AI Glossary
Clear, precise definitions for essential concepts across foundation models, architecture, and agent systems.
Agentic Workflow
An automated pipeline where AI iteratively plans, calls external APIs/tools, evaluates intermediate outputs, and corrects errors to achieve complex objectives.
Context Window
The maximum span of tokens (text, images, audio) that an AI model can hold in its working memory and attend to during a single inference pass.
Mixture of Experts (MoE)
An architectural design where only a specialized subset of neural network weights (experts) is dynamically activated for any given token, enabling massive capacity with lower compute costs.
Reinforcement Learning from Human Feedback (RLHF)
A post-training alignment method that trains a reward model on human preferences to steer LLM outputs toward helpfulness, accuracy, and safety.
Retrieval-Augmented Generation (RAG)
A technique that fetches relevant external data from a vector database or search index and injects it into the prompt before the model generates a response.
Test-Time Compute / Reasoning Model
Models (such as DeepSeek R1 or OpenAI o-series) that spend extra computational cycles generating internal chains of thought before answering to improve accuracy on math, science, and coding.