AI Atlas: the printable guide
Foundations
What AI is, and how machines learn at all.
- Artificial Intelligence (AI): Computers doing things that normally need a human brain, like understanding words or recognising faces.
- Machine Learning (ML): Teaching a computer by showing it lots of examples instead of writing step-by-step instructions.
- Model: The "brain file" an AI produces after learning. You give it an input and it gives back an answer.
- Training Data: The examples an AI learns from.
- Training: The learning phase: the AI makes guesses, gets told how wrong it was, and adjusts itself a little. Repeated millions of times.
- Inference: Using an already-trained AI to get an answer.
- Parameters: The adjustable knobs inside an AI. Training sets their values.
- Neural Network: A web of tiny maths units, loosely inspired by brain cells, that pass signals to each other.
- Deep Learning: Machine learning with neural networks that have many layers.
- Overfitting: When an AI memorises its practice examples instead of learning the general idea, so it fails on new ones.
Language Models
How ChatGPT-style tools read and write.
- Generative AI: AI that creates new things: text, images, music, video or code.
- Large Language Model (LLM): An AI trained on huge amounts of text that writes by predicting the next word, over and over.
- Token: The bite-sized pieces of text an AI reads and writes: often a word, sometimes part of a word.
- Context Window: How much text an AI can "see" at once: its short-term memory.
- Temperature: A setting for how adventurous or predictable the AI's word choices are.
- Prompt: What you type to ask the AI to do something.
- Prompt Engineering: The skill of writing instructions that get good results from AI.
- Hallucination: When AI confidently says something that sounds right but is made up.
- Transformer: The design behind modern language AIs. The "T" in GPT.
- Attention: How an AI decides which other words matter most when understanding each word.
- Multimodal: AI that handles more than one kind of input or output, such as text, images, audio and video.
- Diffusion Model: The kind of AI that makes images by starting with static noise and gradually cleaning it into a picture.
Building with AI
How products connect AI to your data and tools.
- Embedding: Turning words or documents into a list of numbers so that similar meanings end up close together.
- Vector Database: A database that finds things by similar meaning instead of exact words.
- Retrieval-Augmented Generation (RAG): Letting the AI look things up in your documents before it answers.
- Fine-Tuning: Giving an already-trained AI extra training on specific examples to specialise it.
- AI Agent: An AI that doesn't just answer, but takes actions step by step to complete a goal.
- Tool Use: Letting an AI use other software, such as a calculator, search engine or calendar.
Risks & Society
Where AI goes wrong and how we keep it in check.
- AI Bias: When an AI treats some people or groups unfairly because of patterns in its data or design.
- Alignment: Making sure AI does what people actually intend, and does it safely.
- Deepfake: Fake but realistic video, audio or images made with AI.
- Responsible AI: Building and using AI in ways that are fair, safe, transparent and respectful of people.