AI Glossary
Plain-English definitions of every important AI concept. Each entry includes examples, why it matters, and related terms.
AI Agent
An AI system that can autonomously perform tasks, make decisions, and take actions.
AI Alignment
The challenge of ensuring AI systems pursue goals that are beneficial and consistent with human values and intentions.
AI Audit
A systematic evaluation of an AI system's performance, fairness, safety, and compliance with established standards.
AI Benchmark
Standardized tests used to compare AI model performance.
AI Bias
Systematic errors in AI outputs that unfairly favor or disadvantage certain groups based on characteristics like race, gender, or age.
AI Ethics
The moral principles and philosophical frameworks guiding the responsible development and deployment of AI systems.
AI Governance
Frameworks, policies, and regulations that guide the responsible development, deployment, and use of AI systems.
AI Hallucination
When an AI generates false or fabricated information that sounds plausible.
AI Inference Optimization
Techniques that make AI models generate responses faster and cheaper without reducing output quality.
AI Jailbreak
Techniques to bypass AI safety restrictions and get prohibited outputs.
AI Memory
Mechanisms that allow AI systems to retain and recall information across conversations and sessions.
AI Orchestration
Managing and coordinating multiple AI models, tools, and workflows to complete complex end-to-end tasks.
AI Reasoning
An AI model's ability to perform logical deduction, multi-step problem solving, and chain-of-thought analysis.
AI Red Teaming
Systematically testing AI systems by attempting to make them produce harmful, biased, or incorrect outputs.
AI Safety Training
Techniques used to make AI helpful, harmless, and honest.
AI Watermarking
Embedding invisible statistical patterns in AI-generated content to enable detection and attribution of AI outputs.
Anthropic
The AI safety company that created Claude.
Artificial General Intelligence (AGI)
A hypothetical AI system with human-level reasoning across any intellectual task, not just narrow specializations.
Attention Mechanism
How AI models focus on relevant parts of input when generating output.
Backpropagation
The algorithm that computes how much each weight contributed to the error, enabling gradient descent to update them.
Batch Normalization
A technique that normalizes layer inputs across a mini-batch to stabilize and accelerate neural network training.
Chain of Thought (CoT)
A prompting technique where AI shows its reasoning step-by-step to improve accuracy.
Constitutional AI
Anthropic's approach to AI safety using explicit principles.
Context Length
The maximum amount of text an AI can process in one conversation.
Context Window
The maximum amount of text an AI can process in a single conversation.
Data Parallelism
Distributing training data across multiple GPUs that each hold a copy of the model, then synchronizing gradients.
Differential Privacy
A mathematical framework that guarantees individual data points cannot be identified in AI training data or query results.
Embeddings
Numerical representations of text that capture semantic meaning for AI processing.
Explainable AI (XAI)
Techniques that make AI decision-making processes understandable and interpretable to humans.
Federated Learning
A training approach where models learn from data distributed across many devices without the data ever leaving those devices.
Few-Shot Learning
Teaching AI by providing a few examples in the prompt.
Fine-Tuning
Training an AI model on specific data to specialize it for particular tasks.
Function Calling
An AI model's ability to output structured requests to invoke external functions, APIs, or tools during a conversation.
Google DeepMind
Google's AI research lab that created Gemini.
GPU Compute
Using graphics processing units for parallel mathematical operations that power AI training and inference.
Gradient Descent
The core optimization algorithm that adjusts neural network weights by following the slope of the loss function downward.
Grounding
Connecting AI outputs to verifiable sources and real-world data to reduce hallucinations and improve factual accuracy.
Inference (AI)
The process of an AI model generating outputs from inputs (vs. training).
Instruction Tuning
Fine-tuning a language model on instruction-response pairs so it follows human directions reliably.
Large Language Model (LLM)
An AI system trained on vast text data to understand and generate human-like text.
Latency (AI)
The delay between sending a prompt and receiving the first response token.
LoRA (Low-Rank Adaptation)
A parameter-efficient fine-tuning method that trains small adapter matrices instead of modifying the full model.
Mixed Precision Training
Training neural networks using a mix of 16-bit and 32-bit floating-point numbers to save memory and increase speed.
Model Card
A standardized document describing an AI model's capabilities, limitations, intended uses, and evaluation results.
Model Collapse
Degradation that occurs when AI models are trained on AI-generated content.
Model Distillation
Training a smaller "student" model to replicate the behavior of a larger "teacher" model at lower cost.
Model Merging
Combining weights from multiple fine-tuned models into a single model that inherits capabilities from each.
Multi-Agent Systems
Architectures where multiple specialized AI agents collaborate, debate, or coordinate to solve complex tasks.
Multimodal AI
AI that can understand and generate multiple types of content (text, images, audio, video).
Open Source AI Model
AI models with publicly available weights that anyone can download and run.
OpenAI
The company that created ChatGPT and GPT-4.
Output Tokens
The number of tokens an AI generates in its response.
Parameter (Model Size)
The number of learnable values in a neural network, often measured in billions.
Prefix Tuning
A fine-tuning method that prepends learnable virtual tokens to the input without modifying model weights.
Prompt Chaining
Breaking complex AI tasks into sequential prompts where each step's output feeds into the next step's input.
Prompt Engineering
The practice of crafting effective instructions to get better results from AI models.
Prompt Injection
Security attack where malicious instructions are hidden in AI input.
Pruning
Removing unnecessary parameters from a neural network to make it smaller and faster without significant quality loss.
RAG (Retrieval-Augmented Generation)
Combining AI with real-time information retrieval from external knowledge bases.
Responsible AI
The practice of developing and deploying AI systems that are safe, fair, transparent, and accountable throughout their lifecycle.
Retrieval-Augmented Generation (RAG) — Advanced
An advanced architecture that retrieves relevant documents from external sources to ground AI responses in factual data.
Reward Model
A model trained to score AI outputs based on human preferences, used to guide reinforcement learning from human feedback.
Semantic Search
Search that understands the meaning of queries rather than just matching keywords, using vector embeddings.
Sparse Attention
An efficient attention mechanism that processes only a subset of token relationships instead of all pairs.
Speculative Decoding
An inference speedup technique where a small model drafts tokens that a large model verifies in parallel.
Streaming (AI)
Receiving AI responses word-by-word as they are generated.
Synthetic Data
Artificially generated data used to train AI models.
System Prompt
Hidden instructions that define how an AI assistant behaves.
Temperature (AI)
A parameter that controls how creative or random an AI's responses are.
Token (AI)
A chunk of text (roughly 4 characters or 3/4 of a word) that AI models process.
Tool Use
An AI model's ability to select and invoke external tools like search engines, calculators, or code interpreters.
TPU (Tensor Processing Unit)
Google's custom AI accelerator chip designed specifically for tensor operations in machine learning workloads.
Transformer (Architecture)
The neural network architecture behind modern AI like GPT and Claude.
Zero-Shot Learning
AI performing tasks without specific examples, using only instructions.