The AI terms, in plain English.

Core AI terms

AI fluency

The practical ability to use, evaluate, and direct AI tools effectively for real tasks.

LLM

Large language model. A model trained on large amounts of text that can generate text, answer questions, summarize, translate, and more. Examples: GPT-5, Claude, Gemini.

Generative AI

AI systems that produce new content (text, images, audio, video, code) instead of only classifying or scoring existing data.

Prompt

The instruction or question you give an AI model. Better prompts produce better outputs.

Prompt engineering

The discipline of writing, structuring, and refining prompts so AI models reliably produce useful output.

Context window

How much text a model can read at once. Larger windows let the model consider longer documents and conversations.

Token

The basic unit of text a model reads and writes. Roughly equivalent to a word fragment.

Temperature

A setting that controls how varied an AI's output is. Lower is more predictable; higher is more creative.

System prompt

A high-priority instruction set at the start of a conversation that shapes how the AI behaves.

Models and tools

ChatGPT

OpenAI's consumer AI assistant. Free and paid tiers, with paid tiers offering more capable models.

Claude

Anthropic's AI assistant. Known for long context windows and careful reasoning.

Gemini

Google's AI assistant, integrated across Google products.

Perplexity

An AI-powered answer engine that grounds responses in live web search results.

Copilot

Microsoft's AI assistant family, including GitHub Copilot for coding.

Cursor

An AI-first code editor that integrates LLMs deeply into the writing-and-editing flow.

Image models

Midjourney, DALL·E, Stable Diffusion, Imagen, Flux — each with different strengths in style, photorealism, and editability.

Failure modes

Hallucination

When an AI confidently produces wrong information that sounds correct. Detecting hallucinations is a core Iro AI skill.

Confabulation

Same idea as hallucination — invented details that look reasonable but are not grounded in source data.

Prompt injection

A security issue where untrusted content embedded in inputs tries to override the AI's instructions.

Jailbreak

A prompt designed to bypass an AI's safety constraints. Iro AI teaches awareness, not exploitation.

Workflows

Agent

An AI system that pursues a goal across multiple steps, often using tools, memory, or external data.

Workflow automation

Stringing AI actions together to automate a repeatable process.

RAG

Retrieval-augmented generation. A pattern where the AI is given relevant documents at runtime so its answers are grounded in current data.

Vibe coding

Casual term for AI-assisted, prompt-led software building. The skill is judgment: knowing what to ask, what to verify, and what to throw away. See /vibe-coding-course.

Multi-agent system

Multiple AI agents working together, each handling part of a larger task.

Concepts and techniques

Fine-tuning

Training an existing model further on your own examples so it specializes in a task, domain, or style.

Embedding

A numerical representation of text or images that lets AI measure how similar two pieces of meaning are.

Transformer

The neural-network architecture behind modern LLMs, which weighs how words relate to each other using "attention." Introduced in the 2017 paper "Attention Is All You Need."

Multimodal

An AI model that can handle more than one type of input or output — text, images, audio, and video.

Zero-shot prompting

Asking a model to do a task with no examples, relying only on its training.

Few-shot prompting

Giving a model a handful of examples in the prompt to steer its output format and quality.

Chain-of-thought

Prompting a model to reason step by step, which improves accuracy on complex problems.

Reasoning model

An LLM tuned to "think" through problems in steps before answering — useful for math, coding, and logic.

RLHF

Reinforcement learning from human feedback. Training that aligns a model's output with human preferences.

Parameters

The internal values a model learns during training. More parameters can mean more capability and more cost.

Inference

Running a trained model to generate an answer, as opposed to training the model.

Grounding

Tying an AI's answer to verifiable source data — for example via RAG or web search — to reduce hallucination.

Vector database

A database that stores embeddings so AI can retrieve the most relevant content by meaning rather than keywords.

MCP

Model Context Protocol. An open standard for connecting AI assistants to external tools and data sources.

Guardrails

Rules and filters that constrain what an AI can say or do, for safety and reliability.

AGI

Artificial general intelligence. A hypothetical AI that matches human ability across most tasks — not yet achieved as of 2026.

Iro AI terms

Prompt Lab

Iro's active prompt-practice feature. You write prompts and get AI-generated feedback on quality and effectiveness.

Live duels

ELO-ranked 5-question speed rounds against bot opponents in cyberpunk arenas with battle music and podium reveals.

Track

One of four high-level groupings in Iro: Tool Mastery, Creative & Coding, Work & Career, Core Skills.

Path

A single learning path within a track. Iro has 18 paths.

Rank tier

The 6-step progression system: Bronze, Silver, Gold, Platinum, Diamond, Iridescent.

Streak

Consecutive-day usage counter that supports habit formation.

XP

Experience points earned by completing lessons, exercises, and duels.

AI IQ test

Iro's free 10-question quiz at /quiz that ranks users Bronze-to-Iridescent and recommends a starting path.