The Essential AI Glossary

Most people:

  1. Struggle to actually define what AI actually is.
  1. Have been given overly complicated definitions about AI.

In this doc, you'll find definitions to every AI keyword that you've probably heard, but might not understand, that should help with the problems outlined above.

Below, you'll find definitions for:

  1. AI
  1. Machine Learning
  1. Deep Learning
  1. Models
  1. Parameters
  1. Training
  1. Neural networks
  1. Tokens
  1. Transformers
  1. LLMs
  1. Generative AI
  1. GPT
  1. RAG

Anything missing? Let me know.

AI

An approach to automation for tasks that are hard to write down the step by step instructions for (e.g. how to see, how to understand natural language, etc)

Machine Learning

An approach to AI where software learns to do tasks from examples rather than rules

Deep Learning

An approach to machine learning using neural networks

Model

Input/output functions (aka mathematical equations - numbers in -> numbers out) that represent some relationship between the inputs and outputs (e.g. your text request -> an appropriate response, image pixels -> how should my self driving car turn/accelerate)

Parameters

The numbers inside models that control what input/output relationship they represent

Training

The process of changing the parameters of a model to adjust what input/output relationship they represent

Neural Networks

Models that can learn to represent any input/output relationship 

Tokens

Pieces of words (not necessarily whole words) that text inputs to AI systems get broken into before processing (these tokens then get turned into numbers (vectors) so they can be processed by the mathematical AI models 

Transformers

The type of neural network powering state of the art AI models (also the T in GPT)

LLMs

Large Language Models - models with lots of parameters (hence the “large”) trained on examples of text (typically to predict the next word) which amazingly, seems to give them some form of intelligence!

Generative AI

AI that generates realistic looking data (e.g. realistic images, realistic text responses, realistic forecasts)

GPT

Generative Pre-trained Transformer (not General Purpose Transformer) transformer neural networks that have already seen lots of example data (have been pre-trained) and can generate similar looking data

Made with