AI Glossary
Comprehensive dictionary of 185+ AI and machine learning terms. Clear definitions, real examples, and practical explanations.
Browse by Category
Essential AI Terms
Large Language Model (LLM)
An AI system trained on vast text data to understand and generate human-like text.
Prompt Engineering
The practice of crafting effective inputs to get better outputs from AI models.
Hallucination
When AI generates false or fabricated information that appears plausible.
Fine-tuning
Additional training of an AI model on specific data to customize its behavior.
Retrieval-Augmented Generation (RAG)
Technique combining AI generation with external knowledge retrieval for accurate responses.
Tokens
The basic units of text that AI models process, typically word pieces or characters.
Ai applications
(24 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.
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
(59 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.
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 (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.
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.
VLLM
High-performance LLM serving library.
Ai fundamentals
(28 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 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.
Ai models
(17 terms)Anthropic
AI safety company that created Claude.
ChatGPT
OpenAI's conversational AI interface built on GPT models.
Claude
Anthropic's AI assistant known for safety and long context.
DALL-E
OpenAI's text-to-image AI model.
ElevenLabs
Leading AI voice synthesis company for text-to-speech and cloning.
Gemini
Google's multimodal AI model competing with GPT-4.
GitHub Copilot
AI pair programmer that suggests code in your editor.
Google AI
Google's AI research and products including Gemini.
GPT (Generative Pre-trained Transformer)
OpenAI's family of large language models powering ChatGPT.
Llama
Meta's family of open-source large language models.
Meta AI
Meta's AI research division creating Llama and other open-source models.
Midjourney
Popular AI image generator known for artistic quality.
Mistral
French AI company creating efficient open-source language models.
OpenAI
AI company that created ChatGPT, GPT-4, and DALL-E.
Runway
AI video generation company behind Gen-2.
Sora
OpenAI's advanced text-to-video model for minute-long videos.
Whisper
OpenAI's open-source speech recognition model.
Architectures
(5 terms)Autoregressive Model
Model that generates output one token at a time, each prediction conditioned on all previous tokens. Used by GPT and most LLMs.
Decoder-Only
Transformer architecture using only decoder blocks, generating text autoregressively. Used by GPT, Claude, Llama.
Encoder-Decoder
Architecture with separate components to encode input and decode output, used for translation and seq2seq tasks.
Mamba
State space model architecture that offers efficient long-sequence processing with selective state updates, potential Transformer alternative.
State Space Model (SSM)
Alternative to Transformers using recurrent-like computation that can process very long sequences more efficiently.
Companies
(3 terms)Google DeepMind
Google's AI research lab formed from merging Google Brain and DeepMind. Created Gemini, AlphaFold, and other breakthrough AI systems.
Mistral AI
French AI company creating efficient open-weight language models including Mistral and Mixtral, known for high performance at smaller sizes.
Stability AI
Company behind Stable Diffusion and other open-source AI models for image, video, audio, and language generation.
Concepts
(41 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 Bias
Systematic errors in AI outputs that reflect prejudices in training data or model design, potentially causing unfair outcomes.
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.
Batch Inference
Processing multiple inputs together rather than one at a time, improving GPU utilization and throughput.
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.
Content Authenticity
Systems and standards for verifying the origin and history of digital content, distinguishing human from AI-created work.
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.
Inference Optimization
Techniques to make AI model inference faster and cheaper including quantization, batching, caching, and hardware acceleration.
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.
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 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.
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.
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.
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.
Evaluation
(3 terms)Chatbot Arena
LMSYS platform where users vote on anonymous model outputs, creating human preference rankings for AI assistants.
HumanEval
OpenAI's benchmark for evaluating code generation ability, measuring whether generated functions pass unit tests.
MMLU
Massive Multitask Language Understanding benchmark testing knowledge across 57 subjects from STEM to humanities.
Hardware
(1 terms)Models
(4 terms)Flux
Latest open-source text-to-image model from Black Forest Labs, founded by original Stable Diffusion creators. Known for high quality and text rendering.
GPT-4V
Vision-enabled version of GPT-4 that can analyze images alongside text, available through ChatGPT Plus and the API.
LLaVA
Large Language and Vision Assistant, an open-source vision-language model that can understand and respond to image-based queries.
Mixtral
Mixture of Experts model from Mistral AI using 8 expert networks, achieving high performance while only activating 2 experts per token for efficiency.
All Terms A-Z
A
C
F
G
I
L
M
N
O
R
S
T
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.