(a) In their work
The dust has not yet settled on the role that generative AI (genAI) and large language models will have in research and software development and writing. But one thing is clear: the person who wields those tools should understand their work before delegating it to genAI.
Interns at the Institute for Computing in Research need to learn to understand their tools: they will develop their own software architectures with help from their mentors and colleagues at the Institute. Occasionally a few lines of code might come from a search that invokes genAI assistance. When that happens they must give appropriate credit in code comments, as they will do for any human assistance they get.
English language materials, like reports and slides, will not use genAI at all.
When interns accept their job offer we will ask them to write a 3-line “AI pledge” of their own crafting.
Since our interns are paid to do a research job, which involves creating new knowledge, this is crucial. Interns who violate this prescription and their pledge might be subject to early termination and even to not receiving their stipends, since they will not have done the research work.
An example of an AI pledge is by noted researcher Julia Turc at https://diffusion.fyi/about (archived here):
The time you spend reading my content is just as valuable as my time writing it. So I will never feed you AI slop. For transparency, here is how I use AI:
- Researching and understanding
- Writing animation code
- Final pass to refine or compress my (hand-written) text
Of course, as often occurs, there will be exceptions. If a project is about large language models and genAI then student and mentor will use such models. In that case the Institute’s usual position on the use of free/open-source (FOSS) software will apply, with an added proviso that the dust has not yet settled around how genAI relates to the principles of free/open-source software.
(b) In their application
Resumes and cover letters and recommendation letters generated with a large language model will not be considered. We (real humans) read these, so we put our own time into doing so – do not waste it! Just write brief statements “from the heart” and we will be much happier than if you give us AI slop.
Note that if your teacher uses a large language model to write a recommendation letter it is not your fault, so it will not reflect negatively on you. Still, we recommend that you get letters from teachers who will write a quick sincere paragraph about you, rather than wasting our time with a dense page of generic superlatives (which is what “AI slop” recommendation letters look like).