10/29/2026 Debra Levey Larson
Jonathan Freund was an early adopter of artificial intelligence tools, but it wasn’t until this semester that he began using AI as a teaching assistant as well, while developing and teaching a new course on astrobiology. After extensive reading and study of source material, he began using AI to organize lecture notes in topic areas. It also allowed him to rapidly create interactive demonstrations to enhance his students’ experience and their understanding of the material.
Written by Debra Levey Larson
Jonathan Freund was an early adopter of artificial intelligence tools, but it wasn’t until this semester that he began using AI as a teaching assistant as well, while developing and teaching a new course on astrobiology. After extensive reading and study of source material, he began using AI to organize lecture notes in topic areas. It also allowed him to rapidly create interactive demonstrations to enhance his students’ experience and their understanding of the material.
“In the class, we derive equations that have phenomenology related to life, its spread, factors that mediate it and so on. AI nicely exposes the phenomenology buried in the math of those equations to help with student comprehension. With my experience, I can see it in the equations, but it’s clearer to students who are just learning the concepts to see it visually and interactively,” said Freund, who is a professor and head of the Department of Aerospace Engineering in The Grainger College of Engineering, University of Illinois Urbana-Champaign.
With the context of the lecture materials, large language models --a type of AI that’s trained in text, are adept at writing computer codes that are the engines of the demonstration. Freund used it to create demos that are similar to those he has created without AI but says with AI, the demos are far more sophisticated, engaging and much faster to produce.
“With the context of the course material I’ve assembled, I just write a couple sentences with a few suggestions for the prompt. The AI model knows the analysis I'm doing because it has seen my notes. I ask it to add things that it thinks would be interesting. Then I usually refine it once or twice with a couple more prompts.”
Freund said without AI, the information might be a plot showing what each mathematical term does. Instead, students get an AI-generated interactive model with movement. For instance, they can drag a temperature slider and watch how the model changes.
“It’s also great because it runs on any browser so students can download it and play with it outside of class. I tell them that they’ll really understand when they can anticipate what moving a particular slider will do.”
As far as errors go, Freund said he hasn’t seen it get the math wrong. As he lectures, he uses a stylus to write equations by hand that project onto the large screen in the classroom. After class ends, he puts everything he wrote into the model.
“If I just prompt it with, ‘hey, here's what I presented,’ that wouldn’t be enough context. But this isn’t just a transcription tool. It maps that onto the lecture notes I prepared as well as relevant books and papers.”
“It creates a LaTeX file that covers what I presented in class. Then I give it to the students, but not before I reviewed it for errors.”
According to Freund, the AI model also provides a list of what it couldn’t read correctly in his handwriting.
“It asks, ‘did you mean this or this?’ Then I go in and accept the change or correct it. I clear them in about 10 minutes.”
One thing Freund doesn’t leave up to AI is assessing what the students have learned in his class. The only assessments are one-on-one oral exams with him, three times during the semester without using electronic devices.
“It's AI proof and it encourages students to study the material rather than getting good at doing a small class of problems on exams.”
Freund’s current class has about 20 students. He said if there were 100, he wouldn’t be able to do oral exams in this same way, but he is interested in scaling up this model.
“I’ve done oral exams for classes with up to 30 students. I don't know how to scale it larger than that. It might be that I’d need to train teaching assistants or figure out some other assessment tool.”
Freund said he thinks a lot about what AI will be capable of in the future.
“What LLMs can do now versus nine months ago is remarkable. It seems to have crossed the threshold – the depth at which it can manage information and ideas. It’s already great at taking in all this information and bringing to it some level of understanding and concepts.
“All of its information was created on the Internet by people. Does that limit what it can then extrapolate from? We’ll see.”
You can download the demos at https://uofi.box.com/s/a5uz379z1aamow476ygm4bhbqwfmg88o.