The one mental model
If you remember only one idea, make it this one: an AI language model is a very good next-word predictor trained on an enormous amount of text. Think of it as autocomplete that has read the internet. You give it words; it predicts what plausibly comes next, piece by piece, until it has an answer.The seven words
Model — the brain
Model — the brain
The trained system that turns your input into a response. Claude and GPT are models. Different models differ in capability, speed, and cost. Aperium runs on Claude models. When someone says “the model,” they mean the brain — not the app, and not your prompt.
Prompt — what you ask
Prompt — what you ask
Whatever you type or ask. The quality of the prompt is the biggest lever on the quality of the answer: the same task, phrased vaguely versus specifically, produces very different results. There is a whole page on doing this well — see Prompting & getting good results.
Token — a chunk of text
Token — a chunk of text
The unit of text the model reads and writes — roughly three-quarters of a word. You will mostly hear “tokens” when people talk about length or cost.
Context window — what it can hold in mind at once
Context window — what it can hold in mind at once
How much the model can consider at one time: your prompt, any files you attach, and its own reply. Picture a desk of limited size — pile too much on and older things fall off the edge. That is why a very long document or a very long chat can start to “forget” how it began.
Hallucination — confident but false
Hallucination — confident but false
When the model states something false in a confident voice — a made-up number, a citation that does not exist. It happens because the model predicts plausible text and has no built-in sense of true versus false. It is not lying; lying needs intent, and it has no notion of truth, only plausibility.
Agent — talk plus do
Agent — talk plus do
A plain chatbot only talks. An agent can take actions: search a system, read a file, send a message, query data. Giving an agent actions that reach our systems is exactly what makes Aperium more than a generic chatbot.
Tool — one action an agent can take
Tool — one action an agent can take
A specific capability an agent uses to act, such as searching Salesforce or looking up a contract. Chatbot equals talk; agent equals talk plus tools. See How agents, tools & skills work.
The habit to carry forward
Where to go next
How LLMs actually work
Go one level deeper on the mental model: how the model learned, what it is doing word by word when it answers, and why the same mechanism makes it both fluent and confidently wrong.