Lietuviškai
Deutsch
Français
Русский
Español
Polski
Italiano
Українська
Čeština
Português
日本語
Latviešu valoda
Eesti
Magyar
Bahasa Indonesia
Türkçe
हिन्दी
Ελληνικά
العربية
ไทย
Tiếng Việt
Română
Bahasa Melayu
বাংলা
This AI glossary explains 40 terms you will meet in articles, product pages and meetings, each in plain words and in a sentence or two. It is grouped by what the term is about, so you can read one section when you need it, and every entry that has a deeper article on this blog links to it. The aim is that after ten minutes you can follow a conversation about AI without nodding along to words you do not know.
Two notes before the list. First, the field moves quickly and vendors use some words loosely, so definitions here describe the common meaning and mention where usage varies. Second, nothing in this glossary names a particular model or company; the ideas apply to any assistant, including Ask Mio.
AI glossary, part 1: the basics
These ten terms describe what AI systems are and how they are built and run.
| Term | Plain meaning | Go deeper |
|---|---|---|
| 1. Artificial intelligence (AI) | Software that performs tasks we associate with human thinking, such as understanding language, recognising images or making predictions. It is a broad label, not a single technology. | What is an AI assistant |
| 2. Machine learning | A way of building software by showing it many examples so it learns patterns, instead of writing rules by hand. Most modern AI is machine learning. | — |
| 3. Neural network | A machine learning design made of layers of simple connected units whose strengths are adjusted during training. Very large ones power today’s language and image systems. | — |
| 4. Large language model (LLM) | A neural network trained on huge amounts of text to predict what comes next. That one skill, at scale, produces answers, summaries, translations and code. | How AI models differ |
| 5. Model | The trained system itself: the finished set of numbers that turns an input into an output. An assistant may use several models behind one chat window. | Model routing explained |
| 6. Training | The long, expensive process in which a model learns from data. It happens before you use the model, and your chat does not normally change it. | — |
| 7. Inference | Using a trained model to produce an answer. Every time you press send, inference runs, and it is what costs the provider compute. | AI inference explained |
| 8. Parameter | One adjustable number inside a model. Parameter counts, in billions, are a rough measure of size, not of quality. | — |
| 9. Token | A small chunk of text, often a short word or part of one, that a model reads and writes. Usage is often measured in tokens; Ask Mio uses points instead. | Tokens vs points |
| 10. Context window | How much text a model can consider at once, including your question, files and the earlier conversation. Beyond it, older content drops out of view. | AI context windows explained |
Prompting terms
These eight terms describe how you talk to a model and how that conversation can be shaped or attacked.
| Term | Plain meaning | Go deeper |
|---|---|---|
| 11. Prompt | The instruction or question you give the AI, plus any context you attach. Clearer prompts give better answers. | How to write an AI prompt |
| 12. System prompt | Hidden standing instructions that set an assistant’s role, tone and rules before you type anything. You usually cannot see it, but projects let you add your own. | What is a system prompt |
| 13. Zero-shot and few-shot | Zero-shot means asking with no examples; few-shot means including a handful of examples of the output you want. Examples often improve consistency. | Zero-shot vs few-shot |
| 14. Chain of thought | Asking a model to work through a problem in steps before answering, which tends to help on logic, maths and planning tasks. | Chain-of-thought prompting |
| 15. Temperature | A setting that controls how adventurous the wording is. Low values give steadier answers; high values give more varied, sometimes less reliable ones. | AI temperature explained |
| 16. Role prompting | Starting with “act as a…” to steer style and focus. It changes tone more than it adds real expertise. | Role prompting explained |
| 17. Prompt chaining | Splitting a big task into a sequence of smaller prompts, each building on the last, so every step can be checked. | Prompt chaining explained |
| 18. Prompt injection | An attack in which hidden text in a web page or document tells the AI to ignore its instructions. Relevant whenever an assistant reads outside content. | Prompt injection explained |
Quality, safety and behaviour
These eight terms describe how well an assistant behaves and where it tends to fail.
| Term | Plain meaning | Go deeper |
|---|---|---|
| 19. Hallucination | A confident answer that is false or invented, such as a fake citation. It happens because models produce likely text, not verified facts. | AI hallucinations explained |
| 20. Bias | Systematic skew in outputs inherited from training data or design, which can disadvantage groups or favour certain viewpoints. | AI bias explained |
| 21. Sycophancy | The tendency to agree with you or flatter you instead of correcting you. Asking for criticism explicitly helps. | AI sycophancy |
| 22. Guardrails | Rules and filters that keep an assistant within safe or intended behaviour, such as refusing harmful requests. | AI guardrails explained |
| 23. Knowledge cutoff | The date after which a model has no training data. For anything newer it needs web search or documents you provide. | AI knowledge cutoff explained |
| 24. Latency | The delay before an answer appears. Longer inputs, bigger models and extra reasoning steps all increase it. | AI latency explained |
| 25. Benchmark | A standard test used to compare models. Scores are useful signals but rarely predict how a model performs on your own task. | AI benchmarks explained |
| 26. Reasoning model | A model that spends extra effort working through a problem before it answers, trading speed for accuracy on hard questions. | AI reasoning models explained |
Techniques and architecture
These ten terms describe how models are designed and improved.
| Term | Plain meaning | Go deeper |
|---|---|---|
| 27. Transformer | The neural network design, built around attention, that underlies almost all modern language models. Attention lets each word weigh every other word. | Transformer models explained |
| 28. Embedding | A list of numbers that captures the meaning of a piece of text or an image, so similar things end up close together. It powers semantic search. | AI embeddings explained |
| 29. RAG (retrieval-augmented generation) | Searching your documents for relevant passages first, then letting the model answer using them. It grounds answers in your own material. | RAG AI explained |
| 30. Fine-tuning | Further training a model on a specific set of examples to change its behaviour or style. Often a good prompt achieves enough without it. | Fine-tuning vs prompting |
| 31. Mixture of experts | A design in which a model contains many specialist parts and activates only a few per request, so large models run faster than their size suggests. | Mixture of experts explained |
| 32. Quantization | Storing a model’s numbers with less precision so it needs less memory and runs faster, usually with a small loss in quality. | — |
| 33. Small language model | A compact model that is cheaper and faster than a large one and can match it on narrow tasks, even on a phone or laptop. | Small language models explained |
| 34. Open-weight model | A model whose trained numbers are published so others can run it. It is not always fully open source, because training data and code may stay private. | Open-weight vs closed models |
| 35. Multimodal AI | A system that handles more than text: images, documents, audio or video, in or out. | What is multimodal AI |
| 36. Diffusion model | The method behind many image generators: start from random noise and remove it step by step, guided by your description. | — |
Using AI in practice
The last four terms describe how assistants act on your behalf and how their output is labelled.
| Term | Plain meaning | Go deeper |
|---|---|---|
| 37. Agent | An AI system that plans and carries out multi-step tasks, using tools along the way, instead of only answering one question. | AI agents explained |
| 38. Model routing | Automatically choosing the best model for each request, so you do not have to pick one yourself. | Model routing explained |
| 39. Tool use | When a model calls an outside function, such as a calculator, a web search or a code runner, and folds the result into its answer. | AI assistant tools explained |
| 40. Content provenance | Information attached to a file that says how it was made, including whether AI generated it. Watermarks and metadata labels are two forms. | AI watermarks and provenance |
Terms people confuse
A few pairs cause most of the confusion in everyday conversation.
- AI vs machine learning vs LLM. AI is the broad field, machine learning is the main method inside it, and an LLM is one kind of machine learning model that works with text.
- Training vs inference. Training teaches the model once, at the provider; inference is every answer you request afterwards. Your conversation is inference, which is why a model does not learn from your chat unless a provider chooses to use it, and Ask Mio does not use chats or files to train models.
- Context window vs memory. The context window is what the model can see right now. Memory is information an assistant stores and brings back in later chats; see AI projects and memory.
- Fine-tuning vs prompting. Fine-tuning changes the model; prompting changes the instructions. For most business uses, better instructions and examples are the cheaper first step.
- RAG vs fine-tuning. RAG gives the model facts at answer time; fine-tuning changes how it behaves. If the goal is accurate answers about your documents, RAG is usually the right tool.
- Open-weight vs open source. Open-weight means the trained model can be downloaded; open source also implies code, and often data, under an open licence.
How these terms show up in Ask Mio
You will not need most of this vocabulary to use Ask Mio, because the product handles it for you. You never pick a model: Mio routes each request to a model suited for the task, fast ones for chat, coding models for code, long-context models for documents and image models for design. Usage is counted in points, not tokens: a normal chat reply is 1 point, a long or complex one 3, a coding answer 5 to 10 and a generated image 20. Each plan has a monthly amount and a 5-hour allowance. Research mode uses web search and cites sources, which is the practical answer to the knowledge cutoff, and projects with memory let you keep instructions and files together, which is how you get the effect of a system prompt without writing code. A tour of what the assistant can do is in Ask Mio features.
The free plan gives 600 points a month with Chat and Write modes and needs no card. Paid plans start at €5 a month for 2,000 points with Research, file uploads and memory. All prices are in EUR with 21% VAT added at checkout, from the live pricing page.
How to learn the rest
You do not need to memorise any of this. A practical approach: when a term appears that you do not know, ask the assistant to explain it in one sentence with a real example, then ask it to tell you what would be different if the term did not apply. Keep a short list of the ones that matter to your work. For the standards behind responsible use, the NIST AI Risk Management Framework is a clear, free reference, and for the legal side in Europe see the EU AI Act explained in plain language. For a broader overview of how language models work, Wikipedia’s article on large language models is a reasonable starting point, to be checked against current sources.
Frequently Asked Questions
What is the difference between AI and machine learning?
AI is the broad goal of software doing tasks that normally need human intelligence. Machine learning is the main method used today: instead of writing rules, developers show a system many examples and it learns patterns. Most AI products you meet are built with machine learning.
What is a token and why does it matter?
A token is a small chunk of text, often a short word or part of a word, that a model reads and writes. Many providers price usage in tokens and models have token limits. Ask Mio hides this and charges in points per answer instead.
Is a hallucination a lie?
No. A lie requires intent, and a model has none. A hallucination is a fluent but false output produced because the model generates likely text rather than looking up verified facts. The remedy is to ask for sources, check important claims and use web search for current facts.
Does the AI learn from my conversations?
That depends on the provider. Training happens separately, before you use a model, and some services use customer chats to improve their models while others do not. Ask Mio does not use chats or files to train models, and you can export or delete your data at any time.
What does it mean that I never choose a model in Ask Mio?
Mio routes each request to the model best suited for it, so quick questions use fast models, code goes to coding models and documents go to long-context ones. You describe what you want and get an answer; paid plans get the stronger models for the heavier jobs.
Where should a beginner start?
Start with prompt, context window, hallucination and token, which explain most everyday surprises. Then read about RAG and model routing to understand how assistants use your documents and choose models. Try each idea in a real chat; the terms stick quickly once you have seen them in action.
The Bottom Line
This AI glossary gives you the 40 terms that matter most, in plain words. If you remember four things, make them these: models predict likely text, so verify important facts; a prompt with context and examples beats a vague one; training and your chat are different things; and usage is measured in tokens or points depending on the service. Bookmark the page and return when a new word appears, then try Ask Mio free to see the ideas at work in your own questions.