knowledge base
Plain-English definitions, technical mechanics, and software examples for essential artificial intelligence, large language model, and autonomous agent concepts in 2026.
browse by domain
9 terms indexed
1 term indexed
1 term indexed
1 term indexed
4 terms indexed
Agentic AI refers to autonomous AI systems capable of pursuing complex multi-step goals without continuous human intervention. Unlike standard chatbots that respond to single prompts, AI agents can break objectives into tasks, plan workflows, use external tools, execute code, and self-correct based on results.
An AI hallucination occurs when a generative language model outputs factually incorrect, ungrounded, or fabricated information with high confidence. Hallucinations arise from the probabilistic nature of next-token prediction and gaps in training data.
Chain-of-Thought (CoT) reasoning is an AI prompting and architectural technique that instructs a model to break complex problems into sequential intermediate steps before presenting the final answer. This dramatically improves accuracy in mathematical logic, planning, and multi-step debugging.
A context window is the maximum number of tokens (words and characters) a large language model can process and remember in a single interaction. It encompasses the user's prompt, uploaded files, and previous conversation history, determining how much information the model can analyze simultaneously.
Reinforcement Learning from Human Feedback (RLHF) is a machine learning alignment technique that trains AI models using human rankings and preferences. It optimizes models to be helpful, honest, and harmless, transforming raw text predictors into safe, conversational assistants.
Retrieval-Augmented Generation (RAG) is an AI architecture that enhances large language model responses by fetching verified data from external knowledge bases or private documents before generating an answer. This grounds AI responses in real-time facts, eliminates hallucination risks, and bypasses model retraining costs.
Temperature is a hyperparameter that controls the randomness and creativity of an AI model's output. A temperature of 0 produces deterministic, highly focused responses, while higher values (e.g., 0.8 to 1.2) introduce variety, unpredictability, and creative vocabulary.
Tokens are the basic atomic units of text that large language models process, measure, and generate. A token can be a single character, a subword, or a whole word. Tokenization is the algorithmic process of converting raw text into numerical token IDs.
The Transformer is the foundational deep learning neural network architecture introduced in 2017 that powers virtually all modern generative AI. It uses self-attention mechanisms to process entire sequences of data in parallel, capturing long-range contextual relationships efficiently.
Discover and compare verified software tools implementing RAG, diffusion, autonomous agents, and multimodal reasoning in our full AI directory.