Read the narration transcript
Welcome to AI Basics from BoulevardLabs. This narration is AI-generated. The course was designed by the BoulevardLabs AI training team, led by Dr. Robert Finn. It's a free, self-paced introduction for working professionals who are new to AI. Over six short lessons, you'll learn what AI assistants actually do, how to ask for what you need, how to check the results, how to keep your organization's data safe, and how to build AI into your work. Each lesson ends with a short practice exercise and a three-question quiz. Pass all six quizzes and you can print a certificate of completion. If your organization hasn't approved an AI tool yet, ask your manager first, or use made-up, non-work content only.
Lesson one is called What AI is, and what it isn't. Once you understand how these tools produce an answer, it's much easier to know when to trust them and when to check them.
When people say AI at work today, they usually mean generative AI. These tools write, summarize and answer questions in ordinary language, and most run on a large language model, or LLM. An LLM is trained on a very large amount of text. From that text, it learns patterns, like which words tend to follow which, or how a memo is usually structured. When you type a request, it produces the response that best fits those patterns, one piece at a time. Here's the key idea for this course. The model predicts likely text. It isn't looking up an answer in a database, and on its own it isn't searching the web. It has no built-in sense of whether something is true. It produces what sounds right. Usually that's also correct. Sometimes it isn't, and it sounds just as confident either way.
Right below this section, there's an interactive that shows this in action. It starts a sentence about quarterly revenue and offers a few possible next words, each with its odds. Pick a word yourself, or let the model take the most likely one. Keep going and you'll end up with something like, revenue rose by twelve percent over last year. Now look at the number. Twelve percent wasn't chosen because it was true. It was chosen because it was a likely word in that spot. Eight percent and fifteen percent were nearly as likely. The odds shown are illustrative, but the point holds. Pause the audio here and build the sentence a few different ways.
An AI assistant is strong at work that involves shaping words and ideas. It's good at drafting a first version of an email, a job posting or a meeting agenda. It can summarize a long document or messy notes into the main points. It can turn notes into a table, or formal text into plain language. It can explain an unfamiliar term or formula, and brainstorm ten options when you're stuck on one. And it can write or fix an Excel formula, a short script or a query, which you then test.
The same design that makes it fluent also creates blind spots. It can make arithmetic mistakes, like adding a column or computing a percentage, especially in long calculations. Precise facts, like names, dates, dollar figures and quotes, can come out wrong but plausible. Its knowledge stops at a training date, unless the tool you're using is connected to search. It doesn't know your organization's policies, clients, budgets or org chart unless you give it that material or your tool is connected to it. And it rarely says I don't know, unless you ask it to flag uncertainty.
The names you're most likely to see are Microsoft 365 Copilot, ChatGPT, Claude and Gemini. Some are built into software you already use, and others are standalone chat tools. Most come in more than one version. A free consumer version and a paid business or enterprise version can have different features, data handling and administrative controls. That difference matters a great deal, and lesson four covers it. For now, the rule is simple. Use the version your organization has approved. If you aren't sure which that is, ask.
Next on the page is a worked example. It asks an assistant to explain what a three-way match is in accounts payable, in plain language, in under two hundred words, with one simple example using made-up numbers. That plays to the tool's strengths and keeps real data out, though you'd still check the explanation against your team's own procedures. Then try the exercise called Find the edges. In your approved assistant, ask it to explain a term from your field that you know well, in three sentences. Next, ask a precise factual question you can check. Finally, ask it to add up ten made-up numbers with decimals, and check the total yourself with a calculator. Compare the three answers. If you caught even one error, or felt the need to double-check something, the exercise did its job.
Let's wrap up. AI assistants predict likely text. They don't know what's true. They're strong at drafting, summarizing, restructuring and explaining. And you should use the version your organization approved, and check facts and numbers. Take the three-question quiz at the bottom of the page, then join us for lesson two, Asking well.
A text prediction engine, trained on a lot of text
When people say "AI" at work today, they usually mean generative AI: tools that write, summarize and answer questions in ordinary language. Most of these tools run on a large language model, often shortened to LLM.
An LLM is trained on a very large amount of text. From that text it learns patterns: which words tend to follow which, how a memo is usually structured, what a polite decline sounds like. When you type a request, it produces the response that best fits those patterns, one piece at a time.
That is the key idea for this whole course. The model is predicting likely text. It is not looking up an answer in a database, and on its own it is not searching the web. It has no built-in sense of whether something is true. It produces what sounds right, and most of the time what sounds right is also correct. Sometimes it is not, and it will sound just as confident either way.
Loading the interactive example…
What it does well
Because it is good with patterns in language, an AI assistant is strong at work that involves shaping words and ideas.
- Drafting: a first version of an email, a job posting, a policy summary or a meeting agenda.
- Summarizing: turning a long document or a messy set of notes into the main points.
- Restructuring: converting notes into a table, a paragraph into bullets, or formal text into plain language.
- Explaining: walking you through an unfamiliar term, process or spreadsheet formula.
- Brainstorming: giving you ten options when you are stuck on one.
- Code and formulas: writing or fixing an Excel formula, a short script or a query, which you then test.
Where it falls short
The same design that makes it fluent also creates blind spots. Knowing them up front will save you trouble.
- Arithmetic: it can make mistakes adding a column or computing a percentage, especially in long calculations.
- Precise facts: names, dates, dollar figures, citations and quotes can come out wrong but plausible.
- Recent events: its knowledge stops at a training date, unless the tool you are using is connected to search.
- Your organization: it does not know your policies, clients, budgets or org chart unless you give it that material or your tool is connected to it.
- Knowing when it is wrong: it rarely says "I don't know" unless you ask it to flag uncertainty.
The tools you will hear about
The names you are most likely to see are Microsoft 365 Copilot, ChatGPT, Claude and Gemini. They work in broadly similar ways. Some are built into software you already use, such as email, documents and spreadsheets. Others are standalone chat tools.
Most of these products come in more than one version. A free consumer version and a paid business or enterprise version of the same product can have different features, different data handling and different administrative controls. That difference matters a great deal, and Lesson 4 covers it. For now, the rule is simple: use the version your organization has approved, and if you are not sure which that is, ask.
Using AI for what it is good at
This plays to the tool's strengths: explaining a general concept in plain language. It does not ask for anything specific to your company, and the made-up numbers keep real data out of it. Still, check the explanation against your team's own procedures, which may differ from the general version.
Find the edges
- Open the AI assistant your organization has approved.
- Ask it to explain a term from your field that you already understand well, in three sentences.
- Then ask it a precise factual question you can check, such as the year a well-known law in your field was passed.
- Finally, ask it to add up a list of ten made-up numbers with decimals. Check the total yourself with a calculator.
- Compare the three answers. Note which ones you trust and which you would have to verify.
How you’ll know it worked: You should see a clear, useful explanation, a factual answer that may or may not be right, and a sum you had to confirm independently. If you caught even one error or felt the need to double-check something, the exercise did its job.
Use only non-sensitive information, in a tool your organization allows.
Key takeaways
- AI assistants predict likely text. They do not know what is true.
- They are strong at drafting, summarizing, restructuring and explaining.
- Use the version your organization approved, and check facts and numbers.





