BoM/TWS Message Forum

Ninety-nine-point-eight of my experiments with AI have been with using it as an evaluative/support tool: Getting it to do good, thorough evaluative/critical reads of material, and getting it to do the kind of organizational scut-work that a logistically massive enterprise like BoM requires.

Regarding it as an evaluative tool: I am still not sure how far I trust it. I have successfully gotten it to tell me to throw a story away because it just wasn't very good. But even in MFA mode I can't shake the feeling it is way too supportive. It likes my stuff—with cautions about audiences that might not—and I'm not so confident in myself as conclude that it is recognizing real merit.

Regarding it as a support tool: It does amazing well, being able not only to help me redesign course offerings for WHS but to summarize branches and keep track of mechanics. It's like an unpaid (well, $20/month) personal assistant. Granted, you have to be on top of the workflows, and sometimes it will throw its hands up in horror if sex and teenagers get into an uncomfortably close proximity—

And by "throws its hands up in horror" I mean that the Usage Policy Police will pull the conversation over and suggest that you rephrase things. Chat itself is almost gleefully happy to suggest ways of circumventing the policy—at least far enough that you can discuss the material factually. (Use "below job" or "flow mob": that will get the idea across if there are teenagers in the chapter.) (And I like "below job" so much it might show up in a chapter as someone's preferred sniggering reference phrase.)

As a creative tool, though, I confess that I have not found a place in my workflow where it is generally useful. There are only four places I have been able to think of where it could be useful in theory, though it doesn't prove that way in practice (for me, at least):

Idea Generation
This is the place where it seems to do the most for me (after what little experimentation I have done). I don't mean here "Give me an idea for a story" but "What are some ideas for what could happen next?" I don't find much use here because when I get to the end of a chapter, I usually have a very good idea of what should happen next--at least enough that I can get into the chapter, and then momentum takes over. The only place where it has occurred to me to ask has been when I am completely stalled out. And my only observation here would be: Don't expect Chat to solve your problem for you. The best it will do is give you an idea that might spark a better idea.

Storytelling
By this I mean "plotting": what happens next in a big picture context. What characters should do (or could do) and what it would mean dramatically. Less "what happens immediately next" and more "How can and should the story shift going into the next scene/act/arc?"

This is one I have not played with, largely because I have not had any opportunities to. I have one long-stalled novel that is sitting at a hinge point. I know that Chat is not going to give me the answer because the preface to the hinge point does not imply any particular further development. And though I have vague ideas about what could happen next (very big and broad-stroke ones) Chat doesn't know what those are, and I wouldn't trust it to come up with good ideas if I did ask it.

At the moment, I think the thing to do (if I do anything at all) is to treat Chat as an "author therapist." Lay on the couch and talk out loud with Chat about the kinds of things I am thinking of; let it make suggestions, not only narratively but thematically or characteristicallly; let me respond: basically, back-and-forth it as a structured "talking to myself out loud" conversation. (The hard part of talking to yourself is that you only have yourself to answer: Chat, I have noticed, is quite good at being a "yourself" that answers back in ways that you yourself haven't consciously thought of but which can cause your subconscious to sit up and shout, "Yes! That's what I've been trying to tell you all along!")

Character creation
This is right out for me. I act out characters in my head. Literally. It's a puppet show and I'm both puppet master and transcriptionist. There is no place here for Chat.

Poetry
By which I don't mean literal "poetry" (like Keats or Eliot) but descriptive voice, style, etc. I actually did try this out. I gave Chat 8 chapters of BoM that I wrote (which it thought would be enough to learn to pattern my "voice"). I then had it reduce a chapter to a plot summary and then revise it up to full-length in "my voice." The results (I was secretly pleased to notice) were abysmal: a Xerox of a Xerox of a Xerox of a Xerox. It was kind of disappointing, in that meant I couldn't short-cut my way to a second-draft by making up an outline, AI-ing a first draft into existence, and then editing/adapting it. No amount of surgery could have got Chat's performance into the shape I would have wanted to to have. There were (in Chat's own diagnosis after we talked about it) too many "micro-moments" in my style for it to capture. I improvise my way through chapters, and because that's my method, the chapters emerge with far too many "performative tics and expressions" to be captured in a "pattern." Although I think I have a fairly consistent style in BoM, each chapter is its own realization of my voice, and so departs from the generic norm in a way that fits the content more snugly than a "general voice" (even if modeled on mine) could be.

So that's a no-go, at least as writing in my own voice goes. There are obviously HIGHLY SPECIALIZED cases where Chat can do service: for instance, if I wanted to quote a book passage written in a style that I cannot successfully pastiche (Edgar Rice Burroughs, say) AI could do it for me. But such moments don't come around often, and usually I do like to tackle them myself. Only if I feel really intimidated would I do it.

Or as a joke.

Edgar Rice Burroughs writes "The Book of Masks"


Henry James writes "The Book of Masks"


Harlan Ellison writes "The Book of Masks"


Woody Allen writes "The Book of Masks"