AI Vocabulary

This list unfortunately begins with the most negative term. Not our goal, really, but some of us can’t stand lists in any other order than alphabetical. (You may call it OCD, we call it love for logic and structure.) Anyway, we hope this helps you navigate yourself in the realm of AI without feeling like you’ve lost the will to live.

AI Slop

Low-effort, low-quality content generated by AI that’s usually pumped out fast, without much editing or care. It might look fine at a glance, but it’s often shallow, repetitive, or just plain wrong. Think: the fast food of AI content. Easy to churn out, not always nourishing.

API (Application Programming Interface)

A way for software to talk to the model directly. Send a prompt, get a response - no need for a UI. Perfect for developers and tech-savvy tinkerers.

Attention

A core part of modern language models. It helps the model focus on the most relevant words when making predictions. (Like a bouncer for words - letting the important ones into the VIP section of the model’s brain.)

Context Window

The model’s short-term memory. It defines how many tokens it can hold in mind at once. If your prompt is too long, it might forget what you said at the beginning - just like the rest of us during long meetings.

Embedding

A way of turning words into numbers - specifically, points in a high-dimensional space. Words that mean similar things are placed close together, like king and queen, or taco and delicious.

Fine-tuning

A second round of training with more specific data. It’s like sending a generalist model to finishing school for medical writing, customer service, or pizza topping recommendations (yes, that too).

Generative AI

AI that doesn’t just analyse - it creates. Text, images, code, poems about breadsticks - you name it. All based on patterns it learned during training.

Hallucination

When the model makes things up. It’s not lying - it just doesn’t know it’s wrong. It might confidently tell you a made-up “fact” with the enthusiasm of a kid explaining quantum physics.

Inference

The moment the model turns your input into output - like solving a puzzle on the fly. It’s not learning here; it’s applying what it already knows to guess what should come next.

Latency

The time between asking the model a question and getting a reply. Lower latency = faster response. Think of it as the model’s reaction time.

Model Architecture

The blueprint of the model - how it’s built under the hood. Layers, neurons, attention: all the geeky stuff that determines how it understands and generates language or other media.

Parameters

The knobs and dials inside the model - billions of them, imagine that. These numbers were fine-tuned during training and shape how the model responds to anything you throw at it.

Prompt

The text you give the model to get a response - the better your prompt, the better the output - kind of like giving good directions to a very enthusiastic overachiever. A clear, well-structured prompt usually gets a better answer.

Temperature

A setting that controls how creative the model gets. Low temperature = safe, predictable answers. High temperature = more surprises, for better or worse. (Use with caution when emailing your boss.)

Tokens

Tiny chunks of text that a language model reads and understands. Most common words are a single token (like “cat” or “coffee”), but long or unusual words can be split into several (like hypersensitivity → hyper–sens–itivity). Think of tokens as the bite-sized pieces AI snacks on.

Training Data

The giant pile of text the model learned from. Books, websites, articles - you name it. From this, it picked up grammar, facts, tone, and just enough weird internet humor to keep things interesting.

We hope this helps you navigate yourself in the jungle of modern technology trends. If you ever need someone to build a solution for your process automation with all these shiny bits of tech weilded in a high-funcioning machine, let us know, we’re happy to help!

You can also subscribe here to get AI news and advice delivered straight in your inbox.