> learn about ai / brand new to this
Words you will hear
Thirty words, each explained in a sentence or two, so nobody at the table can lose you.
7 min read
Read this once and most of the jargon stops being jargon. Bookmark it for the rest.
The basics
AI. The broad word for machines doing things that used to need a person. Today it usually means a large language model in a chat window.
Model. The trained program itself. Claude Opus 5 is a model. ChatGPT is an app that talks to a model. See what a model is.
LLM. Large language model. A model trained on huge amounts of text that predicts what comes next. Every major chatbot is one.
Prompt. Whatever you type to the model. A question, an instruction, a pasted document. The instructions we write for your agent are one long prompt.
Chat. One conversation. The model remembers what was said in this chat and forgets it when you start a new one, unless the app saves a note.
Token. The small chunks a model reads and writes in. Roughly three quarters of a word. Prices and limits are counted in tokens, which is why you see the word on pricing pages.
Context window. How much a model can hold in mind at once, counted in tokens. A big window means it can read a whole book in one go. Fill it up and it starts losing the beginning, which is why long chats get forgetful.
Hallucination. When a model states something false with full confidence. It is not lying, it is guessing what a right answer looks like. See what AI actually is.
How models are made
Training. The process of building a model by showing it enormous amounts of text and adjusting it until its guesses improve. Takes months and costs a fortune. You never do this.
Parameters. The internal numbers a model adjusts during training. More parameters usually means more capable and more expensive to run. A big model today has hundreds of billions to a few trillion.
Knowledge cutoff. The date after which the model has read nothing. Ask about something newer and it will not know unless it can search.
Fine-tuning. Training an already trained model a little more on a narrow set of examples so it behaves a certain way. Most people never need it.
Open weights. A model whose trained numbers are published for anyone to download and run on their own computer. Meta's Llama, DeepSeek and Mistral do this. Claude and ChatGPT do not.
Reasoning, or thinking. A mode where the model works through a problem step by step before answering. Slower and better at hard things. Most current models can do it.
Multimodal. A model that handles more than text: images, audio, video. Most current models are.
Using it
System prompt, or instructions. Standing instructions that apply to a whole conversation. In Claude they live in a Project, in ChatGPT in a Project or custom instructions. Your agent's instruction block is one.
Tool. Something a model can use besides its own memory: web search, a calendar, an inbox, a brokerage. Tools are how a chatbot becomes useful. See what an agent is.
Connector. A tool you plug into your AI app, usually with a sign in. Gmail is a connector. Robinhood is a connector.
MCP. Model Context Protocol. The open standard that lets any AI app plug into any tool the same way. Why one Robinhood connector works in Claude, ChatGPT, Grok and Cursor alike.
Agent. An AI with a job, a set of tools and a set of rules, that can take steps on its own to finish the job. The thing this site builds.
Scheduled task, automation, routine. Different apps' names for the same thing: an agent that runs on a timer without you opening a chat.
API. The way programs talk to a model directly, without a chat window. Developers use it. You do not need to, but "API key" is how a program proves it is allowed to.
RAG. Retrieval augmented generation. A fancy name for "look the relevant documents up first, then answer". How chatbots answer questions about your own files.
Prompt injection. A trick where text the agent reads (an email, a web page) contains instructions aimed at the agent. The reason a good agent has hard rules it will not break for anyone. See staying safe.
Companies and products
OpenAI. Makes ChatGPT and the GPT models. Anthropic. Makes Claude. Google DeepMind. Makes Gemini. xAI. Makes Grok. Meta. Makes Llama, open weights. See the companies.
Cursor, Claude Code. AI tools built around writing software. Powerful, and they run agents too, but they assume you are comfortable in a code editor or a terminal.
Copilot. Microsoft's name for AI inside Windows and Office, powered mostly by OpenAI models.
Money words, briefly
Free tier. What you get without paying. Usually a smaller model and lower limits.
Usage limits. How many messages or how much thinking you get per day or week on a plan. Every company has them and changes them often.
Credits. How some services meter model use. A dollar of model time turned into a number.
the one thing to do next
The fastest way to make this stick is to do one real thing with it today. Say who you are and we hand you one, prompt included.
Pick who you are