I use AI almost every day, for both personal and professional tasks.
In the early days of AI – two years ago! – people sold compilations of “prompts” – magical incantations to get the AI genie to do something useful. Fortunately, that phase didn’t last long. Today I just treat AI like an assistant. This assistant knows a lot. They just graduated with a PhD in… well, pretty much every subject. But they are flaky, unreliable, and inconsistent. They need to be given very clear instructions, and their work needs to be carefully checked.
Rather than talking in generalities, I’m going to show how I use AI for a specific creative task.
Writing Lectures with AI
I teach computer science at a university in Silicon Valley. Sometimes, when I teach a new class or need to cover a new topic, I write a lecture from scratch. In the past, that took me about six hours. With AI, I’ve reduced the time to about 2 1/2 hours.
I start with very detailed instructions for my AI assistant. I include background and context, and I state specifically what output I want:
I am teaching an undergraduate course in the computer science department at a university in Silicon Valley. The course is “Object Oriented Design,” and we use the Java programming language. Write a lecture on the topic “Encapsulation, Delegation, and Refactoring Design Principles.”
The lecture should consist of 60-70 PowerPoint Slides. (Don’t actually create the slides; give me the lecture in outline form and I’ll make the slides myself.). Each slide should have 5-8 detailed bullet points. The bullet points should be understandable sentences or phrases, not just high-level topics. For each slide, include explanatory notes for me to use in the lecture. Focus on how the design principles solve real-world programming problems.
Where appropriate, show examples in the Java language. Examples may be in “before and after” form, showing how bad code can be improved by using the design principles.
Use current best-teaching practices to engage the students.
Given those instructions, the AI will produce a pretty reasonable lecture. I then go through the lecture slide-by-slide. As I do this, I create the actual powerpoint slides.
Sometimes I just make simple changes such as rephrasing something to fit my lecture style. I may delete points that seem like unnecessary tangents. I may add points that I think are important. I may split one slide into two, or combine slides. This is pretty much just copy-editing.
Sometimes I come across something that doesn’t make sense, or something I disagree with. This may be because:
- I don’t understand the topic as well as I should
- The AI has made an error
- The AI has done a poor job of explaining the point
So I go back to my assistant:
Slide 17 is confusing. Can you explain the point you’re trying to make?
Sometimes the explanation will open up new questions. I may ask follow-on questions, or have the AI look up some references for me. Eventually, I may rewrite the questionable slide myself. Or, I may hand the task back to the AI:
OK, I think you’ve made some good points with regard to slide 17, but I think a better explanation is needed. Please expand that into 5 slides, and include a sample program in Java.
Also, as I go through the slides, I will sometime ask for illustrations to supplement the text:
Looking at the program on slide 26, please create a UML Class Diagram that clearly shows how the Delegation Principle is being used.
Or,
Slide 31 is important and I think we need to emphasize the point. Can you create a cartoon / meme to reinforce that?
In short, I treat AI as an assistant and collaborator. I can typically use about half the slides verbatim. Others require editing and clarification, and sometimes (as I’ve shown) I use AI to help me think through the changes that are needed. Occasionally I find that I don’t like the order in which things are presented, and I’ll re-arrange the slides.
After I’ve gone through the entire lecture, and created the PowerPoint slides, I upload the file to the AI with the following instructions:
I’ve uploaded a set of slides for my lecture, based on our work. Can you review the slides and let me know if you spot any errors, or things that could be improved?
Based on the feedback, I may make some final adjustments. The end result is a better lecture than I could have written myself, in less than half the time.
Related Uses
I use AI in many other ways for my classes. Here are a few more examples:
Updating Lectures
Sometimes I have a lecture that needs to be revised or updated. Here’s an example prompt:
I am teaching a graduate course in the computer science department at a university in Silicon Valley. The course is “System Architecture.” I’ve just uploaded a lecture on “The Current State of Quantum Computing.” The lecture was written two years ago, and needs to be updated to reflect recent progress. There may also be important things I missed. Can you prepare a revised lecture, following the outline of my existing lecture? Stick to my existing lecture as much as possible; limit changes to correcting incorrect information and adding new information.
The lecture should consist of 60-70 PowerPoint Slides. (Don’t actually create the slides; give me the lecture in outline form and I’ll make the slides myself.). Each slide should have 5-8 detailed bullet points. The bullet points should be understandable sentences or phrases, not just high-level topics. For each slide, include explanatory notes for me to use in the lecture.
Use current best-teaching practices to engage the students.
Once the AI creates the updated lecture, I go through the same review-and-edit cycle that I use for a new lecture. It’s much shorter and faster, because we’re starting with a good foundation.
Writing Quizzes and Tests
Preparing tests takes a lot of work. I might give a multiple-choice test at the end of a learning unit (multiple choice tests are not ideal, but can be automatically graded, which saves a lot of time). In the past, writing the test would take me 3-4 hours. With AI, I can cut that to an hour:
I am teaching an undergraduate course in the computer science department at a university in Silicon Valley. The course is “Object Oriented Design,” and we use the Java programming language. I’ve just uploaded slides for six lectures. Based on these lectures, please prepare 60 multiple choice questions that cover the material.
The questions should range in difficulty from moderate to very challenging, but they should not be “trick” questions. For each question, show the correct answer first, and then show three plausible but incorrect answers.
Also, for each question, shown an explanation justifying the correct answer.
In this example, I asked for 60 questions. I actually planned on giving a 40-question test, but I ask for more questions than I need. That way, if I don’t like some of the questions, I simply throw them out rather than trying to “fix” them.
I also use AI to create short-answer questions, essay questions, and programming problems. Those typically take a little more work, and back-and-forth, to craft. For a midterm exam, I might aim for 40 multiple choice questions, three essay questions, and three programming problems. I ask the AI to create more, so I can choose which to use.
Grading
Finally, I use AI for grading. This is a bit controversial. People are very wary of having AI grade students’ work.
It’s important to understand that in most university courses, assignments and exams are graded by a human “grader,” following a grading rubric (a standard or set of guidelines) prepared by the professor. The professor then reviews the grades before they are posted.
I am now using AI (or allowing my human grader to use AI), and then reviewing the output.
Just as with the tasks I’ve already described, using AI saves a lot of time and work. But more importantly, it allows me to give students much better feedback.
Here’a an example:
I am teaching an undergraduate course in the computer science department at a university in Silicon Valley. The course is “Object Oriented Design,” and we use the Java programming language. I’ve just uploaded an assignment that was given to the students, along with a detailed grading rubric.
I’m now going to upload the assignments the students turn in, one at a time. For each, assign a score from 0 to 100, based on the grading rubric. Show the scores for each aspect of the assignment, as specified in the rubrics.
Include detailed feedback for the student. Explain why points were deducted, and what the student could have done differently to earn full points.
I am very impressed with AI’s ability to give insightful feedback and advice. This goes beyond understanding the material and grading correctly; the comments that the AI provides are genuinely useful for students, and at least as good as I could write myself.
I’ve been doing this for two semesters. Sometimes students question the grading, and sometimes I agree with their concerns and change the scores. However, this is also true of human grading. I don’t think the number of complaints or errors has changed, and I think the feedback is much more comprehensive than what I – and my human graders – were able to provide in the past.
Using AI to Help Students Directly
As I incorporated AI into my workflow, I found that I was often using to explain things to me. If there was a topic that I didn’t fully understand, I’d ask AI to explain it to me. In the past, I would have:
- Gone through some textbooks looking for an explanation
- Done a web search
- Watched some YouTube videos
- Asked a colleague.
Now, I simply ask AI. If I’m still not clear, I engage in a back-and-forth dialog. I ask more questions. I ask for examples. I cna even ask AI to quiz me!
It’s a very effective way to learn. And I encourage my students to do the same.
“If there is anything from today’s lecture you didn’t understand, come by my office hours on Thursday. Or, you can upload the slides to ChatGPT and have it answer your questions.”
Are Students Using AI to Cheat?
Students are certainly using AI to do their assignments. And unless they are explicitly told to do that, it is cheating.
This is nothing new. Studies show that about 60% of university students cheat, and that’s been true for many years. We try to fight it. For example, I give more weight to in-class exams, and less weight to assignments. But we can’t prevent it.
At the start of each semester, I give a short talk about what students will learn in the class, and why it’s valuable. I tell the students, “These are important skills that you will need to be successful in your careers. If you pay attention in class, and you do all the assignments and exams, you will learn these skills. If you have someone else do your work, or you use AI to do your work, you won’t learn the skills, and that will hurt your career.”
University students are adults. They are free to make their own choices. I can only encourage them to make the right choices.
Is There Still a Need for Faculty?
For now, certainly. I see myself as creating the course.
- I curate the topics – I decide what’s important and what isn’t.
- I build the framework and outline for the course.
- I determine the subject and general content for each lecture.
- I present that material, and try to engage the students at a human level, oftne with personal stories and anecdotes.
- I give direction to the AI, and I review, edit, and approve its work.
I don’t know if that will be true in ten years. It’s possible that university-level courses will be entirely created and taught by AI. But for now, I am using AI – an incredibly powerful tool – to deliver better courses with less work.
Note About the AI
In this post, I’ve talked about “AI” generically.
I am currently using ChatGPT for my university work, for three reasons:
- The university has a contract so all faculty members get premium accounts
- I’m most familiar with ChatGPT
- It does a pretty good job
I’ve played with other tools, and I’m sure I could get similar results using another AI engine.
I almost always use Gemini to create illustrations, infographics, cartoons and memes. In my opinions, it’s light-years ahead of ChatGPT for effectively conveying information in graphic form.
All of this will change quickly! This represents the state of things, as I see it, in June 2026.