A Byte-Sized History of Technology at Stanford GSB

Over the past two years, artificial intelligence has come crashing into the classroom at Stanford GSB.

“My life was made a lot more interesting by having to deal with this,” jokes Dan Iancu, an associate professor of operations, information, and technology, describing the pedagogical challenges created by AI tools, specifically large language models (LLMs).

In highly technical courses like Advanced Optimization and Simulation Modeling, Iancu aims to equip MBA students with the know-how to determine when to turn to an LLM for an immediate answer as well as the savvy to apply its power thoughtfully and effectively.

“What’s important is to develop a process where students still learn the fundamental concepts, which is a way to think about problems broadly,” Iancu explains. “Whatever the problem is, how do you quantify the important things and estimate these from data to answer the question or solve the problem?”

Iancu thinks of LLMs as a workflow, which he embeds in course elements such as coding and modeling. The programming language Python, for example, “was a bottleneck for some of our students, and it was something one had to understand in order to be able to move to more interesting concepts in class — how to capture fairness, for example. But LLMs are really good at Python, and one can now outsource some of this work.”

While Iancu still devotes class time to teaching Python, his students can now work with an LLM to write code, a shift that requires them to develop prompting skills and the ability to assess the results AI delivers. “To be able to verify that what they’re getting from the tool is correct, they have to understand the coding and the model,” he says. “I encourage them to not immediately go to the tool but to think first on their own, because I want to make sure that they have the correct model and are solving the right problem in the right way. I want to make sure students actually engage critically with the LLM.”

For students who are struggling to keep up, LLMs are no replacement for working through problems with an instructor. They’re also no substitute for conceptual mastery, and Iancu says evaluating comprehension remains a moving target. “Ideally, I’d assess students without access to the tools, because you just want to test the concepts. But that feels a little duplicitous because it doesn’t emulate the real world, and whenever I teach, I want to be as close as possible to actual use cases. I don’t think I have a full recipe just yet.”

Iancu understands why some of his colleagues prohibit the use of LLMs in their classrooms, a policy he believes will remain feasible in the short term. “It’s not trivial how to strike the right balance,” he adds. “But students have access to these tools and they are becoming more powerful, so if we don’t allow the tools, we’re adding guardrails and making them less productive than they could be.”

It’s an equilibrium that may prove increasingly difficult to maintain as the technology continues its rapid progress. “If you’d asked me five years ago where we’d be today, I would’ve said this was thirty years out.” — Michael McDowell