The jargon, ticked off.

Every AI term you've nodded along to in a meeting, explained properly in a sentence or two. No maths degree required.

32 terms

Agent: A model given tools and a goal, allowed to loop: plan, act, observe, repeat, until the job is done.
Aspect ratio: The shape of the output frame. Set it deliberately — models compose very differently at 16:9 versus 1:1.
Chain of thought: Asking a model to reason step by step before answering. Slower, but much better on maths, logic and multi-step tasks.
Context window: How much text a model can hold in mind at once (prompt + reply). Go over it and the earliest parts drop off the edge.
Diffusion model: An image or video model that starts from noise and repeatedly denoises it into a picture guided by your prompt.
Distillation: Training a smaller, cheaper model to imitate a bigger one. Most of the quality, a fraction of the cost.
Embedding: A list of numbers representing meaning. Similar ideas land near each other, which is what makes semantic search work.
Few-shot prompting: Showing the model two or three worked examples so it copies the pattern instead of guessing at it.
Fine-tuning: Further training a base model on your own examples so it adopts a style or task without being told every time.
Generative art: Art produced by a system you design — code, rules or models — where the output is partly outside your direct control.
Guardrails: Checks around a model — input filtering, output validation, refusal rules — that stop bad responses reaching users.
Hallucination: When a model states something confidently wrong. It isn't lying — it's predicting plausible text with no fact-checker attached.
Inference: The act of actually running a trained model to get an answer. Training builds the brain, inference asks it a question.
Inpainting: Regenerating just a masked part of an image while leaving the rest untouched. Ideal for fixing hands and swapping details.
Keyframe: A anchor frame in a video generation that defines where a shot starts or ends; the model fills the motion between them.
Latent space: The compressed map of concepts a model has learned. Generation is really navigation through that space.
MCP: Model Context Protocol — a shared standard for connecting models to external tools and data sources.
Multimodal: A model that handles more than text — images, audio or video in, and sometimes out.
Negative prompt: A list of things you explicitly don't want in the output — common in image generation to kill artefacts and clichés.
Prompt engineering: Writing instructions that reliably get the output you want: clear role, clear task, clear format, clear constraints.
RAG: Retrieval-Augmented Generation: fetch relevant documents first, then ask the model to answer using only those. Grounds answers in your own data.
Rate limit: A cap on requests or tokens per minute. Hit it and you get a 429 — back off and retry rather than hammering.
Seed: The number that determines the random starting point. Same seed plus same prompt equals a reproducible image.
Streaming: Sending the reply token by token as it's generated, so the interface feels instant instead of frozen.
Structured output: Forcing the model to answer in a fixed shape, usually JSON matching a schema, so your code can parse it safely.
Style reference: An image fed alongside your prompt so the model borrows its look — palette, texture, lighting — for new subjects.
System prompt: The standing instructions a model reads before your message — persona, rules and boundaries for the whole conversation.
Temperature: A dial for randomness. Low means predictable and repetitive, high means creative and occasionally unhinged.
Token: The chunks a model reads and writes — roughly 4 characters of English. Pricing and limits are counted in tokens, not words.
Tool calling: Letting a model invoke your functions or APIs with structured arguments instead of just describing what should happen.
Upscaling: Increasing an image's resolution after generation, adding plausible detail rather than simply stretching pixels.
Vibe coding: Building software by describing what you want in natural language and iterating on what the AI produces, rather than typing every line yourself.