AI Glossary
40 essential AI terms explained in plain language. Built for founders, product leaders, and corporate teams who want clarity without the jargon.
The Foundations
AI (Artificial Intelligence)
Computers doing tasks that normally require human thinking. Understanding language, recognizing images, making decisions, spotting patterns.
Read definitionMachine Learning (ML)
Instead of programming rules manually, you show a computer thousands of examples and let it figure out the patterns itself.
Read definitionDeep Learning
A type of machine learning using many layers of processing to handle complex patterns. The technology behind breakthroughs in image recognition and voice assistants.
Read definitionGenerative AI (Gen AI)
AI that creates new content: text, images, code, audio, video. This is the category ChatGPT, Claude, and Midjourney belong to.
Read definition
Models and Providers
Model
A computer program trained to process input and generate a response. Think of it as a very well-read intern who has consumed the entire internet.
Read definitionLLM (Large Language Model)
A type of AI model trained on massive amounts of text to read, write, summarize, and reason in natural language.
Read definitionGPT (Generative Pre-trained Transformer)
Generative: it creates new content. Pre-trained: it learned from massive text data first. Transformer: the architecture that lets it understand context.
Read definition
How Models Learn
Training / Pre-training
Teaching a model by showing it massive amounts of data. The model adjusts millions of internal settings until it gets better at predicting patterns.
Read definitionFine-tuning
Taking a pre-trained model and doing additional focused training on your specific data, like your company's support conversations.
Read definitionRLHF (Reinforcement Learning from Human Feedback)
Human reviewers rate AI responses, and the model learns to produce more of the preferred style. This is how models become helpful and conversational.
Read definitionSynthetic Data
Artificially generated data that mimics real data. Used when real data is limited, sensitive, or fully exhausted.
Read definition
How Models Work
Transformer
The neural network architecture behind modern AI. Its key innovation lets the model look at all words at once instead of one by one.
Read definitionToken
The basic unit of text a model reads and writes. Sometimes a full word, sometimes part of a word. 'ChatGPT is smart' is roughly 5 tokens.
Read definitionContext Window
How many tokens a model can 'see' at once. Think of it as working memory. Small window = a few pages. Large window = an entire book.
Read definitionParameters / Weights
The internal numbers inside a model adjusted during training. More parameters generally means larger, more capable, but more expensive.
Read definition
Using AI in Practice
Prompt
The input you give an AI model. A question, instruction, or piece of context. Everything you type into ChatGPT or Claude is a prompt.
Read definitionPrompt Engineering
The skill of crafting prompts that get better, more reliable results. Small changes in phrasing can lead to dramatically different outputs.
Read definitionSystem Prompt
A hidden instruction developers set to control AI behavior. Example: 'You are a helpful support agent. Always be polite. Never discuss competitors.'
Read definitionRAG (Retrieval-Augmented Generation)
A technique that lets AI search your documents or database before generating a response. Like giving it an open-book test instead of relying on memory.
Read definitionEvals (Evaluations)
Structured tests to measure AI performance on specific tasks. You define what 'good' looks like and measure results.
Read definitionInference
When a trained model runs and generates a response. Every ChatGPT answer is inference happening.
Read definition
AI Agents and Automation
Agent
An AI system that takes actions on your behalf to accomplish a goal. Unlike a chatbot, an agent can plan, break tasks into steps, and use tools.
Read definitionTool Use / Tool Calling
When AI connects to external tools (search, databases, APIs) to complete tasks instead of relying only on its own knowledge.
Read definitionMCP (Model Context Protocol)
An open standard that makes it easy for AI to connect to external tools: calendar, CRM, Slack, codebase.
Read definitionVibe Coding
Building apps using AI tools by describing what you want in plain language rather than writing code. Tools like Lovable, Cursor, and Replit make this possible.
Read definition
Quality and Risk
Hallucination
When AI generates a confident response that is factually wrong or made up. It fills gaps with plausible-sounding fiction.
Read definitionTemperature
A setting controlling how creative or predictable AI responses are. Low = consistent, focused. High = creative, varied.
Read definitionBenchmark
A standardized test to compare AI models. Like a university entrance exam for AI.
Read definitionRed-teaming
Actively trying to break or exploit an AI system to find safety problems before deployment. Stress-testing for AI.
Read definition
Infrastructure
API (Application Programming Interface)
The way software talks to an AI model. Your app sends a prompt, the model processes it, sends back a response.
Read definitionCompute
The processing power (mainly GPU chips) needed to train and run AI. Expensive and one of the biggest cost drivers.
Read definitionOpen Weights / Open-Source Models
When an AI company releases model parameters so anyone can download, run, and modify it. Meta's Llama and Mistral are key examples.
Read definitionLatency
Time from sending a request to getting a response.
Read definition
Sales Pitch Decoder
AI-native
The product was built with AI at its core, not bolted on after the fact.
Read definitionGrounded
The AI has access to real data (usually via RAG) so it's less likely to hallucinate.
Read definitionMulti-modal
The AI handles multiple types of input: text, images, audio, video. Not just one.
Read definitionZero-shot
The AI can do a task it wasn't specifically trained for, just from your instructions.
Read definitionFew-shot
You give the AI a few examples of what you want, and it learns the pattern on the fly.
Read definitionEmbeddings
A way to represent text as numbers so AI can search by meaning, not just keywords.
Read definitionVector Database
A database designed to store and search embeddings. Makes RAG work at scale.
Read definition