Independent resource · not affiliated with or endorsed by Clemson University

AI Workshops by ProvenLabs

Resource · Plain language

The AI Primer.

Part one

Common terminology

The basics: what AI is

Artificial intelligence (AI)
Software that performs tasks normally associated with human intelligence, such as understanding language, recognizing patterns, drawing conclusions, and making decisions. An umbrella term that spans many different methods and products.
Machine learning
A way of building software where the system improves at a task by finding patterns in large amounts of data, rather than following rules a programmer wrote by hand. It powers most of today's AI products.
Generative AI
AI that creates new content, such as text, images, audio, or code, in response to a request. It produces original material rather than only classifying or retrieving things that already exist.
Large language model (LLM)
An AI system trained on enormous amounts of text to predict and produce language. It can draft, summarize, answer questions, and explain in ordinary writing. ChatGPT, Claude, and Gemini are all built on large language models.

How an AI is built

Training data
The collection of text, images, or other examples a model learns from while it is built. The range and quality of that data shape what the model knows and where it is weak.
Token
The small unit of text an AI reads and writes, usually a short chunk of a word. Input, output, cost, and capacity are all measured in tokens rather than words or characters.
Fine-tuning
Additional training that adapts a general model to a narrower purpose, such as a particular drafting style or subject area, by training it further on a focused set of examples.

Talking to an AI

Prompt
The instruction or question you give an AI to tell it what you want. A prompt can be a single sentence or a detailed set of directions, examples, and constraints.
Prompt engineering
The practice of writing and structuring prompts to get reliable, high-quality results. It covers being specific, supplying examples, setting a role, giving constraints, and refining the wording based on what the model returns.

What the AI works with

Context
All the information an AI has in front of it for a task: the prompt, any documents provided, and the earlier turns of the conversation. A model can only use what is in its context at that moment.
Context window
The maximum amount of information, measured in tokens, a model can hold at once. When a conversation or document runs past the window, the oldest material drops out and is no longer available.
Context engineering
The practice of curating everything a model sees during a task, the instructions, supplied documents, available tools, conversation history, and memory, so it has the right information and not a flood of irrelevant detail. Many practitioners consider it the discipline that matters most for reliable AI agents.
Memory
A system's ability to retain information across separate sessions, so it can recall facts, preferences, and past work the next time you use it. Without memory, every session starts from a blank slate.
Grounding
Tying an AI's response to a specific, trusted set of information so the answer rests on real sources rather than on what the model happens to remember from its general training. A grounded answer can show the documents, records, or systems it was drawn from. That is the difference between an AI citing a real document pulled from your own files and inventing one that sounds plausible.
Retrieval-augmented generation (RAG)
The most common technique for grounding an AI. When a question comes in, the system first searches a chosen body of documents for the most relevant passages, drops them into the prompt, and only then asks the model to answer. The result is built from those specific sources rather than from the model's general training. Pronounced “rag.”
Hallucination
When an AI states something false while sounding completely confident, such as citing a source that does not exist. It is the central reason AI output must be checked by a person before anything important rests on it.

When AI does more than chat: agents and automation

Workflow automation
A fixed, predefined sequence of steps that runs on its own once triggered. Plenty of it involves no AI at all, such as rules, templates, and if-this-then-that triggers between apps. AI can be added to one or more steps, but the path itself is set in advance, which keeps the outcome predictable and well suited to routine, repeatable tasks.
Tool use
An AI's ability to take actions beyond writing text, such as searching a database, running a calculation, sending an email, or updating a record. Tools are what let an AI affect the world rather than only describe it.
AI agent
An AI configured to pursue a goal with a degree of independence, choosing its own actions, using tools, and continuing until the task is done, rather than waiting for an instruction at every step.
Agentic workflow
An automated process in which the AI decides the steps itself. Given a goal, it chooses what to do, responds to what it finds along the way, and adjusts its approach. It suits open-ended work that calls for judgment rather than a fixed script.
Loop
The repeating cycle an agent runs to get work done: take an action, observe the result, decide the next action, and repeat until the goal is met. “Human in the loop” means a person reviews or approves steps within that cycle.
Skills
Reusable packages of instructions and resources that teach an AI to perform a specific task to a set standard, loaded only when relevant. They let a team capture its own procedures, such as a house style, so the AI follows them consistently.
Harness
The supporting structure built around an AI model that makes it useful and safe, including the tools it can access, the rules and guardrails it must follow, and the memory it can draw on. A strong harness often matters more to results than the raw power of the model.
Agent swarm
A group of specialized AI agents coordinated by a managing layer, each handling the part of a job it does best, much like a team with a lead. Also called a multi-agent system.

The product landscape

AI assistants
The chat products most people mean when they say “AI”: ChatGPT, Claude, Gemini, Copilot. Used through a chat interface on the web or an app, answering questions and producing work turn by turn within the conversation.
ChatGPT Edu
The university-licensed version of ChatGPT. It runs under an institutional contract, and content entered there is not used to train models outside it, which is the practical difference from a free personal account. At Clemson it is the sanctioned environment for University work.
Custom GPT
A configured assistant with fixed instructions and its own knowledge files, built once inside ChatGPT and reused every cycle. The fastest way to turn a process you've mapped into a tool other people can use.
Coding agents
AI built to work inside a codebase from a terminal or editor, such as Codex or Claude Code. They read a project's files, make changes across them, run commands, and complete multi-step technical tasks.
Reasoning model
A model trained to work through a problem in explicit steps before giving an answer. It is slower and more expensive but performs better on tasks that require careful, multi-step thinking.
Connectors
Live links from an AI to outside software and data, such as mail, calendars, and file drives, through a common interface. Connectors change where your data goes, which is why organizations, Clemson included, typically require an approval request before one is enabled.
Open source vs closed models
Closed models, such as Claude, GPT, and Gemini, run on the provider's servers and are used through an account. Open source models, such as Llama, can be downloaded and run on your own hardware. The trade-off is mainly control and privacy versus convenience.
Answer engine optimization (AEO)
The practice of structuring a website's content so that AI answer engines, such as ChatGPT, Claude, Perplexity, and Google's AI Overviews, are likely to cite or reproduce it when generating an answer. AEO is to AI what SEO was to traditional search rankings: as more people begin their research inside a chat interface rather than a list of blue links, getting represented inside those answers becomes a separate discipline.

Part two

Where adoption stands today

Globally, AI is still in its first chapter. Of the roughly 8.3 billion people on Earth, about 1.4 billion have ever used AI, around 17 percent. Press play on the graphic to watch adoption build from 1950 to today.

  • Never used AI ~6.9 billion people~83%
  • Free chatbot ~1.3 billion people~16%
  • Paid subscription ~45 million people~0.5%
  • AI workforce ~28 million people~0.3%
  • Dev tools ~25 million people~0.3%
  • Vibe coding ~18 million people~0.2%
  • AI academics ~2 million people<0.1%

Each dot in the graphic is roughly 3.2 million people. 2,500 dots fill out the world’s 8.3 billion. Figures are 2026 projections; the segments above match the colored dots in the graphic.

Visualization by Damian Player, drawing on 12 named sources including the Microsoft AI Diffusion Report H2 2025, Stanford HAI AI Index 2025, OECD & LinkedIn AI Workforce Trends, the PwC 2025 Global AI Jobs Barometer, and UN World Population Prospects 2024.

Inside higher education

The campus picture moves faster than the global one. As of spring 2026, about half of US instructors touch AI weekly, students still out-use faculty everywhere measured, and most institutions are governing all of it with policies that arrived later than the behavior.

52%

of US instructors use AI at least weekly

Tyton Partners, spring 2026

25%

of faculty use it daily, the highest point since Spring 2023

Tyton Partners, spring 2026

61%

of US students use AI at least weekly

Tyton Partners, spring 2026

17%

of faculty worldwide rate themselves advanced users

Digital Education Council, 2025

1 in 3

daily AI users report a decreased workload, about 4x the rate of monthly users

Tyton Partners, Time for Class 2026

Higher-ed sources