AI Glossary

Key AI/ML terms explained simply

LLM (Large Language Model)

A large AI model trained on massive text data to understand and generate human language. Examples: GPT-4, Claude, Llama.

Transformer

The neural network architecture behind modern AI. Uses "attention" to understand context in text.

Token

The basic unit AI processes. Can be a word, part of a word, or character. ~1 token ≈ 4 characters.

Fine-tuning

Training a pre-trained model on specific data to improve it for a particular task.

Prompt

The input you give to an AI model. Quality of output depends on quality of prompt.

Embedding

Converting text into numbers that capture meaning. Used for search and similarity.

Temperature

Controls randomness. Higher = more creative, lower = more predictable.

Context Window

Maximum tokens AI can see at once. GPT-4o has 128K tokens.

Zero-shot Learning

AI performs a task without specific training, using general knowledge.

Few-shot Learning

Giving AI a few examples in the prompt to understand the task.

RAG (Retrieval Augmented Generation)

Feeding relevant documents to AI to generate more accurate answers.

Hallucination

When AI generates confident but false information. A known limitation.

Inference

Running an AI model to generate output. Different from training.

Parameters

Internal values the model learns. More parameters usually = more capable.

Quantization

Reducing model size by using fewer bits. Makes it run faster on less RAM.

System Prompt

Instructions that set AI's persona and behavior for all interactions.