I understand AI much better now. In my last piece, I mentioned I’d been goofing around but hadn’t gotten into any substantive creation. Two things have happened since then.
First, my friend Cal Newport broke down our current AI, how it works, and what it can and can’t do in an article in the New Yorker and then in more detail on his podcast. You can listen, but it’s better to watch because he goes all professor, and the visuals helped it make sense. Understanding what these LLM AI programs can and can’t do is definitely helping me sleep better.
Second, I decided to see how well both Bard and ChatGPT did on a screenplay adaptation of a short story.
Reader—it sucked. And I’m kinda thrilled. I shouldn’t be as happy about it as I am, to be honest. But hey, it means we creatives are still needed, at least a bit.
The result it gave me was truly awful. I’m talking base, boring, bland, lacking any uniqueness or spark of life, and trope-addled. The indefinable something (voice) that takes a basic script to something both shootable and watching was completely missing. I’ve read some incredible scripts. Their magic lies in how the screenwriter can evoke a scene in just a few sentences or words. A brilliant screenplay isn’t always enough—clearly, how it is cast, acted, and directed is essential. But it’s hard to make silk from a sow’s ear.
[[An aside:
JT: Bard, where does the phrase silk from a sow's ear come from?
Bard: The phrase "you can't make a silk purse out of a sow's ear" is a proverb that means that you cannot improve something that is inherently bad or worthless. The phrase is thought to have originated in the 16th century, and it is often used to discourage people from trying to improve something that is beyond repair.
There are a few different theories about the origin of the phrase. One theory is that it comes from the fact that it is impossible to make a purse out of a sow's ear. Sows' ears are made of coarse, thick skin, and they are not suitable for making delicate items like purses. Another theory is that the phrase comes from the fact that it is difficult to make something beautiful out of something ugly. A sow's ear is considered to be an ugly thing, and it is therefore difficult to make something beautiful out of it.
The phrase "you can't make a silk purse out of a sow's ear" is still used today, and it is a reminder that it is important to be realistic about our expectations. We should not try to improve something that is beyond repair, and we should not expect to get something for nothing.]]
Thank you, Bard. Helpful! Though I take umbrage that my story might not have been good enough to create an excellent adaptation, I don’t think there’s a direct in/out correlation here.
I fed in various commands for this particular screenplay adaptation, and none made it into what I was hoping for—a template I could use to adapt one of my stories. Looks like I’ll have to master Final Draft after all.
The other thing I tried was outlining a Taylor Jackson short. While the basic premise was…okay, it lacked any creative magic. I tried as myself, in the style of a Ron Howard movie, in the style of a David Fincher movie, and in the style of Aaron Sorkin. I was left with a few atmospheric directions and a deeper setting, but nothing that felt unique. It was a general locked room mystery with palm trees. Sometimes they bent in a breeze. Sometimes they were shadows against a dark sky. (umm…)
I started getting somewhere when I pulled out a specific scene, gave closer directions, and then asked, “why?”
“Why” made a big difference. The refined story outline was a bit more original.
Then I pushed Bard into the core question that kickstarts every good story— “What if…?”
“What if” is the topic of another essay, but adding “why” and “what if” to the queries about each plot point in the outline turned something generic into something somewhat useable. It helped refine the story I was considering.
Knowing what I now know about how the models work, I can much more easily conceptualize the program's limitations and tailor the queries. The issues of copyright infringement notwithstanding, we’re still in a very basic AI world. Will it become better? Maybe? Cal points out that ChatGPT’s founders are looking to make their AI model smaller and more eloquent rather than larger and more broadly capable, which will possibly help this specific issue.
More importantly: Will I use this outline? Meh…probably not. I do have a short story planned for Taylor and Baldwin—you’ll have to read THE WOLVES COME AT NIGHT to see what it might be (ahem 💍)—and my story idea is happily more original than anything Bard or Chat GPT could generate, even with refinements to directorial style and the more specific queries. But for idea generation, for what I don’t want to do, it was invaluable.
Why? Because if AI thinks I should zig, I should clearly zag. Just in case my fellow scribes out there are using these AI programs for idea generation and story outlines, it seems wise for me to stick to using it for research and non-fiction support rather than anything fiction-related. I know everything we do is derivative, but there’s derivative, and then there’s derivative, you know?
I found myself willingly turning to Bard several times this week. Weirdly specific research queries are its forte, for sure. An excellent historical author, Deanna Raybourn, once told me that if you need to learn about a particular topic or time, the children’s section of the bookstore or library is the first place to start because the books can distill complex historical moments into a few simple concepts.
So can Bard. Just like the example above, as it turns out, asking for research mid-paragraph gives a simple, straightforward synopsis that lets me add whatever little thing I need and keep moving forward instead of all-stopping to deep dive into what I need to know. (Don’t worry, I mark each section to double-check. Trust me, I do not believe it wholly yet.)
I still have many questions about the ethics of using AI for story generation, as there’s no way for it to generate anything that hasn’t been written by someone else because that’s how AI has learned to “think.” But for the rest? I see the value and will continue experimenting and reporting back. Many authors are happily engaged with this new technology, and I’m not there yet, but an uneasy alliance might be forming.
My question for you today: Would you ever read a book created by AI with or without an author’s guidance?
Knowing a story was written by AI would automatically turn me off. The sticky question, of course, is how would we know in the first place?
I might try reading an AI story, but I’m not sure. It might depend on the author….
Susan