AI dictionary
AI concepts, straight to the point.
Fundamentals
- Agent
- An agent is an AI model placed in a loop with tools, deciding its own steps to complete a task: it reads files, runs commands, queries systems and checks the result.
- Context window
- The context window is how much text a model can consider at once, adding up instructions, conversation, files and its own answer. It is measured in tokens.
- Harness
- A harness is the software around a model that turns it into an agent: the loop that calls the model, the tools it can use, the context it receives and the permissions that limit it.
- Inference
- Inference is when an already trained model is used to generate an answer. Every request to a model is an inference, and that is what providers serve and charge for.
- Model
- A model is the trained system that takes text (and sometimes images or audio) and generates the answer. It is the engine behind any AI tool, but on its own it can't open files or use tools.
- Prompt
- A prompt is the instruction you send to a model. It includes the request and all the context that helps get the answer right: role, situation, format and examples.
- Skill
- A skill is a set of instructions kept in a file that the agent loads only when the task calls for it. It teaches the agent a procedure without repeating the same prompt.
- Token
- A token is the unit of text a model reads and writes. A word can be one or several tokens. Model pricing, speed and context limits are all measured in tokens.
Market
- Chinese and US models
- The leading AI models come from labs in the United States and China. US models lead in capability and are closed. Chinese models usually have open weights and much lower prices, with differences in regulation and content.
- Frontier model
- A frontier model is the most capable general-purpose model available at a given time, the one pushing the limit of what AI can do. It is a moving target: today's frontier is tomorrow's baseline.
- Open weights
- An open-weight model is one whose weights (the trained parameters) anyone can download and run. It is not the same as open source: the training data and code almost never come with it.
- Provider
- A provider is whoever serves you the model, usually through an API. It can be the lab that trained the model, a cloud reselling many models, or a company specialized in running open models.