AI Fluency Glossary | AIRBORNE Patriot Solutions

1. The Fluency Framework: Four Core Skills

Being "fluent" with AI means working with it in ways that are effective, efficient, ethical, and safe. A useful way to break that down is four competencies:

Skill	The question it answers	Sub-skills
Delegation	What should I hand to AI, and what should I keep?	Problem awareness: know your goal and the shape of the work before involving AI. Platform awareness: know what a given system can and can't do. Task delegation: split the work so humans and AI each do what they're best at.
Description	How do I tell the AI what I want?	Product: define the output (format, audience, style). Process: define the approach (e.g., step-by-step instructions). Performance: define the behavior (concise vs. detailed, challenging vs. supportive).
Discernment	Is what I got actually good?	Product: judge the output (accuracy, relevance, coherence, fit). Process: judge how it got there (logic gaps, skipped steps, bad reasoning). Performance: judge how it behaved (was its style useful to you?).
Diligence	Am I using it responsibly?	Creation: choose systems and interaction styles thoughtfully. Transparency: be honest about AI's role with anyone who needs to know. Deployment: verify and stand behind anything you use or share.

Note

Memory hook: Decide → Describe → Judge → Own.

2. Three Ways Humans and AI Work Together

Mode	Who's driving	Example
Automation	Human specifies the exact task; AI carries it out.	"Convert these 40 rows into a formatted table."
Augmentation	Human and AI think together, iterating back and forth.	Brainstorming a strategy and refining it over several turns.
Agency	Human sets up the AI's knowledge and behavior, then it acts independently, possibly with other people or systems.	An agent that triages incoming requests on your behalf within rules you defined.

3. How the Technology Works

Term	Plain-language meaning
Attention	The mechanism that lets a model weigh which earlier words matter most when interpreting the current one. It's how the model knows what "it" refers to in a sentence.
Bias	Systematic tilts in output that unfairly favor or disadvantage certain groups or viewpoints, usually inherited from training data.
Context window	How much the model can "see" at once: the conversation so far plus any files you shared. It has a hard limit that varies by model.
Embedding	A list of numbers representing a token's meaning. Similar meanings sit close together, like pins on a map where "coffee" is near "espresso" and far from "tractor."
Fine-tuning	Extra training after pre-training that teaches a model to follow instructions, be helpful, and avoid harmful output.
Generative AI	AI that creates new content (text, images, audio, code) rather than only sorting or labeling existing data.
Hallucination	A confident, plausible-sounding statement that is wrong. Includes made-up facts, figures, and citations.
Knowledge cutoff	The date after which a model has no built-in knowledge, because its training data stops there.
Large language model (LLM)	A generative model trained on huge amounts of text to understand and produce language. Most chat assistants are built on one.
Neural network	A computing system of interconnected nodes in layers that learns patterns from data. Loosely inspired by, but distinct from, the brain.
Next-token prediction	The core thing an LLM does: given everything so far, pick the most fitting next token, then repeat. A full reply is many predictions strung together.
Parameters	The internal numeric values that determine how a model processes and relates information. Modern LLMs have billions.
Pre-training	The first, largest training phase, where the model learns language patterns and general knowledge from vast text.
Reasoning (thinking) model	A model designed to work through a problem step by step before answering. Better at multi-step, logical, high-stakes tasks, but slower.
Retrieval-augmented generation (RAG)	Connecting a model to external knowledge sources so it can look things up, improving accuracy and reducing hallucinations.
Scaling laws	The observed pattern that models improve predictably as they get bigger and train on more data and compute. New abilities sometimes appear at certain scales without being explicitly programmed.
Temperature	A setting for randomness. Higher means more varied and creative output; lower means more predictable and focused output.
Token	The small chunk of text a model actually works with: roughly a word or part of a word (about ¾ of an English word on average). Tokens are also the unit context is measured in.
Transformer	The model architecture (introduced in 2017) behind modern LLMs. It processes text in parallel and tracks relationships between words across long passages using attention.

The pipeline in one line:

your words → tokens → embeddings → attention picks what matters → predict next token → repeat → response

4. Prompting Vocabulary

Term	Meaning
Chain-of-thought prompting	Asking the model to work through a problem in steps, so it breaks down complex tasks instead of jumping to an answer.
Few-shot (n-shot) prompting	Showing examples of the input→output pattern you want. n is the number of examples. Zero-shot means none.
Output constraints / formatting	Spelling out length, structure, format, or other traits of the response.
Prompt	Everything you give the model: instructions plus any documents you've shared.
Prompt engineering	Designing prompts that reliably produce the output you want. It's clear communication plus AI-specific techniques.
Role / persona	Telling the model what expertise, character, or voice to adopt, from broad ("act as a UX expert") to specific.
System prompt	The standing instructions behind a conversation that shape every reply.
Think-first approach	Explicitly asking the model to reason before giving its final answer.

5. Everyday Working Terms

Term	Meaning
Agent	An AI that runs multi-step work using tools (browsing, reading files, running commands), not just answering.
Artifact	A file or document the AI produces, as opposed to a chat reply.
Compact	Condensing a long conversation into a summary so you can keep going with less context used.
Context	Everything in the conversation so far. It's finite, so spend it wisely.
SOP	Standard operating procedure: a written recipe for a recurring task that an AI can follow consistently.
Subagent	A helper agent the main agent spins up with its own separate context to handle one piece of work.
TL;DR	"Too long; didn't read." The short version.

Previous: 50 marketing promptsNext: Working with an AI Assistant

Want this trained into your team’s habits?

APS delivers hands-on enablement so people use the AI tools they already have, confidently and safely. Tell us where your team is stuck.

Talk to us 888-878-2118

Educational and general, not legal advice. Follow your organization’s policies and applicable law.