AI Terms Explained in Plain English
I spent way too long nodding along to AI jargon I didn't actually understand, so here's the explainer I wish someone had sent me.
Okay so here's the thing. When I first started getting into AI tools, I felt like I was constantly half-understanding everything. People would throw around words like "LLM" or "tokens" or "fine-tuning" and I'd just kind of nod and then quietly google it later. It was exhausting.
So I'm writing the post I wish existed when I started. These are the terms that actually matter, explained the way I'd explain them to a friend over coffee. No textbook definitions. No unnecessary jargon inside the jargon explanation. Let's just talk about it.
LLM (Large Language Model)
This is basically what ChatGPT, Claude, Gemini, and most of the AI tools you're using are built on. An LLM is an AI that was trained on a massive amount of text, we're talking billions of documents, websites, books, code. And learned to predict what comes next in a sentence, over and over, until it got really good at generating language that sounds coherent and useful.
"Large" refers to the size of the model, which is measured in parameters (more on that in a sec). The bigger the model, generally the more capable it is. That's the basic idea. It's not magic, it's a very sophisticated pattern-matching system trained on an enormous amount of human writing.
Parameters
Okay this one sounds scarier than it is. Parameters are basically the internal settings inside an AI model, like billions of tiny knobs that got tuned during training to help the model produce good outputs. When people say a model has "70 billion parameters" they're talking about the size and complexity of those internal weights.
You don't need to think too hard about this one. More parameters generally means more capable, but also more expensive to run. That's basically all you need to know for everyday use.
Tokens
This one is actually important to understand because it affects how you use these tools and how much you pay for them. A token is roughly a word or part of a word. "cat" is one token. "unbelievable" might be two. "ChatGPT" is probably one or two tokens depending on the model.
Why does it matter? Because AI models have a "context window", a limit on how many tokens they can process at once. Think of it like short-term memory. If you paste in a 50-page document and ask a question, the model needs enough context window to hold all of that. And when companies charge for API usage, they charge per token. So knowing what tokens are helps you understand both limits and costs.
Context Window
This is the amount of text, measured in tokens, that an AI can "see" and work with at one time. Early models had tiny context windows, like 4,000 tokens. Now some models have context windows of 200,000 tokens or more, which means they can read something like a short novel all at once.
Practically speaking: if you're having a long conversation with an AI and it seems to forget what you said earlier, you might be running up against the context window limit. The older stuff just falls out of view. It's not the AI being dumb, it literally can't hold it all at once.
Prompt
A prompt is just what you type to the AI. Your question, your instruction, your request. "Write me a cover letter" is a prompt. "Summarize this email in three bullet points" is a prompt. "Act like a sarcastic assistant who hates Mondays" is also a prompt.
"Prompt engineering" is the slightly overhyped term for getting good at writing prompts that get you good results. I think of it less as engineering and more as just... learning how to communicate clearly with a system that takes instructions very literally. Give it more context, be specific about what you want, tell it what format you need. That's most of it.
Hallucination
Honestly this is one of the most important terms to know because it directly affects how you should use AI. A hallucination is when an AI confidently tells you something that's completely made up. Not uncertain, not hedged, just stated like a fact, and wrong.
I've seen AI tools invent citations that don't exist, make up statistics, and confidently give me incorrect information about real people. It's not lying in the intentional sense. The model doesn't know it's wrong. It's generating what sounds plausible based on patterns, and sometimes plausible-sounding is not accurate.
This is why I always say: for anything where the facts actually matter, verify it. AI is a starting point, not a source of truth.
Fine-tuning
Fine-tuning is when someone takes a base AI model and trains it further on a specific dataset to make it better at a particular task or style. So a company might take a general-purpose model and fine-tune it on their customer support data so it gets really good at handling their specific questions. Or someone might fine-tune a model on medical literature to make it more useful for clinical applications.
You might not need to do this yourself, but you'll see it referenced a lot when companies talk about how they customized their AI tools.
Inference
Inference is just the AI actually running and generating a response. When you type a message and hit send, the model performs inference. Training is how the AI learned everything. Inference is it using what it learned. Most of what you interact with in day-to-day AI tools is inference.
Multimodal
A multimodal AI can work with more than one type of input or output. Text and images, for example. Or text, images, and audio. ChatGPT with vision is multimodal, you can paste in a photo and ask questions about it. Models that can generate images from text prompts are multimodal. It just means the AI isn't limited to one format.
RAG (Retrieval-Augmented Generation)
This one sounds fancy but the concept is pretty intuitive. RAG is when an AI is connected to an external database or set of documents, and before answering your question, it searches that database to find relevant information, then uses that information to generate a response.
Why does this matter? Because it helps reduce hallucinations and lets the AI answer questions about things that weren't in its original training data, like your company's internal docs, or news that happened after its training cutoff. A lot of enterprise AI tools use RAG under the hood.
API
API stands for Application Programming Interface, which is a fancy way of saying "the thing that lets software talk to other software." When developers build apps that use AI, they're usually connecting to an AI company's API, basically sending requests and getting responses back programmatically, instead of using a chat interface.
If you're not a developer this probably doesn't affect your daily usage much. But if you ever try to build something with AI or use tools that let you connect AI to other apps, you'll hit this term constantly.
That's most of the core vocabulary. Honestly once these clicked for me, reading about AI tools got so much easier. I stopped feeling like I was catching every other word. You'll start seeing these terms everywhere and now you'll actually know what they mean, not just roughly, but actually.
Emily in AI
Emily in AI is a plain-English guide to AI tools, tips, and beginner guides. Every tool gets tested and written up without the hype or the jargon, so you can figure out what actually helps. New posts every week.
About Emily in AI →
