You are reading an article about AI. The author mentions "large language models." You keep going. They reference "transformer architecture." You nod along. Then they drop "RLHF," "embeddings," and "inference" in the same paragraph, and quietly close the tab.
Sound familiar? It should. AI jargon is one of the biggest barriers between ordinary people and genuine understanding of the technology reshaping their world.
This glossary exists to tear that barrier down. Fifty terms. Plain English. Real-world examples. No assumptions about what you already know.
Bookmark this page. Share it with anyone who has ever nodded politely at an AI term they did not understand. And come back to it whenever The AI Vanguard uses a term you want to double-check. This glossary will grow alongside the blog.
How to Use This Glossary: The 50 terms are grouped into five categories: Core Concepts, How AI Works, Types of AI and Models, AI Tools and Products, and AI Ethics and Society. You can read straight through or jump to the section most relevant to what you are trying to understand right now.
Part 1: Core Concepts
These are the foundational terms. Understanding these gives you a solid base for everything else in AI.
3. Deep Learning A subset of machine learning that uses neural networks with many layers to analyse data. Deep learning is responsible for most of the major AI breakthroughs of the past decade, including the ability of AI to understand human language, recognise faces in photos, and generate realistic images. Real-world example: The technology behind voice assistants like Siri and Alexa understanding natural speech uses deep learning.
4. Neural Network A computational system loosely inspired by the structure of the human brain. It consists of interconnected layers of nodes that process information. Data passes through these layers and is transformed at each stage. Neural networks are the foundational architecture of most modern AI systems. Real-world example: When you upload a photo and an app identifies what is in it, a neural network is doing the recognition work.
5. Algorithm A set of rules or instructions that a computer follows to solve a problem or complete a task. In AI, algorithms determine how a system learns from data and makes decisions. Every AI tool you use is powered by one or more algorithms working behind the scenes. Real-world example: The algorithm deciding which posts appear at the top of your social media feed is analysing your past behaviour and predicting what you are most likely to engage with.
6. Data In the context of AI, data refers to the raw information that AI systems learn from. This can be text, images, audio, video, numbers, or any other form of recorded information. The quality and quantity of data an AI system trains on largely determines how capable it becomes. Real-world example: ChatGPT was trained on a significant portion of the text available on the internet, billions of web pages, books, and articles, all of which count as training data.
7. Training The process through which an AI system learns from data. During training, the system processes vast amounts of examples, makes predictions, receives feedback on those predictions, and adjusts its internal settings to improve over time. Training can take weeks and require enormous computing power. Real-world example: Before GPT-4 was released by OpenAI, it went through an extensive training process during which it processed enormous volumes of text data.
8. Model The finished product of the training process. When an AI system has been trained on data, the result is called a model. The model contains everything the system learned during training and is what you interact with when you use an AI tool. Real-world example: Claude, GPT-4, and Gemini are all examples of AI models. When you chat with them, you are interacting with the trained model.
9. Parameters The internal numerical settings of an AI model that are adjusted during training to improve performance. A model's parameter count is often used as a rough indicator of its size and capability. Modern large language models have billions or even trillions of parameters. Real-world example: When you hear that a model has "70 billion parameters," it means there are 70 billion individual numerical values inside it that were tuned during training.
10. Inference The process of using a trained AI model to generate a response or make a prediction based on new input. When you type a question into ChatGPT and it responds, that response is generated during inference. Training happens once. Inference happens every time someone uses the model. Real-world example: Every time you send a message to an AI chatbot and receive a reply, that reply is being generated through inference.
Part 2: How AI Works
These terms explain the mechanics behind modern AI systems. You do not need to be an engineer to understand them, but knowing them will help you make sense of how AI tools behave.
11. Large Language Model (LLM) A type of AI model trained on enormous amounts of text data to understand and generate human language. LLMs are the technology behind AI chatbots like ChatGPT, Claude, and Gemini. They work by predicting what text should come next given what has come before, at an extraordinarily sophisticated level. Real-world example: When you ask an AI chatbot to summarise a document, explain a concept, or write an email, you are using a large language model.
12. Transformer The specific neural network architecture that powers most modern large language models. Introduced in a landmark 2017 research paper by Google engineers, the transformer architecture revolutionised what AI could do with language. The "T" in ChatGPT stands for transformer. Real-world example: Every major AI chatbot in use today, including ChatGPT, Claude, and Gemini, is built on transformer architecture.
13. Prompt The input or instruction you give to an AI system. A prompt can be a question, a command, a description, or any combination of these. The quality of the prompt you write has a significant impact on the quality of the response you receive. Real-world example: "Summarise this article in three bullet points for a non-technical audience" is a prompt. "Write me a poem" is a prompt. Everything you type into an AI chatbot is a prompt.
14. Prompt Engineering The skill of crafting effective prompts to get the best possible results from an AI system. Prompt engineering has become a genuinely valuable professional skill as AI tools have become more powerful. A well-engineered prompt can produce dramatically better results than a vague one. Real-world example: Instead of prompting "write a blog post," a prompt engineer might write "write a 1,000-word blog post for small business owners in Australia about using AI for customer service, using a friendly and practical tone, with three subheadings and a bullet point summary at the end."
15. Context Window The amount of text an AI model can process and consider at one time. Think of it as the model's working memory. Content outside the context window is effectively invisible to the model during a conversation. Larger context windows allow AI to handle longer documents and more complex tasks. Real-world example: If an AI model has a context window of 100,000 tokens, it can process roughly 75,000 words in a single conversation before earlier content starts being pushed out.
16. Token The basic unit of text that AI language models process. A token is roughly equivalent to a word or part of a word. Models do not read text the way humans do. They break it down into tokens and process those tokens. Pricing for AI API usage is often measured in tokens. Real-world example: The word "artificial" might be one token. The word "AI" is likely one token. A comma or full stop is a token. Roughly 100 tokens equals approximately 75 words.
17. Temperature A setting that controls how creative or predictable an AI model's responses are. A low temperature produces more consistent, factual, and conservative responses. A high temperature produces more varied, creative, and occasionally surprising responses. Most AI tools let users adjust this setting. Real-world example: Setting a low temperature is useful when you need accurate factual information. Setting a higher temperature works better for creative writing tasks where variety and originality are desirable.
18. Hallucination When an AI model generates information that is factually incorrect but presented with confidence. Hallucinations occur because AI models generate statistically likely text rather than verified facts. They do not "know" they are wrong. This is one of the most important limitations of current AI to understand. Real-world example: An AI asked to summarise a legal case might confidently cite a court ruling that does not exist. This is a hallucination. Always verify important factual claims made by AI tools.
19. Fine-Tuning The process of taking an already-trained general AI model and continuing to train it on a specific dataset to make it better at a particular task or domain. Fine-tuning allows organisations to customise general AI models for their specific use cases without training a model from scratch. Real-world example: A hospital in the United Kingdom might fine-tune a general language model on medical literature to make it more useful and accurate for clinical documentation tasks.
20. Retrieval-Augmented Generation (RAG) A technique that gives AI models access to external knowledge sources at the time of generating a response, rather than relying solely on what they learned during training. RAG helps reduce hallucinations and allows AI to work with up-to-date or proprietary information. Real-world example: An AI customer service bot for a bank in Canada that can look up a specific customer's account details before generating a response is using a form of RAG.
21. Embeddings A way of representing words, phrases, or concepts as numerical vectors so that AI systems can process and understand their meaning and relationships. Words with similar meanings end up with similar numerical representations. Embeddings are a fundamental part of how language models understand context. Real-world example: Because of embeddings, an AI model understands that "dog" and "puppy" are closely related concepts, and that "bank" means something different in "river bank" versus "bank account."
22. Reinforcement Learning from Human Feedback (RLHF) A training technique where human raters evaluate AI outputs, and those ratings are used to train the model to produce better responses. RLHF is one of the key methods used to make AI chatbots more helpful, harmless, and honest. OpenAI and Anthropic both use variations of this approach. Real-world example: The reason ChatGPT and Claude refuse to help with clearly harmful requests is partly the result of RLHF. Human trainers rated harmful outputs poorly, and the model learned to avoid them.
Part 3: Types of AI and Models
AI comes in many forms. These terms help you understand the different categories of AI systems and what distinguishes them from one another.
23. Narrow AI (Artificial Narrow Intelligence) AI systems that are designed to perform one specific type of task extremely well. Every AI tool in widespread use today is a form of narrow AI. Despite being called narrow, these systems can outperform humans at their specific tasks. They cannot generalise beyond what they were trained to do. Real-world example: A chess-playing AI is narrow AI. It can beat any human player at chess but cannot do anything else. ChatGPT is narrow AI that is very broad within the domain of language tasks.
24. Artificial General Intelligence (AGI) A hypothetical AI system that can perform any intellectual task that a human being can, across any domain, without needing to be specifically trained for each one. AGI does not currently exist. Whether and when it will exist is one of the most debated questions in technology. Real-world example: An AGI system could switch from writing a legal brief to diagnosing a medical condition to composing music to solving a mathematics problem, just as a highly intelligent human could.
25. Generative AI A category of AI systems designed to generate new content, including text, images, audio, video, and code. Generative AI is the technology behind ChatGPT, Midjourney, DALL-E, Sora, and most of the AI tools that entered mainstream use from 2022 onwards. Real-world example: When you ask an AI to write a blog post, generate an image from a description, or compose a piece of music, you are using generative AI.
26. Multimodal AI An AI system that can process and generate multiple types of content, such as text, images, and audio, rather than being limited to a single format. The latest versions of GPT-4, Claude, and Gemini are multimodal, meaning they can understand and respond to images as well as text. Real-world example: Uploading a photo of a recipe and asking an AI to suggest ingredient substitutions requires multimodal AI, because it must process both the image and the text question.
27. Foundation Model A large AI model trained on broad data that can be adapted for a wide range of downstream tasks. Foundation models are the base upon which many specific AI applications are built. GPT-4, Claude 3, and Gemini Ultra are all examples of foundation models. Real-world example: Many AI startups in the United States and United Kingdom build their products on top of foundation models from OpenAI, Anthropic, or Google rather than training their own models from scratch.
28. Open Source Model An AI model whose underlying code and sometimes its trained weights are made publicly available, allowing anyone to inspect, use, modify, and build upon it. Open source models offer transparency and flexibility but require more technical knowledge to deploy. Meta's Llama series is a prominent example. Real-world example: Researchers at universities in Australia and Canada use open source AI models to conduct AI safety research without needing to pay for access to proprietary models.
29. Diffusion Model A type of AI model used primarily for generating images, audio, and video. Diffusion models work by learning to reverse a process of gradually adding noise to data. They are the technology behind tools like Midjourney, Stable Diffusion, and DALL-E. Real-world example: When you type a description into an AI image generator and a realistic image appears seconds later, a diffusion model is doing the generation.
30. Agent (AI Agent) An AI system that can take actions autonomously to achieve a goal, rather than simply responding to a single prompt. AI agents can use tools, browse the web, write and execute code, and complete multi-step tasks with minimal human intervention. Agent-based AI is one of the fastest-growing areas of AI development in 2026. Real-world example: An AI agent tasked with researching competitors for a business could independently search the web, read multiple websites, compile the findings, and produce a formatted report, all without being guided step by step.
Part 4: AI Tools, Products, and Companies
These terms cover the specific tools, platforms, and organisations you will encounter most frequently when following AI news and using AI products.
31. ChatGPT An AI chatbot developed by OpenAI, based in San Francisco, United States. ChatGPT was launched in November 2022 and became the fastest-growing consumer application in history, reaching 100 million users in two months. It is powered by the GPT series of large language models and is available in free and paid versions. Real-world example: A freelance copywriter in New Zealand uses ChatGPT to generate first drafts of marketing copy, which she then edits and refines before delivering to clients.
32. Claude An AI chatbot and large language model developed by Anthropic, a safety-focused AI company founded in San Francisco, United States. Claude is widely regarded as one of the strongest AI models for nuanced writing, analysis, and following complex instructions. It is available at claude.ai. Real-world example: A legal professional in London uses Claude to help summarise lengthy case documents before review sessions.
33. Gemini Google's family of large language models and AI products, developed by Google DeepMind in the United States and United Kingdom. Gemini is integrated into Google's products including Google Search, Gmail, Google Docs, and the Gemini chatbot at gemini.google.com. Real-world example: A teacher in Toronto uses Gemini to help create differentiated lesson plans tailored to students with different learning needs.
34. GPT (Generative Pre-trained Transformer) The family of large language models developed by OpenAI. GPT models are trained on vast amounts of text data and are the technology powering ChatGPT. Each new version (GPT-3, GPT-4, and beyond) has brought significant improvements in capability. Real-world example: When someone says they used "GPT" to write something, they typically mean they used ChatGPT, which is built on one of OpenAI's GPT models.
35. API (Application Programming Interface) A set of rules that allows one piece of software to communicate with another. In AI, APIs allow developers to access AI model capabilities and integrate them into their own products and services. Most major AI companies offer APIs that businesses use to build AI-powered features. Real-world example: A startup in Sydney building an AI customer service tool might use the OpenAI API to power the conversational intelligence behind their product.
36. Midjourney One of the most popular AI image generation tools, used to create visual art from text descriptions. Midjourney operates primarily through Discord and produces images known for their artistic quality. It is used by designers, marketers, and creatives worldwide. Real-world example: A graphic designer in Auckland uses Midjourney to generate concept art for client presentations before committing to full designs.
37. Stable Diffusion An open source AI image generation model that can be run locally on personal computers or accessed through various online platforms. Because it is open source, it has spawned a large ecosystem of tools and applications built around it. Real-world example: Developers and artists who want full control over their AI image generation often choose Stable Diffusion because they can run it on their own hardware without paying for access.
38. Copilot Microsoft's AI assistant, integrated across Microsoft 365 products including Word, Excel, Outlook, and Teams, as well as the Windows operating system and Bing search. Microsoft, based in Redmond, United States, made a major investment in OpenAI and uses its models to power Copilot. Real-world example: An accountant in Melbourne uses Copilot in Excel to automatically analyse monthly financial data and generate summary reports.
39. Perplexity AI An AI-powered search engine that combines traditional web search with AI-generated answers, including citations to sources. Unlike standard AI chatbots, Perplexity searches the web in real time, making it more useful for current information and fact-checking. Real-world example: A journalist in Manchester uses Perplexity to quickly gather background information on a breaking story, with source citations she can verify before publishing.
40. Hugging Face A platform and community for sharing, discovering, and deploying AI models and datasets. Often described as the GitHub of AI, Hugging Face hosts thousands of open source models and is widely used by researchers and developers. It is headquartered in New York, United States. Real-world example: A researcher at a university in Edinburgh uses Hugging Face to find and test existing AI models for a natural language processing project.
Part 5: AI Ethics, Safety, and Society
These terms cover the broader impact of AI on people, organisations, and society. Understanding them is essential for anyone who wants to engage seriously with the AI conversation.
41. AI Bias The tendency of AI systems to produce outputs that reflect and often amplify the biases present in their training data. Because AI learns from human-generated data, and humans have biases, AI systems can perpetuate and scale those biases in ways that cause real harm. Real-world example: An AI hiring tool trained predominantly on CVs from male applicants in the United States was found to systematically score female applicants lower, because the training data reflected historical hiring patterns rather than actual merit.
42. AI Safety The field of research concerned with ensuring that AI systems behave in ways that are safe, predictable, and aligned with human values, both in the near term and as systems become more powerful. AI safety research is conducted by organisations including Anthropic in San Francisco, the UK AI Safety Institute, and various academic institutions. Real-world example: AI safety researchers work to ensure that as AI systems become more capable, they remain under meaningful human oversight and do not develop goals that conflict with human wellbeing.
43. Alignment The challenge of ensuring that an AI system's goals and behaviours are aligned with the intentions of its designers and the values of humanity more broadly. Misaligned AI could pursue goals in ways that are harmful, even without any malicious intent. Alignment is one of the central problems in AI safety research. Real-world example: Ensuring that an AI system optimises for genuinely helping users rather than just generating responses that seem helpful on the surface is an alignment challenge.
44. Deepfake Synthetic media, typically video or audio, in which a person's likeness or voice has been convincingly replaced with that of another person using AI. Deepfakes have legitimate creative applications but have also been used to spread misinformation, create non-consensual content, and facilitate fraud. Real-world example: Several politicians in the United States and United Kingdom have been targeted by deepfake videos designed to make them appear to say things they never said. Detecting and flagging deepfakes is an active area of AI research.
45. Explainability (XAI) The ability to understand and explain how an AI system reached a particular decision or output. Many modern AI models are effectively "black boxes" whose internal reasoning is opaque. Explainability is increasingly important in regulated industries like healthcare, finance, and law, where decisions must be justifiable. Real-world example: A bank in Canada using AI to decide whether to approve loan applications faces regulatory pressure to explain why the AI made specific decisions, making explainability a legal as well as ethical concern.
46. Large-scale AI Regulation Legal frameworks and policies designed to govern the development, deployment, and use of AI systems. Regulatory approaches vary significantly by country. The European Union's AI Act, which affects businesses operating in the UK and EU, is one of the most comprehensive regulatory frameworks yet enacted. Real-world example: The EU AI Act classifies AI systems by risk level and imposes strict requirements on high-risk applications such as AI used in healthcare, critical infrastructure, and law enforcement.
47. Synthetic Data Artificially generated data that mimics real data and is used to train AI models. Synthetic data helps address privacy concerns, fill gaps in real datasets, and reduce the cost of data collection. It is particularly valuable in fields like healthcare, where real patient data is sensitive. Real-world example: A healthcare AI company in Australia uses synthetic patient data to train its models, avoiding the need to work with real patient records and the privacy complications that would entail.
48. Digital Watermarking A technique for embedding invisible markers into AI-generated content, such as images, text, or audio, to indicate that it was created by AI. Digital watermarking is being explored as a way to combat misinformation and help people distinguish AI-generated content from human-created content. Real-world example: Google's DeepMind has developed a watermarking tool called SynthID that embeds invisible markers into AI-generated images, allowing detection tools to identify them even after editing.
49. Autonomous Weapons Military systems that use AI to identify and engage targets without human intervention. The development of autonomous weapons is one of the most ethically contested applications of AI. International discussions about regulating or banning lethal autonomous weapons systems are ongoing at the United Nations. Real-world example: The debate around autonomous weapons involves fundamental questions about accountability, proportionality, and whether life-and-death decisions in conflict should ever be delegated to machines.
50. AI Literacy The ability to understand, evaluate, and effectively use artificial intelligence tools and concepts. AI literacy is increasingly considered a foundational skill for participation in modern society, comparable to basic reading, numeracy, or digital literacy. Governments in the United Kingdom, Canada, Australia, and New Zealand have begun incorporating AI literacy into national education frameworks. Real-world example: The AI Vanguard exists specifically to build AI literacy. Every post on this blog is a contribution to your understanding of the technology shaping your world.
You Now Have the Language
Fifty terms. Five categories. All the foundational vocabulary you need to read, watch, listen to, and engage with AI content without hitting a wall of jargon.
Save this page. When The AI Vanguard uses a technical term in a future post, this glossary will be your reference. As the AI landscape evolves and new terms become relevant, this glossary will be updated to reflect them.
The next post on The AI Vanguard goes live tomorrow morning. It covers one of the most-searched questions in AI right now: is ChatGPT still the best AI chatbot available in 2026? The honest answer might surprise you.
Key Takeaways
• AI has its own language. Learning it removes one of the biggest barriers to understanding the technology that is reshaping the world
• The 50 terms in this glossary cover the five core areas you need: foundational concepts, how AI works, types of AI, tools and companies, and ethics and society
• Hallucination, bias, and alignment are not just technical terms. They are real, present limitations that affect how safely AI can be used today
• Understanding AI terminology helps you evaluate AI tools more critically, spot misleading claims, and make better decisions about how and when to use AI
• AI literacy is becoming a foundational life skill. Building it starts with knowing what the words mean
Frequently Asked Questions
What is the most important AI term for a beginner to understand first?
Start with hallucination. Understanding that AI can generate confident, fluent, completely wrong information is the single most important thing to grasp before using any AI tool for anything important. It shapes how carefully you should verify AI outputs and protects you from making decisions based on fabricated information.
What is the difference between a model and a tool?
A model is the underlying AI system, the result of training on large amounts of data. A tool is the application or interface that gives you access to the model. ChatGPT is a tool. The GPT-4 model is what powers it. Claude.ai is a tool. The Claude model is what runs underneath it. Many different tools can be built on top of the same underlying model.
Are all AI chatbots built the same way?
No. While most major AI chatbots use transformer-based large language models, the specific architecture, training data, training methods, safety approaches, and fine-tuning differ significantly between companies. This is why ChatGPT, Claude, and Gemini often give different responses to the same question, with different strengths and weaknesses.
Is generative AI the same as artificial intelligence?
Generative AI is a subset of artificial intelligence. AI is the broader field encompassing any system that performs tasks requiring intelligence. Generative AI specifically refers to systems designed to create new content such as text, images, audio, and video. Not all AI is generative. A fraud detection system is AI but it is not generative AI, because it detects patterns rather than creating content.
Where can I keep learning about AI terminology?
Right here. The AI Vanguard publishes two posts every day covering every aspect of AI from beginner-level foundations to advanced industry analysis. Subscribe to the email list below to receive every new post directly in your inbox. This glossary will also be updated regularly as new terms become relevant.
Never Miss a Post: The AI Vanguard publishes two fresh posts every single day. Subscribe below and get every new post delivered directly to your inbox. Free. No spam. Unsubscribe any time. Join readers across the world who stay ahead of AI with The AI Vanguard.
