Glossary
AI terms in plain language. With examples and posts to explore further.
25 terms
- API
An interface that lets programs talk to each other. One program sends a request and receives data or a result.
Example: The website asks an AI service for an answer and displays the result in the chat.
- Benchmark
A defined test that compares models under similar conditions. Its score describes performance on those particular tasks.
Example: Several models solve the same coding tasks. The test measures how many solutions work.
- Chunk / Chunking
A chunk is a small section of a longer document. Chunking splits text so relevant parts can be found individually.
Example: A handbook is split into sections about signing up, billing and cancellation.
Appears in
- Cloud
Computing or storage provided over the internet. You rent the resources you need instead of running every machine yourself.
Example: You rent a GPU for an AI job and release it when the job finishes.
Appears in
- Cloudflare
A provider for running websites and services. It can handle requests, deliver files and store data.
Example: Here, Cloudflare delivers the site and processes a newsletter signup.
Appears in
- Context
The information a model receives for a particular answer. It includes the question, earlier messages and sometimes relevant text excerpts.
Example: A billing question is sent together with the relevant passage from a handbook.
Appears in
- Context window
The limited amount of text a model can consider in one request. It is usually measured in tokens.
Example: If a whole handbook does not fit, only relevant sections are sent.
- Embedding
A list of numbers that represents features of a text. These numbers help find texts with similar meanings, even when they use different words.
Example: A question about holiday allowance can find a section about vacation policy.
Appears in
- GPU
A processor that can perform many similar calculations at once. This helps with the large numerical computations in AI models.
Example: A GPU computes many parts of a language model in parallel to generate an answer.
Appears in
- Inference
Using an already trained model on a new input. It may generate an answer or recognize something in an image.
Example: You ask a question and the language model computes an answer.
- LLM / Language model
An AI model that works with language and generates text. It uses learned patterns and the supplied context to compute its answer.
Example: A model writes a clear answer from a question and relevant handbook excerpts.
Appears in
- Next.js
A toolkit for building websites with React. It handles pages, navigation and connections between the browser and server.
Example: Here, Next.js builds the post page and the form for your questions.
Appears in
- On-prem
Software runs on hardware you own or manage yourself. You look after the machines, operations and maintenance.
Example: A company runs its AI model on its own GPU server.
Appears in
- OpenRouter
A service for accessing different AI models. It forwards a request to the selected model and returns its response.
Example: The website uses one interface to call a language model.
Appears in
- Prompt
The input or instruction given to an AI model. A concrete task and relevant information help the model produce a useful answer.
Example: Summarize these three paragraphs in two simple sentences.
- R2
A file storage service from Cloudflare. It holds files that a website can load when needed.
Example: When you start a video, the player loads the video file from R2.
Appears in
- RAG
The AI first finds relevant information, then uses it to answer. The retrieved passages are sent to the language model together with the question.
Example: A question about vacation days first retrieves the relevant section from a company handbook.
Appears in
- Reasoning
Working through a task step by step. A model can form intermediate steps, check options and revise its solution.
Example: For a math problem, it identifies the needed values and then checks the calculation.
- Remotion
A tool that renders programmed images and animations into a video. Elements can appear and move at specific times in the film.
Example: A drawn document appears while the narrator talks about documents.
Appears in
- Retrieval
Finding and returning relevant information. In RAG, this step supplies the passages used for the answer.
Example: The search finds the three most relevant sections from a hundred handbook pages.
Appears in
- Sandbox
A separate environment for running and checking code. It limits which files and services that code can access.
Example: A Python example computes 1 + 2 + 3 in a separate test environment and returns 6.
Appears in
- Token
A small piece of text processed by a language model. A token can be a whole word, part of a word or a character.
Example: A long word may split into several tokens, so token counts and word counts differ.
- Tool call
The model requests a tool to search, calculate or perform an action. It can then use the result in its answer.
Example: Instead of guessing a calculation, the model runs a Python program.
Appears in
- Utilization
How much of the available computing capacity is actually used. Unused capacity means hardware spends time waiting.
Example: A GPU running at full capacity for six of 24 hours has an average utilization of 25 percent.
Appears in
- Vector
An ordered list of numbers. For embeddings, it represents features of a text so similarities can be calculated.
Example: The numbers in a question vector are compared with the numbers representing stored passages.
Appears in