OpenAI released ChatGPT on November 30th, 2022. I knew immediately that it would change how educators. I thought I would encourage students to use ChatGPT help them write code in R for my population genetics coursein Fall 2023. Fall 2023 arrived and I didn’t know how to tell them to use it, so I didn’t even mention it. Fall 2023 was also the last time I taught in a classroom. I retired in July 2024. ChatGPT, Claude, and other large language models (LLMs) had little effect on classroom instruction before I retired.
I know from talking with colleagues and from reading the Chronicle of Higher Education, Inside Higher Ed, and Times Higher Education that college faculty are struggling to adapt to a new world. Some are embracing LLMs. More are banning them. Some are encouraging students to use them and helping them understand how to use them. More are returning to in-class assignments and blue book exams.
Maybe it’s because I’m retired, but I’m less worried about LLMs than many of my colleagues who teach. I remember the ruckus that exploded when hand calculators became common. People were sure that students would no longer be able to do arithmetic.1 But the world hasn’t fallen apart. Engineers still build bridges that don’t fall down, and pharmaceutical companies still make drugs with the right composition and concentration. No one uses a slide rule.2 We need to make a similar transition with LLMs.
Students should learn subjects, not regurgitate answers. Learning means reasoning independently, evaluating evidence and argument, and applying that knowledge in new contexts. I presume that students still learn how to add, subtract, multiply, and divide in grade school even though most of them will use calculators or spreadsheets when they do calculations later in life. In the same way, students must learn the principles of philosophy by grappling with them, not by asking ChatGPT to produce an essay about them. Education should not be collecting facts. It should be learning how to think. As instructors we’ve put too much emphasis on the products our students produce and too little on the processes they used to produce them.
We must focus on evaluating the processes. Term papers and take-home exams were once a way to do that. We were at our best when we evaluated how students reasoned, not only what they produced. Doing that in an age of LLMs requires a new approach. Let’s flip the classroom more completely.
Set aside class periods where students write papers or solve problems—with pen (or pencil) and paper. For homework have students transcribe their solutions into an LLM and prompt the LLM to critique the solution. When the student returns, have them annotate a paper copy of the LLM critique with pen (or pencil) and evaluate that.
This approach will require colleges to change scheduling practices. The class periods for problem solving will need to be longer than the typical 1-hour lecture, and it will probably be necessary to add the problem solving period to the existing lecture schedule rather than reducing the number of lectures.3
I wrote the first draft of this post before I happened across this post on Substack by Liz Dubelman. I highly recommend that you read it. Dubelman describes an approach very similar to what I describe here, but she describes it in more detail and much more effectively than I did.
- To be sure, some younger people don’t recognize non-sensical results from a calculator, spreadsheet, or statistical package, which often arise because of input errors. ↩
- But I learned to use one in college, and I still have the one my father used in college in my office. ↩
- This is an example of how scheduling practices might be changed. The details will depend on the subject and the academic calendar of the college or university. ↩