Zero to AI Guide · 9 min read
AI Vocabulary: 50 Terms Every Professional Needs to Know
Plain-English definitions of 50 essential AI terms — from LLMs and prompt engineering to hallucination and RAG. No jargon, no PhD required.
AI conversations are full of terms that sound technical but aren't complicated once they are explained clearly. Here are 50 of the most important — in plain English.
A
- Agent
- An AI system that can take actions autonomously — browsing the web, writing code, sending emails — not just answer questions. Agents can complete multi-step tasks without constant human input.
- Anthropic
- The AI safety company that makes Claude. Founded in 2021 by former OpenAI researchers.
- Artificial Intelligence (AI)
- Software that performs tasks that previously required human intelligence — reading, writing, reasoning, recognising images, making decisions.
- Autonomous AI
- AI that can operate independently over extended periods without step-by-step human instruction. See also: Agent.
B
- Benchmark
- A standardised test used to measure an AI model's performance. Benchmarks allow fair comparison between different models.
- Bias
- When an AI systematically favours or disfavours certain outputs due to patterns in its training data. A hiring AI trained on historical data may inherit historical biases.
C
- Chain of Thought (CoT)
- A prompting technique where you ask the AI to reason step by step before answering. Produces better results on complex problems.
- Claude
- Anthropic's AI assistant. Known for careful, nuanced responses and strong performance on writing, analysis, and coding tasks.
- Context Window
- The maximum amount of text an AI model can process at once, including the conversation history and any documents you share. Measured in tokens.
- Completion
- The text an AI generates in response to a prompt. 'Text completion' was an early framing for what AI does — you give it the start, it finishes.
D–E
- Diffusion Model
- The type of AI behind image generators like Midjourney and DALL-E. It works by learning to reverse a process of adding random noise to images.
- Embedding
- A way of representing text as numbers that capture meaning — words with similar meanings get similar numbers. Embeddings power search and recommendation systems.
- Emergent Behaviour
- When an AI model displays capabilities that weren't explicitly trained — they 'emerge' from scale. Large models often surprise researchers with what they can do.
F–G
- Few-Shot Prompting
- Giving an AI a few examples of what you want before asking it to do the task. More effective than zero-shot (no examples) for many tasks.
- Fine-Tuning
- Taking a pre-trained model and training it further on specific data to improve performance on a specific task or domain.
- Foundation Model
- A large AI model trained on broad data that can be adapted for many tasks. GPT-4, Claude, and Gemini are all foundation models.
- Generative AI
- AI that creates new content — text, images, code, audio, video — rather than simply classifying or predicting.
- GPT
- Generative Pre-trained Transformer. The architecture behind OpenAI's models. 'GPT-4o' is the current main model.
- Guardrails
- Rules and filters built into AI systems to prevent harmful, offensive, or dangerous outputs.
H–I
- Hallucination
- When an AI generates plausible-sounding but factually incorrect information. A major limitation. Always verify important facts from AI.
- Human-in-the-Loop
- A system where humans review and approve AI outputs at key stages, rather than AI operating fully autonomously.
- Inference
- The process of running an AI model to generate outputs. 'Inference time' is how long it takes to generate a response.
- Instruction Tuning
- Training an AI specifically to follow instructions, making it more useful as an assistant. Most modern LLMs are instruction-tuned.
L–M
- Large Language Model (LLM)
- The type of AI behind ChatGPT, Claude, Gemini, etc. Trained on vast amounts of text to understand and generate human language.
- Latency
- How long it takes for an AI to respond. Lower latency = faster responses. Relevant when building AI-powered products.
- Machine Learning (ML)
- A subfield of AI where systems learn from data rather than being explicitly programmed. LLMs are built using machine learning.
- Model
- The trained AI system. When people say 'the model', they mean the underlying AI engine — not the product wrapper around it.
- Multimodal
- An AI that can work with multiple types of input and output — text, images, audio, and video. Claude and GPT-4o are multimodal.
N–O
- Natural Language Processing (NLP)
- The field of AI that deals with understanding and generating human language. All modern chatbots and writing assistants use NLP.
- Neural Network
- The underlying structure of most modern AI — loosely inspired by the human brain. Consists of layers of interconnected mathematical operations.
- OpenAI
- The company behind ChatGPT and GPT-4. Originally non-profit, now for-profit. One of the two leading AI labs alongside Anthropic.
- Output
- What the AI produces in response to your input. Also called a 'completion' or 'response'.
P–R
- Parameters
- The numerical values that define how an AI model behaves — the 'weights' learned during training. GPT-4 reportedly has over 1 trillion parameters.
- Pre-training
- The initial phase of training an LLM — processing enormous amounts of text to learn language patterns. Extremely computationally expensive.
- Prompt
- The instruction, question, or input you give to an AI model. The quality of the prompt directly determines the quality of the output.
- Prompt Engineering
- The skill of writing prompts that produce consistently high-quality AI outputs. Often the difference between a useful AI and a mediocre one.
- RAG (Retrieval-Augmented Generation)
- A technique that allows AI to retrieve relevant information from a database before generating a response — giving it access to up-to-date or private knowledge.
- RLHF (Reinforcement Learning from Human Feedback)
- A training technique where humans rate AI outputs and the model is trained to produce higher-rated outputs. Used to align AI to human preferences.
S–T
- System Prompt
- An instruction given to an AI before the conversation starts — setting its role, personality, constraints. Users often cannot see system prompts.
- Temperature
- A setting that controls how 'creative' or 'random' an AI's outputs are. Low temperature = predictable. High temperature = more varied and creative.
- Token
- The unit AI models use to process text — roughly 3–4 characters or 0.75 words. AI costs and limits are often measured in tokens.
- Training
- The process of exposing an AI model to data and adjusting its internal weights to improve performance. Training large models costs millions of dollars.
- Training Data
- The text (or images, audio, etc.) used to train an AI model. The quality and breadth of training data heavily influences what the model can do.
- Transformer
- The architecture that underlies almost all modern LLMs. Introduced by Google in 2017. The T in GPT, ChatGPT, etc.
V–Z
- Vector Database
- A database optimised for storing embeddings — used in RAG systems to find relevant documents quickly. Examples: Pinecone, Weaviate, Chroma.
- Vibe Coding
- Using natural language to describe what software should do and letting AI write the code. No traditional coding skills required.
- Zero-Shot
- Asking an AI to perform a task without giving any examples. Contrast with few-shot (a few examples) and one-shot (one example).
Tools we use & recommend
Start with the right AI tools
Claude ProRecommended
The AI used in the Zero to AI workshop. Best for writing, analysis, and reasoning.
£18/month →
ChatGPT PlusPopular
OpenAI's flagship. Strong for research, browsing, and image generation.
£20/month →
Notion AIProductivity
AI built into your workspace. Great for notes, docs, and project management.
From £8/month →
Some links may be affiliate links. We only recommend tools we actually use.
Frequently Asked Questions
What is the difference between AI and machine learning?+
Machine learning is a technique used to build AI systems. AI is the broader goal — making computers do intelligent things. Machine learning is one of the main methods used to get there.
What does 'hallucination' mean in AI?+
When an AI generates text that sounds convincing but is factually wrong, that is called hallucination. The AI is not lying — it is pattern-matching incorrectly. Always verify important outputs.
What is a prompt?+
A prompt is the instruction or question you give to an AI model. Better prompts produce better outputs. Learning to write good prompts is the core skill of working effectively with AI.
What does 'context window' mean?+
The context window is the amount of text an AI can 'see' at once — including your conversation history and any documents you have shared. Larger context windows allow AI to work with longer documents.
What is fine-tuning?+
Fine-tuning is the process of taking a pre-trained AI model and training it further on specific data to make it better at a specific task. A customer service bot fine-tuned on your company's FAQs is more useful than a general-purpose AI for that specific job.
One-Day Course · Richmond, London
Stop reading about AI. Start using it.
Zero to AI is a one-day intensive for professionals who want to go from curious to capable. Eight seats. Hands-on from minute one. You leave with a working system, not a slide deck.
Reserve your seat →