Educating for What Comes Next

People around the world send 2.5 billion messages to ChatGPT every day. At St. Edward, students are exploring what makes those answers possible. Discover how building their own artificial intelligence—and their teachers’ exploration of AI's educational potential—is helping prepare them for what comes next.

In Anthony Mortimer’s computer science classroom, understanding artificial intelligence begins with building it.

Under Mortimer’s guidance, Henry Ashcraft ’28 and his classmates are developing a small language model—the same general type of AI behind ChatGPT. Their version runs locally, on one of St. Edward’s servers, with training text and student work staying on campus.

That gives students something a commercial chatbot cannot: a system they can change, test, and study from the inside.

For Ashcraft, that independence is part of the appeal.

I like the idea of owning the AI we use instead of relying on a big company. I also like supporting open-source software, which anyone can use and improve. -Henry Ashcraft '28

A language model learns by processing text and repeatedly predicting what comes next. During training, those predictions are compared with the actual text, and the model’s internal settings are adjusted to improve its performance. Over many repetitions, it learns patterns that allow it to generate responses to prompts.

The model taking shape at St. Edward is far smaller than commercial tools like ChatGPT. Its scale makes it possible to train on school equipment while giving students room to investigate how design decisions affect its performance.

“You don’t need a supercomputer to train an AI,” Ashcraft says. “With the right techniques to make the model smaller and more efficient, regular hardware can do a lot.”

Once the model is ready in the coming weeks, students will begin a series of experiments. Does more training text improve its responses? What does increasing the model’s size accomplish? How much can they reduce its size to make it faster before the quality of its output suffers?

Ashcraft hopes to discover “whether it can do something genuinely useful, like write code.”

Those questions turn the project into an exercise in scientific reasoning as well as programming. Students will make predictions, change variables, and evaluate results. They will weigh competing priorities—speed, computing power, and quality—and use evidence to defend their decisions.

They will also confront a fundamental characteristic of language models: generating plausible text does not guarantee an accurate answer. Testing their own system gives students a practical reason to examine AI’s output closely.

“The real learning happens when students have to explain what worked, what didn’t, and why,” says Mortimer. “If they change the model and get a better response, they need evidence that it actually improved. That pushes them to be precise in both their programming and their thinking.”

The project brings to life an idea President KC McKenna ’00 shared in last week’s reflection: knowing something is only the starting point. What matters is what students can do with that knowledge. In Mortimer’s classroom, they are putting it to work through experimentation, critical thinking, and creation.

Next week, two St. Edward educators will present on AI’s educational potential at the Alliance for Catholic Education Leadership Conference at the University of Notre Dame. Dan Humphrey, who serves as director of IT and teaches Aesthetics & Technology, and Dan Corcoran ’07, academic dean, will draw on their classroom and leadership experience to help fellow educators connect AI’s capabilities with sound instructional practice.

For Corcoran, preparing students to use AI thoughtfully starts with open conversation.

“AI is here, and many young people are already using it,” says Corcoran. “Strong relationships with our students make it possible to talk honestly about that. We want them to feel comfortable saying, ‘Here’s how I used AI, and here’s what it gave me.’ Then we can look at it together, question the response, and explore ideas that take their thinking further.”

Their session will also address practical applications, including how AI can help teachers tailor instruction, provide timely feedback, and streamline routine tasks to create more time for working directly with students. Humphrey and Corcoran will examine student privacy, academic integrity, and bias, helping educators make informed decisions about the tools they bring into their classrooms.

For Humphrey, that thoughtful approach opens the door to ambitious, creative work.

“What excites me is what students can do once they understand the tools,” says Humphrey. “They can test an idea, build something, find a problem, and try again. We want them asking, ‘Is this accurate? How could I improve it? What could I create with this?’ That’s where their knowledge and imagination really matter.”

Back in Mortimer’s classroom, those possibilities will be tested one experiment at a time. Ashcraft and his classmates will investigate what their model can do, where it falls short, and how to improve it.

The technology will keep changing. The ability to ask thoughtful questions, evaluate evidence, and turn ideas into something useful will serve these young men wherever the future takes them.