It probably can. Sort of.
First, a tale from film history. Max Sennett (along with Hal Roach) was an early pioneer of silent comedy. He produced hundreds of reels starring such greats as Charlie Chaplin, Harry Langdon, Roscoe (“Fatty”) Arbuckle, Harold Lloyd, The Keystone Cops, and W. C. Fields. According to histories of early Hollywood comedy, during gag conferences at Sennett’s studio, they sometimes seated a “wild man” among the writers. His job, reportedly, was to generate the raw, outlandish seed-ideas — what some called the “wildies.”
The wild man would sit silently for long periods, perhaps “speechless,” then mutter something like “You take this cloud…” while making vague gestures or “sketching vague shapes in the air.” The writers would pick up on that nebulous spark and, collectively, mold it into a gag or scenario. Who the wild man was, or whether there was a stable of them let loose when needed, was never specified.
It’s a wonderful anecdote, and it taps into popular conceptions of the unpredictable frontier days of early cinema.
Except that it may never have happened.
Apparently, the source for this factoid is an article by film critic James Agee, “Comedy’s Greatest Era” (Life essay / reprints, 1949). Yet this most excellent of film scholars provided no documentation — no studio records, meeting notes, personal letters, or memoirs from multiple participants — that unambiguously confirm the existence of a regular “wild man” post in Sennett’s gag conferences. Yet it’s possible Agee didn’t make it all up. Not intentionally. Most likely, he got the tale from recollections and reminiscences, rather than from contemporaneous records. Maybe it did happen once, and maybe it snowballed into “standard practice.” Who knows?
I snatched this bit of skepticism from ChatGPT. This is one of the chatbot’s most useful traits. It can perform scads of research for you. It can instantly go places that would take you hours to traverse. It has punctured a tale I’d read decades ago in some Agee anthology. So yes, it just helped my creativity in devising this article.
But can AI write short stories? Edit vidcast videos? Massage soundtracks and suggest cuts?
I believe it can. Kind of.
A year ago, I filmed a sunset timelapse on Tampa’s Riverwalk on the Hillsborough River. I must have looked extra-competent with my Fujifilm X-H2 camera mounted on a bulky tripod, because a young woman with her friends and boyfriend stopped and handed me a twenty to photograph them all together. I dutifully sent them the picture, then thought I could turn it into a little vid-drama of a fictional breakup. Compiled with experimental video effects and altered Photoshop stills, the resultant three-minute production awaits you below.
I wanted to use “Baba O’Reilly” by The Who as (sort of) a soundtrack. I know, I know. Everybody has overused this piece so much that it’s become a “moldy oldie.” But if I remastered select chunks of it, I could give it a weird & new & edgy sound. Could.
So I uploaded a temporary copy of my video to Youtube and copied links to both it and a recording of the tune. Then I asked ChatGPT to go to town.
First, the bad news. This modest request turned into a Herculean task of re-elucidation. I had to clarify my requests so often that simple suggestions became complex challenges to my aptitude. The bot didn’t know how much I knew, so it assumed I was an expert audio editor. For several suggestions, it drew me into rabbit warrens of special effects with arcane tweakings.
Chief among ChatGPT’s shortcomings is that it has no ears. It had little idea what sounds cool, so it issued cockamamie suggestions for DaVinci Resolve’s audio component. I said, “Alter the music, but recognizably,” so what did it do? It reversed one clip with a vocal track that sounded like babble, and slowed another one down to 1/9th speed. It could have just as well been an experimental track by a modern music group. It suggested other artistic faux pas as well.
More often, it revealed brazen ignorance of my current software’s screens. It told me to use parameters and settings that no longer existed, or never did. I rejected about half of its suggestions and turned the dial down (in one case, up) on others.
Now the good news. ChatGBT did end up helping, somewhat. It forced me to research edits on my own. When I got stuck, I put the bot on hold and scoured Youtube videos by experts, to find out how to apply this and that effect. My chatbot became the wild man of silent film mythology. Its spinning-out-of-control, devil-make-care, full-speed-ahead, damn-it-all suggestions became contagious, even if I didn’t use most of them. Like caresses of my inner thigh, they got me in the mood. It was worth it. But you be the judge of the result.
What about creative prose? Short stories. Memoirs. I gave it my notes for one chapter in my memoir about my sex education in the early sixties. In it I’d narrated how my classmate Diane had been disciplined by a nun in freshman year for wearing her skirt too high. Her friend objected during the harassment, showing the class a history book illustration of early Egyptian women wearing short skirts. “What,” I asked, “should be the sarcastic nun’s response?” This is what ChatGPT suggested.
‘“Thank you, Miss Kelly,” Sister said with venomous sarcasm. “I’ll be sure to consult Cleopatra next time I’m revising the dress code.”’
That’s actually pretty good (although I cut the overwritten “venomous sarcasm”). ChatGPT supplied me with a flash of humor good enough for the final draft. Alas, such gems were rare. Here’s another paragraph it suggested, the opening one.
“They say Diane Stokes has a knack for trouble, but I think she’s just got more guts than the rest of us. That day in Religion class back in ’61 proved it. I was there, two desks over, when she walked in with her skirt rolled up high enough to make Sister Corde Maria’s jaw drop.”
Three clichés in one paragraph! It tossed me other clunkers too, like hokey similes, kludgy comparisons, and fatty figures of speech that required trimming.
When queried, a chatbot can produce prose chunks, poems, images, effects, even whole videos. But are they good ones? Meh. It’s not its fault. How’s it supposed to know what’s really good, in your opinion? It doesn’t know you like it knows a cesium atom or Mussolini’s March on Rome (1922). It can’t spot clichés that irk, sentimentality that nauseates, and mawkish pictures fit only for laundromat walls. It can’t tell which words are, in Mark Twain’s quote, “more lightning bug than lightning.” Your tastes are your own and sometimes so protean you’re not sure of them yourself. AI is the nice fellow helping you search for your lost sunglasses. “They’ve gotta be here somewhere.” You asked him over, but he’s under no obligation. He doesn’t care whether you find them or not.
So go ahead. Try it out for your next Valentine’s Day ditty. Spend time making your AI prompts as elaborate and specific as possible. Watch videos on how to squeeze out the best results. Read up on its recent improvements. Maybe the day will come when it blows you over with massively clever concoctions and bull’s-eye suggestions.
Does it think so too?
Me: What's in store for LLMs in aiding writers and artists with their creativity?
ChatGPT: LLMs will get better at:
Generating nuanced ideas rather than generic suggestionsMaintaining a consistent tone and style over long projectsAdapting to your voice by learning from your past workOffering counterarguments, twists, subversions—like a real brainstorming partner
(More…)
Well. We’ll see about that.
