step problems logically, rather than just recalling or pattern-matching an answer.
- Reasoning Model A newer type of language model trained to "think" at length, working through a problem internally before giving its answer.
- Red-teaming Deliberately attacking or stress-testing an AI to find its weaknesses, harmful outputs, jailbreaks, failures, before real users or bad actors do.
- Reinforcement LearningRL Training an AI by trial and error, rewarding good actions and penalising bad ones, so it learns a strategy that maximises reward.
- Reinforcement Learning from Human FeedbackRLHF A training method that uses human preferences about which answers are better to make a model more helpful, honest and safe.
- Retrieval-Augmented GenerationRAG A technique that lets a model look up relevant information from your own documents and use it when answering, instead of relying only on what it learned in training.
- ROI of AIROI Whether an AI project actually pays off, its value created versus its full cost; often positive for good use cases, but easy to get wrong.
- Role Prompting Telling an AI to take on a role ("act as a lawyer", "you are a friendly tutor") to shape the style, tone and focus of its answers.
S
- Scaling Laws Predictable patterns showing how AI models get better as you increase their size, training data and compute.
- SDK (for AI)SDK A ready-made code library that makes it easier for developers to build with an AI model, wrapping the raw API in convenient tools.
- Self-supervised LearningSSL Learning from unlabelled data by having the data provide its own answers, for example hiding part of a sentence and predicting it. The engine behind modern foundation models.
- Small Language ModelSLM A language model with relatively few parameters, small enough to run cheaply, on-device or offline, while still being capable for many tasks.
- Speech-to-TextSTT AI that turns spoken audio into written text; the technology behind voice typing, captions and voice assistants.
- Structured Output Making an AI return its answer in a strict, machine-readable format (like JSON) so other software can use it reliably.
- Summarization AI that condenses a long text into a shorter version that keeps the key points; one of the most useful everyday AI tasks.
- SuperintelligenceASI A hypothetical AI far smarter than the best humans across virtually every field; a step beyond even human-level AGI.
- Supervised Learning Training a model on examples that come with the correct answers, so it learns to predict those answers on new data.
- Symbolic AI The older approach to AI based on explicit rules, logic and symbols, in contrast to today's data-driven, learned approach.
- Synthetic Data Artificial training data created by AI or simulation instead of collected from the real world; increasingly important for training modern models.
- System Prompt A behind-the-scenes instruction that sets an AI assistant's role, rules and tone before the user's conversation begins.
T
- Temperature A setting that controls how random or predictable an AI model's output is, low for focused and consistent, high for varied and creative.
- Text-to-Image AI that creates a picture from a written description; type what you want, and it generates a matching image.
- Text-to-SpeechTTS AI that turns written text into natural-sounding spoken audio; the "voice" of assistants, audiobooks and accessibility tools.
- Throughput How much work an AI system can handle in a given time, for example tokens generated or requests served per second.
- Token The small chunk of text, often a word or part of a word, that a language model reads and generates one at a time.
- Tokenizer The component that splits text into tokens before a model can process it, and reassembles tokens back into text.
- Tool Use When an AI model can call external functions, search, code, a calculator, an API, to do things it cannot reliably do with text alone.
- TPUTPU A chip designed by Google specifically for AI workloads, an alternative to GPUs for training and running large models.
- Training Data The collection of examples a model learns from; its quality and content largely determine what the model becomes.
- Transfer Learning Reusing a model trained on one task as the starting point for another, so you need far less data and compute for the new task.
- Transformer The neural-network design, based on an "attention" mechanism, that powers virtually all modern large language models.
- Turing Test A classic thought experiment: if a machine can converse so that people cannot tell it from a human, does it "think"? Historic, but no longer a serious benchmark.
U
V
- Variational AutoencoderVAE A generative model that learns to compress data into a smooth "meaning space" and generate new samples from it; also a building block inside modern image generators.
- Vector Database A database built to store embeddings and find items by meaning rather than exact keywords; the search engine behind RAG.
No terms match your search.