I’ve been tinkering with chatgpt, Claude, & similar, for the past couple of years, mostly for self-education — how do they work, what are they good for? — and occasionally to actually try to get something out of one of them, also in a partially self-educational mode.
Before today, it hadn’t been … hugely successful.
My most annoying & story-worthy incident came when I was prepping a talk about the Establishment Clause, and I wanted to write about how the Supreme Court had effectively killed off the Lemon test, without explicitly overruling it. I had a great idea for a graphic! A little gravestone with Lemon v. Kurtzman on it …. I asked chatgpt to make this for me … and it lectured me about how it couldn’t threaten anyone with violence. We went back and forth and back and forth and I got some graphics that really did not meet the need. Anyway, I made my point in class that night about Lemon, and an extra point about gen-AI.
One of my first experiments was also trying to prompt for graphics — I wanted a mermaid with an alligator tail instead of a fish tail — and was definitely not able to get that. In retrospect, I should have just asked Freddie Baer to collage something for me, because that was definitely what I was visualizing ….
Most frequently, I’ve used these tools to generate ideas or outlines or talking points, and then used those as — well, as negative counters from which I could, shaking my head and rolling my eyes, write my own that was better. Well, more to my needs. No, better.
But! Today I actually, truly, used a gen-AI tool for something and it was helpful!
Necessary backstory: A few years ago, when my university dropped BePress & BePress simultaneously dropped their “expert gallery” or “selected works” product, I failed to export my list of 50+ talks, submissions, papers, etc. I didn’t care, because (see last post) my focus, energy, and fucks to give have been blowing out into vacuum for several years.
But, I figured out quickly that I really did not want to have to recreate that list — even though it was 10 years out of date & so therefore just a snapshot of a few years’ worth of stuff.
My colleague helpfully send me the bepress dump, which was nested folders of xml files & the original PDF uploads.
But now I was faced with my utter inability to effectively muck about with xml & re-generate the humanly readable bibliographic data.
Voilà! Chatgpt to the rescue. I had it analyze the zip file & convert it into a .ris file, then imported the whole thing into Zotero. Then I was able to easily dig out the relevant citations & add to the “writings” section here. Behold the fruits of my labor!
Now I am preening because I Learned a Thing. Not, you know, xml parsing or atom or even ris formats. But a use case for ChatGPT that was actually effective.
Leave a Reply