Glossary of AI Terms

the new AI language

A

• Adversarial Example

• Adversarial Attack

• Agent Architecture

• Agent Evaluation

• Agent Framework

• Agentic Workflow

• Agent Loop

• Agent Memory

• Agent Orchestration

• Agent Planning

• Agent Protocol

• AGI (Artificial general intelligence)

• AI Accelerator

• AI Agent

• AI Alignment

• AI Arms Race

• AI Assistant

• AI Auditing (AI Forensic)

• AI Autonomy

• AI Benchmarking

• AI Benchmarks

• AI Compute

• AI Constitutions

• AI Control Problem

• AI Copilot

• AI Ethics

• AI Fluency

• AI Governance

• AI Guardrails

• AI Infrastructure

• AI Model Zoo

• AI Native

• AI Poisoning (black-hat SEO)

• AI Psychosis

• AI Red Teaming

• AI Safety

• AI Scaling

• AI Sovereignty

• AI Toxicity

• AI Winter

• AI Workflow

• Alignment Problem

• ANI (Artificial narrow intelligence)

• API Rate Limits

• API endpoints

• ASI (Artificial superintelligence)

• AVX-512

• AVX2

• Ablation

• Abliteration

• Activation Function

• Adam Optimizer

• Affirmative AI

• Agent Skill Malware

• Agent-to-Agent (A2A)

• Agentic AI

• AlexNet

• Algorithm

• Alignment Tax

• AlphaGo

• Anthropomorphism

• Apple Silicon

• Artificial Intelligence

• Attention Heads

• Attention Mechanism

• Automation

• Autonomous Agent

B

• Beam Search

• Benchmark Contamination

• BF16

• Black Box Model

• BLEU

• Bloom Filter

• BM25

• Bootstrap Sampling

• BPE (Byte Pair Encoding)

• Boltzmann Machine

• Browser Automation Agent

• Backpropagation

• Batch

• Bayesian Networks

• Bias (in AI)

• Big Data

• Bitter Lesson

• Browser Agent

C

• Calibration (AI Calibration)

• Capability Evaluation

• Classifier

• Classification

• CLIP (Contrastive Language–Image Pretraining)

• CNN (Convolutional Neural Network)

• Connectionism

• Context Engineering

• Continuous Batching

• ControlNet

• Corrigibility

• Cosine Similarity

• CPU

• Cross Attention

• CUA (Computer-Using Agent)

• CUDA

• CUDA Cores

• Catastrophic Forgetting

• Chain of thought

• Chain-of-thought Prompting

• Chat template

• Chatbot

• Checkpoint

• Chinese Room

• Chunk Overlap

• Chunk Size

• Chunking

• Claw

• Closed Source

• Cloud AI

• Computer Use

• Computer Vision

• Constitutional AI

• Content Moderation

• Context Window

• Context Window Management

• Continued Pretraining (or Domain-adaptive Pretraining)

• Contrastive Learning

• Cross Encoder

• Cybernetics

• Curriculum Learning

D

• Dartmouth Workshop

• Data Augmentation

• Data Curation

• Data Governance

• Data Leakage

• Data Mining

• Data Pipeline

• Dataset

• Deceptive Alignment

• Decision Boundary

• Decision Tree

• Decoder

• Decoder-only Models

• Deep Blue

• Deepfake

• Deep Learning

• Diffusion Models

• Dimensionality Reduction

• Distillation

• Distributed Representation

• Distributed Training

• DPO (Direct Preference Optimization)

• DPO vs RLHF

• Domain Adaptation

• Draft Model

E

• E2E (end-to-end learning)

• ELIZA

• ELO Benchmarks

• Edge AI

• Embedding Model

• Embedding Space

• Embeddings

• Emergent Behavior

• Emergent Misalignment

• Encoder

• Encoder-only Models

• Encoder–Decoder Model

• Ensemble Methods

• Epoch

• Ethical AI

• Evaluation

• Evaluation Dataset

• Evolutionary Algorithms

• Expert Systems

• Expert System Winter

• Exploration vs Exploitation

F

• Feature Engineering

• Feature Extraction

• Feature Store

• Fine-Tuning Dataset

• Flash Attention 2

• Foundation Model vs LLM

• FP16

• FP8

• Federated Learning

• Feed-Forward Network (FFN)

• Few-Shot Learning

• Fine-Tuning

• Fine-Tuning vs Continued Pretraining

• Flash Attention

• Foundation Models

• Frontier AI model

• Function Calling (alongside Tool Calling)

G

• GANs (Generative Adversarial Networks)

• Generalization

• Generative Model

• GEO (Generative Engine Optimization)

• GGML

• GGUF (GPT-Generated Unified Format)

• GNN (Graph Neural Network)

• GPT (Generative Pre-Trained Transformer)

• GPU (Graphics Processing Unit)

• GenAI (Generative AI)

• Goal Misgeneralization

• Goodhart’s Law (AI)

• Gradient

• Gradient Accumulation

• Gradient Clipping

• Gradient Descent

• Grounding

• GQA (Grouped Query Attention)

• Guardrails

H

• HAI (Human-Centered Artificial Intelligence)

• Hard Prompt

• HCI (Human-Computer Interaction)

• Hallucinations (in AI)

• Hidden State

• Hugging Face

• Human Evaluation

• Human Feedback

• Human-in-the-Loop

• Human Oversight

• Hybrid Search

• Hyperparameter

I

• INT4

• INT8

• Image Generation

• Image Segmentation

• ImageNet

• Inference

• Inference Server

• Inference vs Training

• Instance Segmentation

• Instruction Following

• Instruction Tuning

• Instrumental Convergence

• Intelligence Explosion

• Interpretability

• Inverse Reinforcement Learning

• IoT (Internet of Things)

• Iterative Refinement

J

• JAX

• JSON Mode

• Jailbreak

• Joint Embedding

• Judge Model

K

• KV Cache

• KV Cache Eviction

• KV Quantization

• Knowledge Cutoff

• Knowledge Distillation

• Knowledge Graph

• Knowledge Representation

• k-NN (k-Nearest Neighbors)

L

• Latent Diffusion

• Latent Diffusion Model

• Language Model

• Language Reasoning

• Layer Normalization

• llama.cpp

• LLM (Large Language Model)

• LLMOps (Large Language Model Operations)

• Logic Programming

• Logits

• Long Context Models

• Latency

• Latent Space

• Learning Rate

• LoRA (Low-Rank Adaptations)

• Local AI

• Local AI hardware

• Local LLM

• Loss Function

• Loss Landscape

M

• Machine Learning Model

• Machine Translation

• MCP (Model Context Protocol)

• ML (Machine Learning)

• MLM (Masked Language Model)

• MLOps (Machine Learning Operations)

• MMLU

• Markov Chain

• Matrix Multiplication

• Memorization (rather than the intended generalization during training)

• Memory vs Context Window

• Mesa Optimization

• Metal

• Mini-batch

• MLX

• Mmap (memory mapping)

• MPS (Metal Performance Shaders)

• MoE (Mixture of Experts)

• Model

• Model Drift

• Model Hub

• Model types

• Model vs Architecture

• Model Collapse

• Model Context Window

• Model Merging

• Model Routing

• Modern AI

• Moravec’s Paradox

• Multi-Agent Collaboration

• Multi-Agent Debate

• Multi-Agent System

• Multi-Head Attention

• Multimodal AI

• Multimodal Foundation Models

• Multimodal Retrieval

N

• NAS (Neural Architecture Search)

• NLP (Natural Language Processing)

• NPU (Neural Processing Unit)

• Neural Network

• Neural Scaling

• Neural-Symbolic AI

• Neural Turing Machine

• Neural Reasoning

• Normalization

O

• Object Detection

• OCR (Optical Character Recognition)

• Ollama

• One-Shot Learning

• ONNX

• OOD (Out-of-Distribution)

• OpenAI Gym / Gymnasium

• Open Source

• Open Source vs Open Weight

• OpenVINO

• Open-Weight Model

• Optical Flow

• Optimization Algorithm

• Optimizer

• Orthogonality Thesis

• Overfitting

P

• Paged Attention

• Paperclips (The Paperclip Maximiser theory)

• Parameter

• Parameters vs Weights

• PEFT (Parameter-Efficient Fine-Tuning)

• Perception

• Perceptron

• Perceptron Winter

• Perplexity

• Persona Prompting

• Pipeline Parallelism

• Pose Estimation

• Positional Embeddings

• Positional Encoding

• Power-Seeking AI

• Practical AI

• Predictive Analytics

• Preference Optimization

• Pretraining

• Pretrained Model

• Private reasoning

• Prompt

• Prompt Caching

• Prompt Chaining

• Prompt Compression

• Prompt Engineering

• Prompt Injection

• Prompt Injection vs Jailbreak

• Prompt Leakage

• Prompt Template

• Prompting

• Pruning

• PTQ (Post-Training Quantization)

Q

• QAT (Quantization-Aware Training)

• Quantization

• Quantization Levels

R

• RAG (Retrieval-Augmented Generation)

• RAG ecosystem

• RAM

• Reasoning Trace

• Reinforcement Learning (RL)

• RL Environment

• RLHF Alternatives

• RLHF (Reinforcement Learning from Human Feedback)

• ROCm

• ROUGE

• Re-ranking

• Reasoning Models

• Residual Stream

• Responsible AI

• Retriever

• Reward Hacking

• RMSE (Root Mean Square Error)

• Robustness

• RoPE (Rotary Positional Embeddings)

• Robotics

• Role Prompting

S

• Safetensors

• Safety Evaluation

• Sandboxing

• Scalable Oversight

• Semantic Embedding

• Semantic Segmentation

• Sequence-to-Sequence Model

• SGD (Stochastic Gradient Descent)

• Shadow AI

• Scaling Hypothesis

• Scaling Laws

• Self-Attention

• Self-Supervised Learning

• Semantic Analysis

• Semantic Search

• Serving

• Simulators Theory

• Single Direction

• Sliding Window Attention

• Slop

• Small Language Models (SLM)

• Softmax

• Sparse Attention

• Sparse Model

• Spatial Intelligence

• Specification Gaming

• Speculative Decoding

• Speculative Sampling

• Stochastic Parrot

• Structured Outputs

• Style transfer

• Supervised Learning

• Sycophantic chatbots (sycophancy)

• Symbolic AI

• Synthetic Data

• Synthetic Data Generation

• System 1 / System 2 AI

• System Prompt

T

• Temperature (AI Temperature)

• Temperature vs Randomness

• Temperature Sampling

• Temperature Scaling

• Tensor

• Tensor Cores

• Tensor Parallelism

• TensorRT

• Test-Time Compute

• Thinking Models

• Throughput

• Token

• Token Budget

• Token Unmaxxing

• Tokenization

• Tokenizer

• Tokens per second (TPS)

• Tokens vs Words

• Tool Calling

• Toolformer

• Tool Use

• Traditional AI

• Training

• Training Compute

• Training Data

• Transfer Learning

• Transformer

• Transformer Attention

• Transformer Layer

• Tree of Thoughts

• Truthfulness Evaluation

• Turing Test

U

• Uncensored (AI Model)

• Uncertainty Estimation

• Underfitting

• U-Net

• Unified Memory

• Unsupervised Learning

• Utility Function

V

• VAE (Variational Autoencoder)

• Validation Set

• Value Alignment

• Vector Embedding

• Verification

• Vision Encoder

• vLLM

• VRAM

• Vector Database

• Vector Search

• ViT (Vision Transformer)

• Vibe Coding

• Vision-Language Models (VLM)

• Vulkan

W

• Weak AI (narrow AI)

• Weight Decay

• Weight Orthogonalization

• Weight Quantization

• Weights (Model Weights)

• Wireheading

• World Models

• Workflow Automation

X

• XAI (Explainable AI)

• XML Prompting

Y

• YAML

• YOLO (You Only Look Once)

• Yield

Z

• Zero-Shot Learning