§ AI Wiki / Glossary
One-line definitions, the AI dictionary.
§ Search this category
Search the Wiki →Anthropic's alignment technique where the model critiques and revises its own outputs against a written set of principles.
A smarter chunking method that uses embedding similarity to split documents at topic boundaries.
An LLM extending its capability by calling external tools, APIs or functions.
An open-source framework from Microsoft Research for building conversation-based multi-agent applications.
Teaching the model a task by showing a handful of examples directly in the prompt.
The discipline of systematically designing what is placed in an LLM's context and how.
The model's ability to learn a new task purely from information in the prompt, without parameter updates.
Training a smaller 'student' model to mimic the behaviour of a larger 'teacher' model.
An AI agent's ability to operate an entire computer via mouse, keyboard, and screenshots.
A classic keyword-ranking algorithm published by Robertson and Walker in the 1990s.
A memory type that recalls specific events or interactions together with their time and context.
Removing weights with negligible impact to shrink a model and speed it up.
A developer-friendly, low-friction open-source vector database.
A language model that runs entirely on a phone, laptop or tablet — no network required.
The process where a trained model takes input and produces output.
A system in which multiple AI agents collaborate, negotiate, or divide labor to accomplish a goal.
An open-source Python framework that organizes role-based agents as a collaborating crew.
A transformer architecture that processes the query and a candidate document jointly to score relevance.
A feedback loop that continuously measures and refines an agent's output.
A family of generative models that produce images, audio or video by iteratively denoising random noise.
An RLHF alternative that directly optimises a model on preference data, skipping the explicit RL loop.
A reasoning approach that explores multiple branches in parallel instead of a single chain, then picks the best path.
Prompting the model to reason step-by-step before producing a final answer.
The process by which a model's weights are updated to learn patterns from data.