AI glossary
The terms that come up in our daily AI news, explained in plain English. New ones are added every week.
A
- attention
- Attention is a method AI models use to weigh which parts of the input matter most when producing an answer.
B
- benchmark
- A benchmark is a standard test used to measure and compare how well an AI system performs on a specific task.
C
- checkpoint
- A checkpoint is a saved snapshot of an AI model at a particular point in training.
- computer-use agent
- An AI assistant that uses a computer like a person does, viewing the screen and clicking, typing, and scrolling to complete tasks.
O
- output token limit
- The maximum number of tokens an AI model can generate in one response.
S
- semantic memory
- Semantic memory is the ability to store and recall facts and meanings, separate from the specific time or place you learned them.
- subagent
- A subagent is a smaller AI helper that a main AI assistant creates to handle one focused task and report back.
T
- top-k sampling
- Top-k sampling is a method that limits an AI model’s word choices to the k most likely options at each step.
- trajectory
- The sequence of steps an AI system takes—including actions, inputs, and results—as it works toward a goal.