Today I’m taking a break from grading final exams to eat a little lunch and think a little thought. Exam week involves a classic Wandlessian paradox: I quite like grading them, as it turns out, since they usually give me a good glimpse of the sorts of learning that stick, but I also admittedly resent them a little bit, as the middle stretch of December is usually right around the time my mind starts fixating on other stuff.

What I think I like most about the style of exam I administer is the humane engagement it involves. I used to give finals that were a little mix-and-match at the level of objective knowledge and subjective questions, but all the adjustments we made in academics to maximize opportunity at the height of the COVID era have led to more venturesome essay-driven modes of response. I know some folks are terribly fretful about the prospect of LLM-generated responses, but to my thinking (and in my mode of evaluation) the risk seems smallish.

LLM stuff has no place at all in my creative life; I view it largely as a toy rather than anything verging on some simulation of genuine intelligence. I’ve used ChatGPT and a couple of image generators, all to results I more or less expected. The former offers useful pedagogical illustrations I can bring to my classes, and it helps to get a feel for the sort of prose it spits out. Were I a more adventurous professor I’d probably challenge one of my intro classes to a race, if only to see if an earnest essayist writing a first draft composes more quickly than an LLM user who has to revise the churned prose to make it sound almost human. My bet is that the original authorship goes much more quickly and smoothly, with far fewer vestiges of robotic lameness. I once asked the program to give me a jillion-dollar screenplay idea, and it gave me the plot of Needful Things, albeit with a 2500-word screed on ethics as a bonus. There’s a bit of tragicomedy to it all, but I’d like to think we’re beginning to figure out how little it’s actually good for. I have no doubt that there’s a certain kind of business mindset that imagines profits flying everywhere from all the time saved, but the fact that creative writers are in high demand to undo the silly things that LLM programs invariably do gives me some comfort regarding those bleak visions of the future.

The image generators make the prospects for LLM content even starker. I’ve not yet been brave enough to run a program live in a class, largely because my own experiments have been so wildly various (ask for a familiar figure in a party hat, for example, and the program might neglect to render any other clothing), but there are some vivid illustrations out there in the world already about what these models do, both by design and inadvertently. There’s some incidental racism and censorship cooked into most programs right out of the box, problems that can only be remedied by aggressively training a home model, never mind the creepy mutations and other errata. The images generated, at least for me, at best can inspire a bit of uncanny-valley fascination. There’s no art in them, no humanity–just strangeness.

At one level, I kindasorta get it. We value ease, or the appearance of ease, and the idea that a machine can produce images or writing at a high standard, allowing us to forgo decades of training and practice, is not without a certain kind of tantalizing interest. But I am, at bottom, an 18th-century scholar, and I’m often reminded of Pope’s “Essay on Criticism,” when he writes: ”True ease in writing comes from art, not chance/as those move easiest who have learned to dance.”

Someday I’ll write on the ethics of all the copyright violation that has made LLM vaguely possible, but today, for the sake of my own humanity, I’d better get back to grading.