AI Use Statement

Large language models (LLMs) are increasingly used within academia and in many other fields. Below are the guidelines through which I govern my own use of these tools, and how I seek to encourage my students and trainees to do so as well:

Generating text

I will never use LLMs to generate first drafts of any text, whether an email or a research proposal. The writing process is in and of itself an exercise in perseverance and limitless practice, and not one that can be supplemented without losing a key element of what makes me love my work. I will, however, use LLMs to help reduce text to fit within character limits, or to reformat a document (e.g., reference lists, overall styling, margin adjustments). I will never copy and paste directly from an LLM without personal editing.

Writing code

I judiciously use LLMs when writing code, with varying degrees of reliance. I do statistical coding primarily in R and was trained to do so before the proliferation of LLMs. These skills are quickly being replaced by automated programs. I still enjoy writing statistical code myself, but I’ve become less reliant on dozens of Stack Overflow tabs in my browser and more reliant on LLMs to answer questions and generate snippets as needed. My general rule is: if I can hand-code an alternative test to see if my LLM-generated code is indeed correct, then I am comfortable to rely on the model. If, however, I am conducting inferential statistics or running a simulation, for example, I will rely primarily on the input from my colleagues who are trained in these fields.

Gone are the days, however, where I spend hours adjusting a ggplot2 figure to be publication-ready. LLMs have saved me heaps of time formatting visualizations and creating interactive dashboards and reports. Likewise, I continue to lean heavily on LLMs to provide formatting (e.g., CSS and HTML) for this website.

Finally, I find LLMs incredibly helpful translation machines. For example, if a collaborator sends me Python code, I typically need LLMs to explain some sections.

Conducting Research

I will never rely on LLMs to conduct foundational literature reviews or summarize research papers that I have not already read myself. This, like writing, is an indispensable act of good science. I will absolutely use LLMs to condense my own notes into cohesive summaries for future reference, and I will allow LLMs to make tables of research findings from papers I have already read.

I am currently exploring the value of using LLMs as a forced critic of my work, providing my own arguments as prompts from which I ask it to argue opposite viewpoints. More on this later…