Cortexa AI Glossary
AI, in plain words.
Short, sourced answers to the questions people ask about AI, from the Cortexa Learn app by Cortexa Consulting. Each page says what it was checked against, and when.
A to Z
A
- Agentic AIWhat is agentic AI?The word is in every product launch. What it promises, and what it leaves out.
- AGIWhat is AGI?Three letters in every headline, and no agreed definition. How to read them.
- AIWhat is AI?The word is everywhere. What it means, in plain terms.
- AI agentWhat is an AI agent?A chatbot gives you an answer. An agent takes steps toward a goal, with tools. What changes when it does.
- AI alignmentWhat is AI alignment?A chatbot's polite tone and its refusals were decided by people. How those choices get built in.
- AI answers in search (AI Overviews)What are AI answers in search (AI Overviews)?That answer above the search results was written seconds ago. What it's built from, and when to click past it.
- AI benchmarkWhat is an AI benchmark?Every model launch comes with a chart of test scores. What those tests are, and what they leave out.
- AI biasWhat is AI bias?An image tool draws nearly every judge the same way. Where that pattern comes from, and how people fix it.
- AI coding assistantWhat is an AI coding assistant?Your phone guesses the next word. Programmers have a version that can now change a whole project.
- AI governanceWhat is AI governance?Your workplace has a rule about who can approve an expense. This is that rule, for the artificial intelligence tools you use at work.
- AI modelWhat is an AI model?The word in every announcement, and the one thing it means.
- AI safetyWhat is AI safety?Before a new car reaches the lot, someone crashes one on purpose. The same kind of care goes into artificial intelligence.
- AI slopWhat is AI slop?Junk made in bulk has a name now. Who makes it, and how to tell it from good work.
- AI watermarks and content credentialsWhat are AI watermarks and content credentials?Some photos now carry a small label you can tap. What it tells you, and what it can't.
- AI, machine learning, deep learning: what's the difference?Three words on one box, and they fit inside each other.
- Attention in AIWhat is attention in AI?How a chatbot works out what "it" means, and why long documents cost more.
C
- Can AI learn from AI feedback (RLAIF, Constitutional AI, RLCD)?Some chatbots are trained partly by another model checking answers against written rules.
- Can we see inside an AI (explainability)?Even the builders can't always say why an answer came out. Researchers are starting to look inside.
- ChatGPTWhat is ChatGPT, and what is a chatbot?The window that says hi, and what's typing back.
- Computer-use agentWhat is a computer-use agent?Some assistants now click and type for you, one screenshot at a time. How, and how to keep it on a short leash.
- Context engineeringWhat is context engineering?A chatbot can only use what it can see. Choosing what goes in front of it is a skill with a name.
- Context windowWhat is a context window?Why a long chat forgets what you said at the start, and what to do about it.
D
- Data poisoningWhat is data poisoning?How a few planted examples can change what a model learns, and how teams catch them.
- Deep researchWhat is deep research?A button that reads for several minutes and hands back a long report. What happens while you wait.
- DeepfakeWhat is a deepfake?A famous face selling a cream they've never heard of. What's going on, in plain terms.
- DistillationWhat is distillation?Why a chatbot's "mini" version can be quick and still pretty capable.
- Do AI detectors work?A checker says an essay is 87 percent machine-written. What that number can and can't tell you.
- Do I have to label AI-made content?A small tag under a video says a machine helped make it. Who put it there, and when one is required.
- Does AI remember me?The chatbot remembered something from weeks ago. Here's where that memory lives.
- Does AI understand what it says?It says it understands. What's going on underneath?
E
- EmbeddingsWhat are embeddings?Your phone finds beach photos you never labeled. The map of numbers behind it.
- EU AI ActWhat is the EU AI Act?Europe wrote a broad law for artificial intelligence, and it reaches well past Europe.
- EvalsWhat are evals?Your work app updated overnight. Better or worse? Evals are how teams find out.
F
- Fine-tune, retrieve, or prompt: which one?Three ways to make a chatbot better at your work, and how to pick the lightest one that does the job.
- Fine-tuningWhat is fine-tuning?When an app says it was "trained for lawyers," this is often what happened.
- Foundation modelWhat is a foundation model?Many different apps can share the same model underneath. Here's what that means for you.
- Frontier modelWhat is a frontier model?The phrase shows up in news and in laws. It names a moment more than a machine.
G
- Generative AIWhat is generative AI?The button that writes the draft, and what it's doing underneath.
- GPUWhat is a GPU, and why does AI need so many?The chip that draws your video games turned out to suit artificial intelligence (AI) too.
- GuardrailsWhat are guardrails?If a chatbot ever told you "I can't help with that," you've met one. What guardrails are, and why there's never just one.
H
- How do agents talk to each other (A2A)?Agents from different companies are starting to pass work to each other. The shared rules that make it possible.
- How do AI image generators work?A picture nobody painted, made from static. How the trick works.
- How do I check an AI answer?One answer you're about to act on, and a minute to check it.
- How do I read an AI headline?One headline says it aced an exam. The next says it can't count letters. Four questions sort them out.
- How do I spot an AI voice scam?A cloned voice can sound perfect. The story it tells can't hide.
- How do reasoning models learn (RLVR)?Remember the answers in the back of a math workbook? Some chatbots learned to think that way.
- How does AI copy a voice?The voice reading your audiobook may never have recorded those lines. How a voice gets learned.
- How does AI make video?A dog surfs a perfect wave. Then the board melts. That melt is a clue.
- How does AI write an answer, one word at a time?Why the answer types itself out, and what that tells you about trusting it.
- How does my phone recognize my face?Your phone doesn't keep a photo of you. It keeps a set of numbers, and checks a fresh set each time you look.
- How is a model like ChatGPT trained?The tidy paragraphs, the friendly opener. Every habit a chatbot has was trained in its own stage.
- How much energy does AI use?Real numbers, both ways: the big total, the small single question, and what's changing.
I
- InferenceWhat is inference, and how is it different from training?You tap Translate, and a second later it's in English. That second has a name.
- Is a bigger model always better?"Biggest ever" makes a good headline. Whether you need it is a different question.
- Is AI dangerous?A straight answer: what the real risks are, and the habits that handle most of them.
- Is it safe to paste work information into AI?A chatbot could tidy your report in a minute. Which account you use decides whose rules apply.
- Is training AI on copyrighted work allowed?Authors and news publishers are taking artificial intelligence companies to court. The question underneath, and where it stands.
- Is what I type used to train AI?There's a switch in your chatbot's settings. What it decides, and where to find it.
J
K
L
M
- Machine learningWhat is machine learning?Your bank texts about a purchase seconds after you make it. Who decided to ask?
- MCPWhat is MCP?The Connect buttons in your chatbot's settings share one plug. What it is, and what stays your choice.
- Mixture of expertsWhat is a mixture of experts?How a model can be enormous and still answer quickly.
- Multi-agent systemWhat is a multi-agent system?A kitchen splits the dinner rush among several cooks. Some artificial intelligence (AI) tools now split big jobs the same way.
- Multimodal AIWhat is multimodal AI?Point your phone at a menu and ask. What's happening when a chatbot reads a picture.
N
O
P
- ParametersWhat are parameters?Headlines count them in billions. What a parameter is, and what the number does and doesn't tell you.
- PretrainingWhat is pretraining?The first and biggest stage of training, and where a chatbot's knowledge and its cutoff both come from.
- PromptWhat is a prompt, and what is prompt engineering?What you type shapes what you get. Asking well is most of the skill.
- Prompt cachingWhat is prompt caching?Why the same long instructions can cost less the second time you send them.
- Prompt injectionWhat is prompt injection?A phishing email fools people. This trick is written for the assistant reading your email.
R
- RAGWhat is RAG?Those small numbered links under a chatbot's answer are a clue.
- Reasoning modelWhat is a reasoning model?That pause before the answer is the model working. When it's worth the wait.
- Red teamingWhat is red teaming?The people paid to make a model misbehave, so you don't meet the problem first.
- RLHFWhat is RLHF?Two answers side by side, and a question: which do you prefer? That choice trains chatbots.
S
- Small modelWhat is a small model, and can AI run on my phone?Turn on airplane mode and type a message. Something still suggests the next word.
- Synthetic dataWhat is synthetic data?Made-up data, made on purpose. Why it helps, and why it still gets checked.
- System promptWhat is a system prompt?The instructions a chatbot gets before you type a word, and how to write your own.
T
- Three ways machines learnWhat are the three ways machines learn?Learning from answers, from patterns, or from a score. How machines learn, and why it matters which.
- TokenWhat is a token?The word on every chatbot price list and limit notice, in plain terms.
- Tool use (function calling)What is tool use (function calling)?A chatbot can't see the sky, yet it knows this weekend's forecast. Here's who checked.
- TrainingWhat is training?"Trained on billions of examples" sounds grand. Underneath is a very patient loop.
- Training dataWhat is training data, and where does it come from?Authors and artists are asking whether their work trained these models. What did the models read?
- TransformerWhat is a transformer?The design inside nearly every chatbot, and how it tells a riverbank from a bank account.
V
W
- What does "human in the loop" mean?Your bank texts "Was this you?" before a big payment. That text has a name.
- What does GPT stand for?Three letters you've probably said out loud, and what each one tells you.
- What happens when AI trains on AI (model collapse)?A copy of a copy loses its detail. What researchers found when models learn from models.
- What's the difference between open and closed models?"Open" is on a lot of models. It can mean the cake, or the recipe.
- Where is AI in my everyday life?Captions, a translated menu, a voicemail you read. That may be artificial intelligence at work three times tonight.
- Which AI tool should I use?Everyone seems to have a favorite chatbot. A ten-minute test tells you which one suits your work.
- Who owns what AI makes?A bakery logo made in five minutes with an image tool. Can anyone else copy it?
- Why do I get a different answer each time?Same question, two different answers. Nothing's broken, and the difference is useful.
- Why does AI agree with me so much?You ask "Are you sure?" and it changes its answer. Why that happens.
- Why does AI make things up?A confident answer can still be a guess. Why chatbots make things up.
- Why does my phone's photo look better than what I saw?Your phone didn't take one picture of that sunset. It took several.
- Will AI take my job?What the forecasts say, in both directions, and what to do this week.
By category
The basics
- What is AI?The word is everywhere. What it means, in plain terms.
- What is machine learning?Your bank texts about a purchase seconds after you make it. Who decided to ask?
- AI, machine learning, deep learning: what's the difference?Three words on one box, and they fit inside each other.
- What is an AI model?The word in every announcement, and the one thing it means.
- What is generative AI?The button that writes the draft, and what it's doing underneath.
- What is a large language model (LLM)?The engine inside today's chatbots, in three plain words.
- What is ChatGPT, and what is a chatbot?The window that says hi, and what's typing back.
- Does AI understand what it says?It says it understands. What's going on underneath?
- What is a neural network?Layers of simple parts that learn by adjusting their connections. How the idea works, and how loose the brain comparison is.
- What are parameters?Headlines count them in billions. What a parameter is, and what the number does and doesn't tell you.
- What does GPT stand for?Three letters you've probably said out loud, and what each one tells you.
- What is AGI?Three letters in every headline, and no agreed definition. How to read them.
- What is a foundation model?Many different apps can share the same model underneath. Here's what that means for you.
How it learns
- What is training?"Trained on billions of examples" sounds grand. Underneath is a very patient loop.
- What is training data, and where does it come from?Authors and artists are asking whether their work trained these models. What did the models read?
- What is inference, and how is it different from training?You tap Translate, and a second later it's in English. That second has a name.
- How is a model like ChatGPT trained?The tidy paragraphs, the friendly opener. Every habit a chatbot has was trained in its own stage.
- What is RLHF?Two answers side by side, and a question: which do you prefer? That choice trains chatbots.
- What are the three ways machines learn?Learning from answers, from patterns, or from a score. How machines learn, and why it matters which.
- What is pretraining?The first and biggest stage of training, and where a chatbot's knowledge and its cutoff both come from.
- What is fine-tuning?When an app says it was "trained for lawyers," this is often what happened.
- What is a transformer?The design inside nearly every chatbot, and how it tells a riverbank from a bank account.
- What is synthetic data?Made-up data, made on purpose. Why it helps, and why it still gets checked.
- What are embeddings?Your phone finds beach photos you never labeled. The map of numbers behind it.
- What is an AI benchmark?Every model launch comes with a chart of test scores. What those tests are, and what they leave out.
- Is a bigger model always better?"Biggest ever" makes a good headline. Whether you need it is a different question.
- What is a GPU, and why does AI need so many?The chip that draws your video games turned out to suit artificial intelligence (AI) too.
- What is attention in AI?How a chatbot works out what "it" means, and why long documents cost more.
- What is distillation?Why a chatbot's "mini" version can be quick and still pretty capable.
- What happens when AI trains on AI (model collapse)?A copy of a copy loses its detail. What researchers found when models learn from models.
- What is overfitting?Great on the examples it learned from, weak on yours. Why that happens.
- How do reasoning models learn (RLVR)?Remember the answers in the back of a math workbook? Some chatbots learned to think that way.
- Can AI learn from AI feedback (RLAIF, Constitutional AI, RLCD)?Some chatbots are trained partly by another model checking answers against written rules.
How it answers
- What is a token?The word on every chatbot price list and limit notice, in plain terms.
- How does AI write an answer, one word at a time?Why the answer types itself out, and what that tells you about trusting it.
- Why do I get a different answer each time?Same question, two different answers. Nothing's broken, and the difference is useful.
- What is a context window?Why a long chat forgets what you said at the start, and what to do about it.
- Does AI remember me?The chatbot remembered something from weeks ago. Here's where that memory lives.
- What is a knowledge cutoff?Why a chatbot can sound completely sure and still be a year out of date.
- Why does AI make things up?A confident answer can still be a guess. Why chatbots make things up.
- What is a system prompt?The instructions a chatbot gets before you type a word, and how to write your own.
- What is a reasoning model?That pause before the answer is the model working. When it's worth the wait.
- What is multimodal AI?Point your phone at a menu and ask. What's happening when a chatbot reads a picture.
- What is a small model, and can AI run on my phone?Turn on airplane mode and type a message. Something still suggests the next word.
- What's the difference between open and closed models?"Open" is on a lot of models. It can mean the cake, or the recipe.
- What is prompt caching?Why the same long instructions can cost less the second time you send them.
- What is a mixture of experts?How a model can be enormous and still answer quickly.
Asking well
- What is a prompt, and what is prompt engineering?What you type shapes what you get. Asking well is most of the skill.
- How do I check an AI answer?One answer you're about to act on, and a minute to check it.
- What is RAG?Those small numbered links under a chatbot's answer are a clue.
- What is a vector database?You typed "send this back" and found "Returns and refunds." The filing cabinet behind that search.
- Fine-tune, retrieve, or prompt: which one?Three ways to make a chatbot better at your work, and how to pick the lightest one that does the job.
- What is context engineering?A chatbot can only use what it can see. Choosing what goes in front of it is a skill with a name.
- What are AI answers in search (AI Overviews)?That answer above the search results was written seconds ago. What it's built from, and when to click past it.
- What is deep research?A button that reads for several minutes and hands back a long report. What happens while you wait.
- What are evals?Your work app updated overnight. Better or worse? Evals are how teams find out.
Agents and connections
- What is an AI agent?A chatbot gives you an answer. An agent takes steps toward a goal, with tools. What changes when it does.
- What is MCP?The Connect buttons in your chatbot's settings share one plug. What it is, and what stays your choice.
- What is vibe coding?Someone you know built an app without writing a line of code. What they did, and where it can go wrong.
- What is an AI coding assistant?Your phone guesses the next word. Programmers have a version that can now change a whole project.
- What is agentic AI?The word is in every product launch. What it promises, and what it leaves out.
- What is tool use (function calling)?A chatbot can't see the sky, yet it knows this weekend's forecast. Here's who checked.
- What is a computer-use agent?Some assistants now click and type for you, one screenshot at a time. How, and how to keep it on a short leash.
- What does "human in the loop" mean?Your bank texts "Was this you?" before a big payment. That text has a name.
- What is a multi-agent system?A kitchen splits the dinner rush among several cooks. Some artificial intelligence (AI) tools now split big jobs the same way.
- How do agents talk to each other (A2A)?Agents from different companies are starting to pass work to each other. The shared rules that make it possible.
Making pictures, voices, and video
- How do AI image generators work?A picture nobody painted, made from static. How the trick works.
- How does AI copy a voice?The voice reading your audiobook may never have recorded those lines. How a voice gets learned.
- How does AI make video?A dog surfs a perfect wave. Then the board melts. That melt is a clue.
- What is AI slop?Junk made in bulk has a name now. Who makes it, and how to tell it from good work.
Trust, fakes, and safety
- What is a deepfake?A famous face selling a cream they've never heard of. What's going on, in plain terms.
- How do I spot an AI voice scam?A cloned voice can sound perfect. The story it tells can't hide.
- Why does AI agree with me so much?You ask "Are you sure?" and it changes its answer. Why that happens.
- What are AI watermarks and content credentials?Some photos now carry a small label you can tap. What it tells you, and what it can't.
- Do AI detectors work?A checker says an essay is 87 percent machine-written. What that number can and can't tell you.
- What is AI bias?An image tool draws nearly every judge the same way. Where that pattern comes from, and how people fix it.
- What is prompt injection?A phishing email fools people. This trick is written for the assistant reading your email.
- What are guardrails?If a chatbot ever told you "I can't help with that," you've met one. What guardrails are, and why there's never just one.
- What is a jailbreak?"Chatbot goes rogue" makes a good headline. Read on, and there's usually a person doing the pushing.
- What is AI safety?Before a new car reaches the lot, someone crashes one on purpose. The same kind of care goes into artificial intelligence.
- What is AI alignment?A chatbot's polite tone and its refusals were decided by people. How those choices get built in.
- What is data poisoning?How a few planted examples can change what a model learns, and how teams catch them.
- What is red teaming?The people paid to make a model misbehave, so you don't meet the problem first.
- Can we see inside an AI (explainability)?Even the builders can't always say why an answer came out. Researchers are starting to look inside.
Your data, rights, and the rules
- Is what I type used to train AI?There's a switch in your chatbot's settings. What it decides, and where to find it.
- Is it safe to paste work information into AI?A chatbot could tidy your report in a minute. Which account you use decides whose rules apply.
- Who owns what AI makes?A bakery logo made in five minutes with an image tool. Can anyone else copy it?
- Is training AI on copyrighted work allowed?Authors and news publishers are taking artificial intelligence companies to court. The question underneath, and where it stands.
- Do I have to label AI-made content?A small tag under a video says a machine helped make it. Who put it there, and when one is required.
- What is the EU AI Act?Europe wrote a broad law for artificial intelligence, and it reaches well past Europe.
- What is AI governance?Your workplace has a rule about who can approve an expense. This is that rule, for the artificial intelligence tools you use at work.
AI and your life
- Where is AI in my everyday life?Captions, a translated menu, a voicemail you read. That may be artificial intelligence at work three times tonight.
- Will AI take my job?What the forecasts say, in both directions, and what to do this week.
- Is AI dangerous?A straight answer: what the real risks are, and the habits that handle most of them.
- How does my phone recognize my face?Your phone doesn't keep a photo of you. It keeps a set of numbers, and checks a fresh set each time you look.
- Which AI tool should I use?Everyone seems to have a favorite chatbot. A ten-minute test tells you which one suits your work.
- Why does my phone's photo look better than what I saw?Your phone didn't take one picture of that sunset. It took several.
- How much energy does AI use?Real numbers, both ways: the big total, the small single question, and what's changing.
- How do I read an AI headline?One headline says it aced an exam. The next says it can't count letters. Four questions sort them out.
- What is a frontier model?The phrase shows up in news and in laws. It names a moment more than a machine.