Slides for a faculty talk at WHU - Otto Beisheim School of Management. The talk asks how we can use AI for reproducible research. Research must be reproducible, whereas LLMs are probabilistic (i.e., the same task can return different results) and can report the same steps differently. Thus, I argue for a skill-first setup, in which a written workflow sets the steps for us and for our agents. The talk shows my current skill template, a paper audit skill as an example, and a reproducible folder structure with the rules that every AI agent must follow.
The slides are available as a PDF.
The deck is one Quarto file, slide_deck.qmd, which renders to slide_deck.pdf on the WHU beamer template. Two slides show a companion repository, the research pipeline example.
The repository is available at github.com/victorvanpelt/ai-research-talk.
ai-research-talk/
├── slide_deck.qmd the deck's source; this is the file to edit
├── slide_deck.pdf the rendered slides
├── template/
│ ├── talk_template.sty WHU beamer template
│ └── 00_whu_logo.png WHU logo that the template loads
├── images/ pictures and figures shown on the slides
│ └── folder_structure.tex figure of the reproducible folder structure
├── references/
│ ├── references.bib bibliography
│ └── apa.csl APA 7 citation style
├── _config.yml tells GitHub Pages to skip the Quarto file
├── .gitignore keeps build files out of git
└── README.md this file
git clone https://github.com/victorvanpelt/ai-research-talk.git
cd ai-research-talk
quarto render slide_deck.qmd
quarto install tinytex).Maintained by Victor van Pelt.