Skip to content
ToolScout

AI Glossary

Comprehensive dictionary of 218+ AI and machine learning terms. Clear definitions, real examples, and practical explanations.

Browse by Category

Essential AI Terms

AI Applications

(25 terms)

AI Agent

AI system that can autonomously take actions to accomplish goals.

AI Background Removal

Automatically separating subjects from backgrounds in images.

AI Code Review

Using AI to review code for bugs, security, and quality.

AI Coding Assistant

AI tools that help write, debug, and understand code.

AI Image Editing

Using AI to modify and enhance existing images.

AI Image Upscaling

Using AI to increase image resolution while adding detail.

Computer Vision

AI field enabling computers to understand and analyze visual content.

Content Moderation

Systems and policies for managing harmful content in AI outputs.

Deepfake

AI-generated fake media that realistically depicts people doing or saying things they didn't.

Image-to-Image (img2img)

Generating new images based on existing image inputs.

Inpainting

AI technique for filling in or replacing parts of images.

Named Entity Recognition (NER)

AI technique that identifies and classifies proper nouns in text.

Negative Prompts

Text describing what should NOT appear in generated images.

OCR (Optical Character Recognition)

Technology that extracts text from images and documents.

Outpainting

AI technique for extending images beyond their original boundaries.

Semantic Search

Search based on meaning rather than exact keyword matches.

Semantic Search

Search that understands meaning rather than just matching keywords.

Sentiment Analysis

AI technique that determines the emotional tone of text.

Speech-to-Text (STT)

AI technology that converts spoken audio into written text.

Stable Diffusion

Popular open-source text-to-image AI model that can run locally.

Text-to-3D

AI systems that generate 3D models from text descriptions.

Text-to-Image

AI systems that generate images from text descriptions.

Text-to-Speech (TTS)

AI technology that converts written text into spoken audio.

Text-to-Video

AI systems that generate video content from text descriptions.

Voice Cloning

AI technology that replicates a specific person's voice.

AI Development

(63 terms)

AI Watermarking

Techniques for marking AI-generated content as machine-made.

API (Application Programming Interface)

Interface allowing software to interact with AI models programmatically.

Attention Mechanism

Neural network component that weighs the importance of different input elements.

Autoencoder

Neural network that compresses data to a smaller representation then reconstructs it.

AutoGen

Microsoft framework for building multi-agent AI applications.

Batch Processing

Processing multiple AI requests simultaneously for efficiency.

Benchmark

Standardized tests for measuring and comparing AI model performance.

CFG Scale

Parameter controlling how closely AI follows prompts.

Chunking

Splitting documents into pieces for AI processing.

Constitutional AI

Training approach where AI critiques itself based on a set of principles.

ControlNet

Technique for controlling image generation with precise inputs.

CrewAI

Framework for orchestrating role-playing AI agents.

Data Augmentation

Technique of creating variations of training data to improve model robustness.

Diffusion Model

AI architecture that generates images by gradually denoising random noise.

DreamBooth

Fine-tuning technique for teaching models new subjects.

Embeddings

Numerical representations of text that capture semantic meaning as vectors.

Fine-tuning

Additional training of an AI model on specific data to customize its behavior.

Function Calling

AI capability to invoke external tools and functions with structured outputs.

Function Calling

LLM capability to output structured API calls.

GAN (Generative Adversarial Network)

AI architecture using competing networks to generate realistic content.

GPU (Graphics Processing Unit)

Hardware accelerator essential for training and running AI models.

Hugging Face

Platform hosting AI models, datasets, and tools for the ML community.

Hybrid Search

Combining keyword and semantic search for better results.

Inference

Running a trained AI model to generate predictions or outputs.

Instruction Tuning

Fine-tuning AI models to follow natural language instructions.

Interpretability

Understanding how AI models make decisions internally.

Jailbreak

Techniques to bypass AI safety restrictions and content policies.

LangChain

Framework for building applications with large language models.

Latent Space

Compressed representation of data where similar items are near each other.

llama.cpp

Efficient framework for running LLMs on CPUs and consumer hardware.

LLMOps

Practices for deploying and managing LLM applications in production.

LoRA

Efficient technique for fine-tuning large models with minimal parameters.

LoRA (Low-Rank Adaptation)

Efficient fine-tuning technique that trains small adapter layers instead of full models.

Mixture of Experts (MoE)

Neural network architecture using specialized sub-networks for different inputs.

Model Collapse

Degradation when AI models are trained on AI-generated content.

Model Context Protocol (MCP)

Open standard for connecting AI assistants to external data sources and tools.

Model Merging

Combining multiple AI models into one with blended capabilities.

Model Serving

Infrastructure for running AI models in production environments.

Multi-Agent Systems

Multiple AI agents working together on complex tasks.

NeRF (Neural Radiance Field)

AI technique for creating 3D scenes from 2D images.

NVIDIA

Dominant company in AI hardware, providing GPUs that power most AI systems.

Ollama

Easy-to-use tool for running local LLMs.

Overfitting

When a model memorizes training data but fails on new data.

Prompt Injection

Attack technique that manipulates AI systems through malicious prompts.

Quantization

Technique to reduce AI model size by using lower precision numbers.

Quantization

Reducing model precision to decrease size and increase speed.

Reinforcement Learning

ML approach where agents learn through rewards and penalties.

Reranking

Improving search results by rescoring with a more powerful model.

Retrieval-Augmented Generation (RAG)

Technique combining AI generation with external knowledge retrieval for accurate responses.

RLHF (Reinforcement Learning from Human Feedback)

Training technique using human preferences to align AI behavior with human values.

Sampling Method

Algorithm used to generate images in diffusion models.

Scaling Laws

Mathematical relationships between model size, data, compute, and performance.

Structured Output

Forcing AI to respond in specific data formats.

Supervised Learning

ML approach using labeled examples to learn input-output mappings.

Synthetic Data

Artificially generated data used to train AI models.

TPU (Tensor Processing Unit)

Google's custom AI chip designed for tensor operations.

Training Data

The dataset used to teach AI models patterns and capabilities.

Transfer Learning

Using knowledge from one task to improve performance on a different task.

Transformer

Neural network architecture using attention mechanisms, powering modern AI.

Unsupervised Learning

ML approach finding patterns in data without labeled examples.

Vector Database

Database optimized for storing and querying high-dimensional embedding vectors.

Vector Database

Database optimized for storing and searching AI embeddings.

vLLM

High-performance LLM serving library.

AI Fundamentals

(30 terms)

AI Alignment

Ensuring AI systems pursue goals that match human intentions and values.

AI Ethics

Principles and practices for developing and using AI responsibly.

AI Safety

Field focused on ensuring AI systems are beneficial and aligned with human values.

Artificial General Intelligence (AGI)

Hypothetical AI with human-like general reasoning across all domains.

Artificial Superintelligence (ASI)

Hypothetical AI surpassing human intelligence in all domains.

Bias in AI

Systematic errors in AI outputs that reflect societal prejudices or data imbalances.

Chain of Thought

Prompting technique that improves reasoning by showing steps.

Chain-of-Thought

Prompting technique where AI explains its reasoning step by step.

Context Window

The maximum amount of text an AI model can consider at once.

Deep Learning

Machine learning using neural networks with many layers.

Emergent Abilities

Capabilities that appear in large AI models but not in smaller ones.

Few-shot Learning

AI technique where models learn tasks from just a few examples in the prompt.

Foundation Model

Large pre-trained model that serves as the base for many downstream applications.

Generative AI

AI systems that create new content like text, images, audio, or video.

Hallucination

When AI generates false or fabricated information that appears plausible.

In-Context Learning

Ability of LLMs to learn from examples provided in the prompt.

Large Language Model (LLM)

An AI system trained on vast text data to understand and generate human-like text.

Machine Learning

AI approach where systems learn patterns from data rather than explicit programming.

Multimodal AI

AI systems that can process and generate multiple types of content like text, images, and audio.

Natural Language Processing (NLP)

AI field focused on understanding and generating human language.

Neural Network

Computing system inspired by biological brains, consisting of interconnected nodes.

Open Source AI

AI models with publicly available weights that anyone can use and modify.

Prompt

The input text or instructions given to an AI model to generate a response.

Prompt Engineering

The practice of crafting effective inputs to get better outputs from AI models.

System Prompt

Hidden instructions that set an AI assistant's behavior and personality.

Temperature

Parameter controlling randomness and creativity in AI model outputs.

Tokens

The basic units of text that AI models process, typically word pieces or characters.

World Model

AI system's internal representation of how the world works.

Zero-shot Learning

AI performing tasks without any task-specific examples in the prompt.

Zero-Shot Learning

AI performing tasks without any task-specific examples.

AI Models

(19 terms)

architectures

(6 terms)

companies

(6 terms)

concepts

(59 terms)

Agentic AI

AI systems that can take autonomous actions to achieve goals, making decisions and executing multi-step tasks with minimal human guidance.

AI Benchmark

Standardized tests used to evaluate AI model performance across tasks like reasoning, coding, math, and language understanding.

AI Bias

Systematic errors in AI outputs that reflect prejudices in training data or model design, potentially causing unfair outcomes.

AI Ethics

Field studying moral implications of AI including bias, fairness, transparency, privacy, and societal impact of AI systems.

AI Leaderboard

Rankings of AI models based on benchmark performance, helping users compare capabilities across different systems.

AI Music Generation

Using AI to create original music from text descriptions or other inputs. Models can generate melodies, harmonies, and full songs.

AI Watermarking

Techniques to embed invisible markers in AI-generated content to identify it as machine-created, helping detect deepfakes and misinformation.

Batch Inference

Processing multiple inputs together rather than one at a time, improving GPU utilization and throughput.

Chain of Thought (CoT)

Prompting technique that asks models to show step-by-step reasoning, improving accuracy on complex problems.

Code Completion

AI feature that suggests code as developers type, predicting and completing functions, variables, and logic based on context.

Code Generation

AI creating complete code from natural language descriptions, generating functions, classes, or entire programs.

Constitutional AI

Anthropic's technique for training AI using a set of principles (constitution) that guide the model's behavior and responses.

Content Authenticity

Systems and standards for verifying the origin and history of digital content, distinguishing human from AI-created work.

Content Moderation

Systems that detect and filter harmful, inappropriate, or policy-violating content in AI inputs and outputs.

Context Window

Maximum amount of text (in tokens) an AI model can process at once, including both input and output. Larger windows enable longer documents.

Continuous Batching

Dynamic batching that adds new requests to running batches as slots become available, maximizing throughput without sacrificing latency.

Edge AI

Running AI models on local devices (phones, IoT, laptops) rather than cloud servers, enabling privacy and offline use.

Elo Rating

Rating system adapted from chess used to rank AI models based on head-to-head comparisons in Chatbot Arena.

Explainability

Ability to understand and interpret how AI models reach their decisions, important for trust and regulatory compliance.

Few-Shot Prompting

Providing examples in the prompt to demonstrate desired behavior, helping the model understand the task pattern.

Flash Attention

Optimized attention implementation that is faster and uses less memory by restructuring the computation to be more hardware-efficient.

Guardrails

Safety mechanisms that constrain AI outputs, filtering harmful content and ensuring responses stay within defined boundaries.

Hybrid Search

Combining semantic/vector search with traditional keyword search for more comprehensive and accurate retrieval results.

Inference Optimization

Techniques to make AI model inference faster and cheaper including quantization, batching, caching, and hardware acceleration.

Jailbreak

Prompt techniques designed to bypass AI safety measures and get models to produce otherwise refused outputs.

JSON Mode

LLM output mode that ensures responses are valid JSON, simplifying parsing and integration with downstream systems.

Knowledge Cutoff

The date up to which an AI model's training data extends. The model has no knowledge of events after this date unless given access to search.

KV Cache

Key-Value cache that stores computed attention states during LLM inference, avoiding recomputation for previously processed tokens.

Long Context

AI models with extended context windows (100K+ tokens) that can process entire books, codebases, or lengthy documents in a single query.

Max Tokens

Parameter limiting the maximum length of model output, preventing excessive generation and controlling costs.

Model Context Protocol (MCP)

Anthropic's protocol for connecting AI models to external data sources and tools in a standardized way, enabling richer integrations.

Model Distillation

Training a smaller student model to mimic a larger teacher model, transferring knowledge while reducing size and compute requirements.

Model Serving

Infrastructure for deploying AI models to handle inference requests in production, including APIs, scaling, and monitoring.

Neural TTS

Text-to-speech using neural networks that produces more natural, expressive speech than traditional concatenative or parametric methods.

On-Device AI

AI processing that happens directly on user devices like phones or laptops, keeping data local and reducing latency.

Paged Attention

Memory management technique for LLM inference that handles KV cache more efficiently, enabling better batching and throughput.

Prompt Injection

Attack where malicious instructions in user input attempt to override system prompts or manipulate AI behavior.

Prompt Template

Structured format for constructing prompts with variable placeholders, enabling consistent and reusable prompt patterns.

Pruning

Removing unnecessary weights or neurons from a neural network to reduce size and increase speed while maintaining accuracy.

Red Teaming

Adversarial testing of AI systems to find vulnerabilities, biases, harmful outputs, and ways to bypass safety measures.

Reranking

Second-stage retrieval that rescores initial results using a more sophisticated model to improve relevance ranking.

Retrieval

Finding relevant information from a knowledge base to provide context for AI responses. Core component of RAG systems.

Self-Reflection

Technique where AI models critique and improve their own outputs through iterative refinement, often catching and correcting errors.

Sliding Window Attention

Attention mechanism where each token only attends to a fixed-size window of nearby tokens, enabling efficient processing of long sequences.

Sparse Attention

Attention mechanism that only computes attention for a subset of tokens, reducing computational cost and enabling longer context windows.

Speculative Decoding

Inference optimization using a smaller draft model to predict multiple tokens, then verifying with the larger model, speeding up generation.

Speech-to-Text (STT)

Technology that converts spoken audio into written text. Also called automatic speech recognition (ASR) or voice recognition.

Stop Sequence

Token or string that signals the model to stop generating, preventing runaway outputs and controlling response boundaries.

Streaming

Receiving AI model output token-by-token as it's generated rather than waiting for the complete response, improving perceived latency.

Synthetic Data

Artificially generated training data created by AI models, used to augment real data or create datasets for specific purposes.

System Prompt

Instructions given to an AI model that define its behavior, personality, capabilities, and constraints for the conversation.

Temperature

Parameter controlling randomness in AI outputs. Lower values (0-0.3) are more focused; higher values (0.7-1) are more creative.

Text-to-Speech (TTS)

Technology that converts written text into spoken audio. Modern AI TTS produces highly natural-sounding voices.

Tool Use

Ability of AI models to use external tools like web search, calculators, APIs, and code execution to accomplish tasks beyond pure text generation.

Top-K Sampling

Sampling strategy that only considers the K most likely next tokens, preventing very unlikely tokens from being selected.

Top-P (Nucleus Sampling)

Sampling parameter that considers only tokens comprising the top P probability mass, balancing diversity and quality.

Tree of Thoughts

Extension of chain-of-thought that explores multiple reasoning paths and evaluates them, enabling more complex problem-solving.

Vision Language Model (VLM)

AI model that can process both images and text, understanding visual content and responding to queries about images.

Voice Cloning

AI technology that creates a synthetic voice matching a specific person from audio samples. Used for dubbing, accessibility, and content creation.

evaluation

(3 terms)

hardware

(2 terms)

models

(5 terms)

All Terms A-Z

Fact-Checked Expert Reviewed Regularly Updated
Reviewed by Sarah Chen, Former Product Manager at Atlassian
Our Editorial Standards

How We Research & Review

Our team tests each tool hands-on, evaluates real user feedback, and verifies claims against actual performance. We follow strict editorial guidelines to ensure accuracy and objectivity.

Hands-on testing User feedback analysis Regular updates