§ AI Wiki / Glossary
One-line definitions, the AI dictionary.
§ Search this category
Search the Wiki →A hypothetical AI system matching or surpassing human intelligence across a broad range of cognitive tasks.
The discipline of designing, shipping and maintaining LLM-driven agent systems in production.
The race in which open-weight LLM labs rapidly close the capability gap with closed frontier models.
A new engineering role focused on integrating AI models into products.
AI applications built as a thin product layer on top of one or more LLM APIs.
An open-source CLI pair-programming tool created by Paul Gauthier.
Software development workflows in which AI agents read and write files, run commands, and execute tests.
A new generation of language models that engage in long-form thinking at inference time to tackle complex problems.
Apple's brand for the AI features integrated into iOS, macOS, and its other devices.
A release model where only the trained weights are published, not the training code or dataset.
A large language model whose weights, code, or training material are publicly released.
The San Francisco–based AI company behind the Claude model family, founded with an explicit safety focus.
The schema describing a tool's name, description, and parameter types so the model can call it correctly.
A hypothetical AI that decisively surpasses human intelligence across virtually all cognitive tasks.
When a model refuses harmless or reasonable requests it should have answered.
A modern AI agent where an LLM is the decision-maker, taking multiple steps and using tools.
An autonomous software system that takes its own steps toward a goal, often using tools along the way.
A structured pipeline of LLM-driven steps with explicit branching, rather than a single agent's free-form loop.
A RAG architecture in which an agent decides retrieval steps instead of a fixed pipeline.
A model's capacity to work through a problem in multiple steps — now a central axis of LLM competition.
Sending the model's response token-by-token in real time rather than waiting for the complete answer.
A secondary agent invoked by a parent agent to handle a specific subtask with its own prompt and tools.
AMD's CDNA3-architecture data-center accelerator, positioned as a credible H100 alternative.
Anthropic's alignment technique where the model critiques and revises its own outputs against a written set of principles.