Use of LLMs (Claude, ChatGPT, etc.) in Courses

Large Language Models (LLMs) are very helpful for gathering information (like an advanced search engine) and for executing simple tasks (like a research assistant). I use them myself extensively, and this increased my productivity across a wide range of tasks. There is a caveat to their use in courses, though.

In order to learn to conduct valuable economic analyses in academia, in policy or in the private sector, it is important that students build some foundational knowledge in economics in general and econometrics in particular. How else to know what to even ask an LLM or how to interpret its responses?! Sure, even someone who’s never heard of economics could still pass on a question to it and let it explain in detail, but (i) this would – without thoughtful engagement with the response and many iterative refinements – result in an analysis of poor quality, and/or (ii) it would amount to learning the material then, with the LLM. The bottom line is that, if you want to be of any added value compared to someone without training in economics, you must eventually acquire the foundational economics knowledge that is necessary to conduct valuable empirical economic analyses (with or without the use of an LLM). That’s exactly what university coursework is for.

LLMs can get in the way of learning these foundations. Given that the topics we discuss in core econometrics courses are common knowledge (and generally rather simple in the grand scheme of things out there), LLMs can easily answer the questions in the Problem Sets (PSs) I assign. This is why I prefer to rely on exams and (individual) term papers for assessing student performance; in the former, no LLM is available, and in the latter, students have to perform an actual, in-depth economic analysis of some research question, for which they have to do the reasoning and legwork themselves (while passing on simpler tasks to an LLM). Still, I want to keep assigning PSs. They are like homework that helps students prepare for the exam or for writing a term paper (and, ultimately, for doing economic analyses later in life). Yes, you can simply copy-paste from an LLM without thinking much about the question or its response – similarly as you could always copy from other students –, but you do this at your own risk. You would give up on an opportunity to learn the skills necessary for conducting valuable economic analyses (and writing a good exam). In my courses, PSs account for only a small portion of the grade, and their grading scheme is very generous, so you won’t be punished for imperfectly solving things yourself. (Also, the grading scheme is crude and rewards conciseness, so the teaching assistant doesn’t have to read in detail through all the verbose and repetitive answers that some students and (all) LLMs tend to produce.)

I encourage students instead to use LLMs to enhance – rather than diminish – their learning experience. With an LLM, you have access to essentially all the knowledge humanity has acquired so far. Use it! Discuss with the LLM. It can be your private tutor. It’s like your professor has all the time in the world just for you! To ensure this tutoring is effective, I suggest: (i) tell the LLM about yourself (your background, your preferences on writing, coding, discussing, etc.), (ii) give it access to read your documents (papers, lecture notes, slides, etc.) into it so that it doesn’t work from its memory (which can be wrong or vague) but engages with the same material that you have access to.