Michael Limberger

Glossary

The words as they show up

Terms used in this session, explained in the same coffee-chat as the rest.

API

Application Programming Interface. A way for programs to talk to each other. Ollama provides an API that lets your scripts send requests and receive responses. Think of a waiter taking your order and bringing back food. You do not need to know how the kitchen works.

Base64

An encoding method that converts binary data (like images) into text characters. Lets you embed files in JSON and other text formats. Named because it uses 64 different characters: A-Z, a-z, 0-9, plus, slash.

Binary data

Raw computer data: ones and zeros. Images, audio, and other files are stored as binary. Unlike text, you cannot read binary data directly.

curl

A command-line tool for making HTTP requests. We use it to send data to Ollama's API. The name comes from "Client URL."

Endpoint

A specific URL that provides a particular function. Ollama's /api/generate endpoint generates AI responses. Different endpoints do different things.

HTTP

Hypertext Transfer Protocol. The protocol web browsers use to communicate with servers. Also used by APIs. When you send a request to Ollama, you are using HTTP.

JSON

JavaScript Object Notation. A text format for structured data using key-value pairs. Looks like {"key": "value", "number": 42}. Used by most web APIs including Ollama. Human-readable and machine-parseable.

jq

A command-line tool for parsing and manipulating JSON. The name is short for "JSON query." Essential for working with API responses.

LLM

Large Language Model. An AI trained on massive amounts of text to understand and generate language. GPT, Claude, and Llama are examples. "Large" refers to the billions of parameters (adjustable values) in the model.

Model

A specific trained AI. Different models have different capabilities, sizes, and specializations. A vision model can see images. A code model is good at programming. Models are like different employees with different skills.

Ollama

An application that runs AI models locally on your computer instead of sending data to cloud services. It provides an API that other programs can use to access the AI.

Parsing

Extracting useful information from text or data. When we parse the vision model's response, we are pulling out the YES and NO values we need.

Prompt

The text instruction or question you send to an AI model. Good prompts lead to good responses. Prompt engineering is the skill of writing effective prompts.

Regex

Regular expression. A pattern-matching language for text. Lets you search for complex patterns like "a word followed by a number" instead of exact strings. Supported by most programming languages.

Structured output

AI responses formatted in a predictable way (like KEY: VALUE) that can be easily parsed by code. The opposite of free-form text.

Temperature

A setting that controls randomness in AI output. Low temperature (0.1) gives consistent, predictable responses. High temperature (1.0) gives more varied, creative responses.

Token

The basic unit AI models use to process text. Roughly corresponds to words or word parts. "Hello" is one token. "Internationalization" might be several. Models have limits on total tokens.

Vision model

An AI model that can process both text and images. Regular language models only understand text. Vision models can "see" and describe pictures.

Where to go next

Ollama: ollama.com. API docs: github.com/ollama/ollama docs/api.md. Models: ollama.com/library.

Regex tester: regex101.com.

More AI sessions on this site: AI hub. (Techalicious forum and meetup are historical only; those sites are gone.)