You've Heard the Words. Now Here's What They Actually Mean.
Artificial intelligence is everywhere — in your phone, your search engine, your workplace software, and seemingly every news headline. But the wave of new vocabulary that's come with it? That part hasn't been as easy to absorb.
TechCrunch recently published a comprehensive glossary breaking down the most important AI terms and phrases circulating right now. Whether you've been nodding along in meetings or just trying to keep up with the news, here's a plain-English rundown of what it all means.
Hallucinations
This is one of the most talked-about — and most misunderstood — terms in AI. When an AI model "hallucinates," it confidently produces information that is simply made up. It's not lying in the human sense; it's generating plausible-sounding text that has no basis in fact. This is why you should never rely on AI tools to look up medical dosages, legal statutes, or anything where accuracy is critical without verifying independently.
Large Language Models (LLMs)
This is the engine behind tools like ChatGPT, Claude, and Gemini. A large language model is a type of AI trained on enormous amounts of text data, learning to predict and generate language by recognizing patterns at a massive scale. When you type a question and get a fluent paragraph back, that's an LLM at work.
Prompts and Prompt Engineering
A "prompt" is simply the instruction or question you give an AI model. But "prompt engineering" has evolved into a genuine skill — the art of crafting inputs in a way that consistently gets you better, more useful outputs. Specificity, context, and tone all matter more than most people realize.
Tokens
AI models don't read text the way humans do — they process it in chunks called tokens, which roughly correspond to words or parts of words. When a model has a "context window" of 100,000 tokens, that's how much text it can "hold in mind" at once during a conversation.
RAG (Retrieval-Augmented Generation)
This is a technique that helps AI models give more accurate answers by pulling in real, up-to-date information from an external source before generating a response. Instead of relying purely on what it learned during training, a RAG-powered system can fetch current facts and weave them into its answer — reducing hallucinations in the process.
Fine-Tuning
When a general-purpose AI model is trained further on a specific dataset — say, medical records or legal documents — to make it more useful for a particular domain, that's fine-tuning. It's the difference between a model that can talk about medicine in general and one that's been sharpened to assist actual clinicians.
Inference
This is the moment an AI model actually does its job — when it takes your input and generates an output. Training is expensive and happens once (or periodically); inference is what happens millions of times a day as users interact with the model.
Why This Matters
As AI tools become embedded in everyday work, healthcare, education, and civic life, understanding the vocabulary isn't just for tech enthusiasts anymore. Knowing the difference between a hallucination and a verified fact, or understanding why an AI might produce a confident but wrong answer, is quickly becoming baseline digital literacy.
The full TechCrunch glossary covers many more terms and is worth bookmarking as a reference.
Source: TechCrunch — AI terms glossary, May 2026.
