Today’s post is simple. I want you to do this prompt after you put Claude code into an already existing folder for one of your classes. I’m going assume it is a data science class but frankly, you can make a single change to that. You can even let R be uninstalled and tex uninstalled! Won’t matter — Claude code will install them both for you. But I think once you do this, and see it with you own eyes, you’ll understand.
Also, feel free to add stuff in like “use Stata or python” instead, or tell it also about some stuff you have that you tend to rely on like notes. Tell it more about the students — is it undergrads? Grads? Majors? Percent of each class (ie 20% first year), the background (ie mostly commuter students), and so on. Emphasize iow the audience and what you know of them. Heck if you have a student pdf with their actual majors and years, ask Claude code to look at that before designing the rhetoric of the deck! But otherwise, paste this entire prompt into Claude code once you’ve put the Claude code into that local directory. It must be local as you cannot run code from GitHub cloud best I can tell.
“I want you to design for me an original Beamer style design — something truly original, aesthetically pleasing, but professional for an undergraduate course on data science. I then want you to take the first deck and restructure it. I want you to emphasize a new rhetoric of the same subject matter and key learning points but maintaining my pedagogy as you detect it. I want R scripts embedded in the decks, and also new R scripts accompanying it so that I can do walk through and provide those scripts to them. Remember to me a deck must be beautiful, it must have a consistent narrative flow that nonetheless maintains technical rigor, it must have beautiful slides with optimal cognitive density across all slides — a smooth delivery, not overloaded at the slide level, but distributed and balanced well so that slides do not become too dense — and I want beautiful figures and beautiful tables as I care about the visualization of data. And then compile it. Once you compile, then check and eliminate ALL overfill, overfill, vbox and hbox errors. No matter how small. Recompile. Then have a second agent evaluate the deck for whether these instructions were met and make adjustments based on that recommendation and criticism. I want the figures and tables to be based on R output png and tex also, so it is critical you run the code first and then be sure to have it inserted well. Always be aware of labeling issues with Tikz and graphics from ggplot. You can often easily miss the mislabeling positioning bc it will not show up as compile overfull etc errors. They are more often due to restrictions you placed inadvertently on positions and coordinate placements being wrong. Have a third agent check only the graphics for those problems including numerical accuracy. Then compile a third and last time.”
Try that. You might want to be seated when you see what happens next. Godspeed and good luck. Welcome to the 31st century.