Introduction to AI: the hour in words

Welcome. This hour is an orientation to AI as people actually encounter it, rather than a test of technical knowledge. A name like “ChatGPT” can refer to a service, an app, a subscription, and access to different models; we will separate those pieces before comparing what the tools can do. I would like to start with your experience: have you tried an AI chatbot, and did anything surprise, help, or disappoint you? Your answers will help me choose examples that matter in this room.

Presenter note: Invite one brief response from each person if the group is small. If no one attends, answer the question from your own experience and treat this as a full rehearsal. Point to the five-part map: vendor, product, model, interface, and location.

Generative AI is good at producing a useful continuation from patterns it learned and the context we give it now. That lets it explain, draft, reorganize, compare, and combine ideas across many subjects. But producing a convincing sentence is a different act from retrieving a record or proving a claim. A system may answer from patterns, consult a web page, calculate with a tool, or mix all three. When accuracy matters, the useful question is, “What did you actually check, and can I inspect it?” This is a practical description of how to work with the tool, not a complete theory of intelligence.

Presenter note: Use one simple contrast: “What are some possible reasons sales fell?” invites generation; “What were our sales last Tuesday?” needs a record. Ask whether the audience can tell the difference from a fluent answer alone.

Several companies offer AI systems, but the company is not the product and the product is not one fixed model. OpenAI offers ChatGPT and Codex; Anthropic offers Claude and Claude Code; Google offers Gemini and NotebookLM; Microsoft places Copilot in several different products. A model is the engine that generates a response, while the interface determines how you work with it and which files or tools it can reach. A consumer subscription may give access inside an app without providing separate API usage or permission to automate it. When evaluating an offer, ask which product you are opening, which model and tools are included, what data it can access, where the work runs, and what the plan actually allows.

Presenter note: Return to the five-part map rather than ranking companies. Use only one or two familiar examples. Product names and plans change; avoid promising a feature or price from memory.

A chatbot can be a remarkably broad partner for first drafts: it can translate a dense document into plain language, turn scattered notes into an outline, compare options, suggest questions, and help you understand code or an image. The result improves when you give it a real objective, the audience, relevant context, constraints, and a way to review the answer. Its fluency does not guarantee truth, current knowledge, or good judgment, and it cannot take responsibility for a decision you make. Use it freely to explore; verify important claims against inspectable sources and keep consequential choices under human control.

Presenter note: Ask which task in the list someone would try tomorrow. If time and network allow, use the prepared two-chatbot prompt here to show that answers vary. Have screenshots ready; do not spend the segment signing in.

A browser chatbot usually begins with a conversation: it sees your messages, uploads, and any services you explicitly connect. A terminal or coding agent begins inside a working environment where it may be allowed to read files, run commands, edit a document, and test the result. Both can use powerful cloud models; the distinction here is their reach and the kind of work they can complete. For example, one can advise you how to improve a file, while the other can inspect the file, save a proposed revision, and show the exact difference. Greater reach makes the review of scope, permissions, and changes more important.

Presenter note: Use the short, nonprivate sample file prepared for Demo 2. Keep the original intact and show a before/after or diff. If the agent takes too long, show the prepared result and move on.

Cloud and local describe where computation and data processing happen, not whether you use a browser or terminal. Cloud services are easy to start and often provide the most capable models, but they bring provider dependence, account rules, and data handling choices. A local model can run on your own hardware and may work offline, though capability, setup, maintenance, electricity, and hardware cost become your responsibility. Many useful arrangements mix the two: local files and tools with a cloud model, for example. To judge privacy or cost, trace the whole workflow, including search, uploads, extensions, and any connected APIs.

Presenter note: Ask, “If I use a terminal on my laptop to call a cloud model, is that local AI?” The answer is that the interface is local while the model computation is remote. This tests whether the two axes are clear.

The demonstrations bring the distinctions together. The same question sent to two chatbots reveals different assumptions and answers; a real file shows the difference between advice and a reviewable change; and the completed “Why I Email Myself” campaign in PubDesk shows a larger workflow with drafts, platform versions, publication status, and evidence. None of these examples asks the audience to declare a winning vendor. Look instead for what context the system had, what action it could take, what can be inspected afterward, and where a person kept control.

Presenter note: Seven minutes is only enough for a quick guided recap. Reuse the earlier chatbot and file demonstrations, then spend under four minutes on the PubDesk preview. Keep all three screenshots available. Avoid live logins or publishing.

Today’s map is meant to make the next step easier. Once an AI can work with files, browsers, and connected workflows, the question changes from “Can it answer?” to “How should we direct, check, and authorize its work?” That is the focus of Seminar 2, “From Chatbot to Collaborator.” Before we leave, I would like to know which distinction became clearer, what still feels confusing or unbelievable, and which demonstration you would want to examine more closely. Those answers will shape the next session.

Presenter note: End with the five takeaways rather than another product tour. Record the audience’s own wording, actual timing, and any question you could not answer confidently.