AI Glossary
Plain-English definitions of AI terms — no CS degree required. Understand the technology behind the tools so you can make smarter buying decisions.
Large Language Model (LLM)
The AI "brain" behind tools like ChatGPT, Claude, and Gemini.
Definition
A large language model is a type of artificial intelligence trained on massive amounts of text data. It learns patterns in language — grammar, facts, reasoning, and style — and uses those patterns to generate human-like text. When you chat with ChatGPT or Claude, you are interacting with an LLM.
Why It Matters
LLMs power almost every AI text tool on the market. Understanding what they are helps you evaluate tool quality, pricing (bigger models cost more to run), and limitations (they predict likely text, not guaranteed facts).
Example
When you ask ChatGPT to "write a cold outreach email for my SaaS product," the LLM generates text by predicting what words would most likely follow your prompt, based on its training data.
Tokens
The "units" AI models use to measure text — roughly 1 token equals 0.75 words.
Definition
Tokens are chunks of text that AI models process. A token can be a word, part of a word, or a punctuation mark. The word "understanding" might be split into "under" and "standing" (2 tokens). Most AI tools use tokens to measure usage limits and pricing.
Why It Matters
Token limits determine how much text you can send to and receive from an AI model in one conversation. Understanding tokens helps you choose the right pricing plan and avoid hitting usage limits.
Example
If a tool says it has a "128K token context window," that means it can process roughly 96,000 words (about 200 pages) in a single conversation.
Context Window
How much text an AI can "remember" in a single conversation.
Definition
The context window is the total amount of text (measured in tokens) that an AI model can consider at once. This includes both what you send (your messages, uploaded documents) and what the AI responds with. Once a conversation exceeds the context window, the AI starts "forgetting" earlier parts.
Why It Matters
A larger context window means you can upload longer documents, have more extended conversations, and get responses that consider more information. Claude has 200K tokens, Gemini has 1M tokens, and ChatGPT has 128K tokens.
Example
If you want to upload a 100-page legal contract for analysis, you need a model with at least a 128K token context window. Claude (200K) and Gemini (1M) can handle this easily.
Prompt Engineering
The skill of writing effective instructions to get better results from AI tools.
Definition
Prompt engineering is the practice of crafting your inputs (prompts) to AI models to get the best possible outputs. This includes being specific about format, tone, audience, length, and examples. Good prompts turn generic AI responses into genuinely useful outputs.
Why It Matters
The difference between a mediocre and excellent AI output is usually the prompt, not the model. Learning basic prompt engineering can 5-10x the value you get from any AI tool.
Example
Instead of "Write a blog post about AI," try: "Write a 1,000-word blog post for non-technical startup founders about how to use AI tools to save time on marketing. Include 3 specific examples. Use a conversational, practical tone."
Hallucination
When an AI confidently states something that is incorrect or made up.
Definition
AI hallucination occurs when a language model generates text that sounds plausible but is factually incorrect. The AI is not "lying" — it is predicting likely text patterns, and sometimes those predictions are wrong. Hallucinations can include fake citations, incorrect statistics, or entirely fabricated information.
Why It Matters
Every AI tool can hallucinate. Understanding this limitation is critical for using AI responsibly — always verify important facts, especially numbers, dates, legal claims, and medical information.
Example
You ask an AI to cite sources for a claim, and it generates a URL that looks real but leads to a page that does not exist. Or it states "According to a 2024 McKinsey report..." when no such report exists.
Fine-Tuning
Training an existing AI model on your own data to specialize it for your needs.
Definition
Fine-tuning takes a pre-trained AI model and further trains it on a specific dataset — such as your company's documents, writing style, or industry terminology. This creates a specialized version of the model that performs better on your specific tasks.
Why It Matters
Most businesses will never need to fine-tune a model. Tools like Jasper and Grammarly offer brand-voice features that achieve similar results without technical complexity. Fine-tuning is relevant for companies building AI into their own products.
Example
A legal tech company fine-tunes an LLM on millions of legal documents so it understands legal terminology, case citations, and contract language better than a general-purpose model.
RAG (Retrieval-Augmented Generation)
A technique where AI searches through documents before answering, reducing made-up information.
Definition
RAG combines a search system with an AI model. Instead of generating answers purely from its training data, the AI first retrieves relevant documents from a knowledge base, then generates a response based on those specific documents. This grounds responses in real data.
Why It Matters
RAG is how tools like Perplexity AI work — they search the web first, then synthesize answers from real sources. It is also how Notion AI answers questions across your workspace. RAG significantly reduces hallucinations.
Example
When you ask Perplexity "What is Stripe's current pricing?", it searches the web for Stripe's pricing page, retrieves the current data, and gives you an answer with citations — rather than guessing from training data that might be outdated.
Diffusion Model
The AI architecture behind image generators like Midjourney, DALL-E, and Stable Diffusion.
Definition
Diffusion models generate images by starting with random noise and gradually refining it into a coherent image, guided by your text description. Think of it like a sculptor starting with a block of marble and chipping away until a statue emerges — except the "chipping" is guided by your text prompt.
Why It Matters
All major AI image generators use diffusion models. Understanding this helps explain why they sometimes produce unexpected results — the process involves randomness, so the same prompt can produce different images each time.
Example
When you type "a sunset over mountains, oil painting style" into Midjourney, the diffusion model starts with random pixels and iteratively refines them, using your description to guide each refinement step until a photorealistic sunset painting emerges.
Text-to-Image
AI that converts written descriptions into images — what Midjourney, DALL-E, and others do.
Definition
Text-to-image models take a natural language description (prompt) and generate a corresponding image. The quality and style of the output depend on the model, the prompt quality, and any style parameters you specify.
Why It Matters
Text-to-image tools have made professional visual content accessible to anyone who can describe what they want. This is particularly valuable for startups and small businesses that cannot afford professional designers for every piece of content.
Example
You type "modern minimalist logo for a coffee shop called Brew Lab, clean lines, earth tones" and the AI generates several logo concepts in seconds.
Multimodal AI
AI that can understand and generate multiple types of content — text, images, audio, and video.
Definition
Multimodal AI models can process and generate different types of content. For example, GPT-4o can read text, analyze images, understand audio, and generate text and images in response. This allows more natural interactions where you can show the AI a photo and ask questions about it.
Why It Matters
Multimodal capabilities are becoming standard in AI assistants. You can take a photo of a whiteboard and ask the AI to organize the notes, or upload a chart and ask for analysis. This is especially useful for non-technical users.
Example
You photograph a restaurant receipt in a foreign language and ask Gemini to translate it and categorize the expenses. The AI analyzes the image and provides structured text output.
Machine Learning (ML)
The broader field of AI where computers learn from data instead of being explicitly programmed.
Definition
Machine learning is a subset of artificial intelligence where algorithms learn patterns from data. Instead of writing rules manually ("if temperature > 90, then hot"), you show the algorithm thousands of examples and it learns the patterns itself. LLMs are one type of machine learning model.
Why It Matters
Machine learning is the foundation of every AI tool discussed on this site. Understanding that these tools learn from data (not magic) helps set realistic expectations about what they can and cannot do.
Example
Grammarly uses machine learning trained on millions of well-written documents to learn what "good writing" looks like. It then flags your text when it deviates from those patterns.
Deep Learning
A powerful subset of machine learning using layered neural networks — the tech behind modern AI.
Definition
Deep learning uses artificial neural networks with many layers (hence "deep") to learn increasingly complex patterns from data. The first layers might learn simple patterns like edges in an image, while deeper layers learn complex concepts like faces or objects. All modern AI tools use deep learning.
Why It Matters
Deep learning is why AI has improved so dramatically in recent years. The "deep" architectures can learn subtle, complex patterns that simpler algorithms miss. You do not need to understand the math, but knowing the term helps you follow AI news and conversations.
Example
DALL-E uses deep learning to understand the relationship between text descriptions and images, allowing it to generate new images from text prompts it has never seen before.
Neural Network
The architecture behind modern AI — loosely inspired by how brain neurons connect.
Definition
A neural network is a computing system made up of interconnected nodes (artificial neurons) organized in layers. Data flows through these layers, and the connections between nodes have "weights" that are adjusted during training. Neural networks are the building blocks of deep learning and LLMs.
Why It Matters
Neural networks are the architecture behind every AI tool you use. When someone says a model has "175 billion parameters," they are referring to the number of adjustable weights in the neural network. More parameters generally (but not always) means more capable models.
Example
GPT-4 is a neural network with hundreds of billions of parameters. Each parameter was adjusted during training on internet text to optimize the model's ability to predict and generate language.
Grounding
Connecting AI responses to real, verifiable data sources to reduce errors.
Definition
Grounding is the practice of anchoring AI responses in real data — whether from web searches, uploaded documents, or specific databases. A "grounded" response includes references to actual sources, making it more reliable and verifiable than a response generated purely from training data.
Why It Matters
Grounded AI tools like Perplexity AI are more reliable for factual research because they cite sources. When evaluating AI tools for business use, tools with grounding capabilities are safer for decisions that depend on accurate information.
Example
Perplexity AI grounds its responses by searching the web in real-time and citing specific URLs for each claim. This is more reliable than asking ChatGPT the same question, which relies on training data that may be outdated.
Few-Shot Learning
Teaching AI by showing it a few examples of what you want — a key prompting technique.
Definition
Few-shot learning is a prompting technique where you include a few examples of the desired input-output pattern in your prompt. The AI then follows the pattern for new inputs. This is one of the most practical skills for getting better results from AI tools.
Why It Matters
Instead of describing what you want in words (which can be ambiguous), showing 2-3 examples makes the AI understand exactly what format, style, and tone you need. This works with any AI tool.
Example
Instead of saying "summarize these emails," provide two examples: "Email: [text] Summary: [your preferred summary style]" — then ask the AI to summarize a new email. It will follow the pattern from your examples.
Constitutional AI
Anthropic's approach to making AI helpful, harmless, and honest.
Definition
Constitutional AI is a training technique developed by Anthropic (creators of Claude) where the AI is trained to follow a set of principles or "constitution" that guides its behavior. Instead of relying solely on human feedback, the AI is taught to evaluate and revise its own responses based on these principles.
Why It Matters
This is why Claude tends to give more nuanced, careful responses than some competitors. For business users, this means more reliable outputs for sensitive topics like legal analysis, financial advice, and HR communications.
Example
When asked an ethically complex question, Claude will often acknowledge multiple perspectives and potential risks rather than giving a one-sided answer — this is constitutional AI in action.
AI Code Generation
AI that writes, completes, and modifies computer code based on natural language descriptions.
Definition
AI code generation tools convert natural language descriptions into working code, autocomplete partial code, or modify existing code based on instructions. These tools are trained on billions of lines of code and can generate code in virtually any programming language.
Why It Matters
AI code generation has transformed software development. Tools like GitHub Copilot and Cursor can increase developer productivity by 30-55%. For non-technical founders, tools like Replit can generate entire applications from descriptions.
Example
A developer types "function that validates an email address and returns true/false" and Copilot generates the complete function with proper regex pattern matching, edge case handling, and error management.
Speech-to-Text (STT)
AI that converts spoken audio into written text — used for transcription and meeting notes.
Definition
Speech-to-text AI converts spoken language into written text in real-time or from recordings. Modern STT models achieve 85-95% accuracy and can handle multiple speakers, accents, and technical jargon. They power tools like Otter.ai for meeting transcription.
Why It Matters
STT tools save hours of manual note-taking and make meeting content searchable. For founders and managers who spend significant time in meetings, these tools can be among the highest-ROI AI investments.
Example
Otter.ai joins your Zoom call, transcribes everything in real-time with speaker identification, then generates a summary with action items — all automatically.
Open-Source AI
AI models whose code and weights are publicly available — free to use, modify, and run yourself.
Definition
Open-source AI models are released publicly so anyone can download, run, modify, and deploy them. This contrasts with closed-source models (like GPT-4 or Claude) that can only be accessed through the company's API. Open-source examples include Stable Diffusion, LLaMA, and Mistral.
Why It Matters
Open-source AI gives you maximum control, privacy, and customization — but requires technical ability to set up and maintain. For most business users, commercial tools are more practical. Open-source matters if you need data privacy, custom models, or very high-volume usage.
Example
Stable Diffusion is an open-source image model. You can download it and run it on your own computer for unlimited free image generation, or use it through cloud services like DreamStudio for convenience.
Enterprise AI
AI tools designed for business use with security, compliance, and admin controls.
Definition
Enterprise AI refers to AI tools and plans designed for organizational use. They typically include features like SSO (single sign-on), admin dashboards, usage analytics, data retention policies, compliance certifications (SOC 2, HIPAA), and guarantees that your data will not be used for model training.
Why It Matters
If your company handles sensitive data (financial, health, legal), you need enterprise-grade AI tools. The free tiers of most AI tools may use your data for training. Enterprise plans typically offer data protection guarantees.
Example
ChatGPT Team ($30/user/mo) guarantees that your conversations are not used for training and includes admin controls for managing team access. The free tier does not offer these protections.
AI Content Generation
Using AI to create text, images, video, or audio content — the most common business use of AI.
Definition
AI content generation encompasses any use of AI to create new content. This includes writing articles, generating images, creating presentations, drafting emails, and producing marketing materials. It is the broadest and most immediately practical application of AI for businesses.
Why It Matters
Content generation is where most businesses see the fastest ROI from AI. Whether you are writing blog posts, creating social media graphics, or drafting customer emails, AI tools can reduce production time by 50-80% while maintaining quality.
Example
A marketing team uses ChatGPT to draft a blog post, Midjourney to generate the header image, and Grammarly to polish the final copy — producing in 2 hours what used to take a full day.
LangChain
A popular framework for building AI-powered applications — relevant for technical builders.
Definition
LangChain is an open-source framework that makes it easier to build applications using large language models. It provides tools for chaining AI calls together, connecting to databases, implementing RAG, and building complex AI workflows. It is the most popular framework for AI application development.
Why It Matters
If you are building AI features into your own product (not just using existing tools), LangChain is likely the framework your developers will use. Understanding what it does helps you have informed conversations with your technical team.
Example
A developer uses LangChain to build a customer support chatbot that can search your company's knowledge base, retrieve relevant articles, and generate helpful responses — combining RAG, conversation history, and LLM calls.
Deployment
Publishing your application so users can access it — made easier by tools like Replit and Vercel.
Definition
Deployment is the process of taking your code from a development environment and making it available on the internet for users. Traditionally this required server configuration and DevOps knowledge. Modern tools like Replit, Vercel, and Netlify make deployment as simple as clicking a button.
Why It Matters
For non-technical founders, understanding deployment helps you evaluate tools like Replit (one-click deploy) vs traditional development workflows. Easy deployment means faster iterations and quicker time to market.
Example
You build a landing page on Replit and click "Deploy." Within seconds, it is live on the internet with a public URL you can share with potential customers — no server setup required.
23 terms defined · Updated regularly · Take the quiz to find the right tools for you