Robin Sloan
main newsletter
November 2022

Author’s note

This edi­tion is a bit shorter than October’s. There’s a link below to a new short story — yes, another one! — and I do some­times real­ize it’s a lot to say “I invite you to read this short story” and then fol­low it imme­di­ately with “I invite you to ALSO read this block­buster newslet­ter.”

So, I’ll try to focus on just link­ing to the story … but this is my oppor­tu­nity to add a bit of color about its creation, and I can’t pass that up.

The dance of death: the death blow, 1816, Thomas Rowlandson
The dance of death: the death blow, 1816, Thomas Rowlandson

My new short story was com­mis­sioned by Google and pro­duced using a new AI-powered edi­tor called Word­craft. Titled Author’s Note, it is pre­sented along­side new sto­ries from sev­eral other very impres­sive writ­ers.

I’ll say more about this project below, but first, I want to present the com­plete list of sto­ries I’ve pub­lished in the past year:

There’s also the new edi­tions of Mr. Penumbra’s 24-Hour Book­store and Sourdough, both with their pre­quel sto­ries. All in all, a pretty solid set of offerings!

While I’m men­tion­ing those new edi­tions, the ones with bonus material, I want to apol­o­gize for being so parochial in their promotion. “Hooray, Sourdough’s pre­quel story is now available!” announces Sloan — in his newslet­ter received by many thou­sands of peo­ple out­side the U.S. to whom it is not, in fact, available.

The U.S. edi­tions, and my rela­tion­ship with my U.S. publisher, are so front-and-center for me that it’s easy to for­get about the global picture; but I really shouldn’t, and, in the future, I won’t.

Onward, to my … 

Notes on writing with Google’s AI

I’m going to dive into some detail here; it’s prob­a­bly not for every­one, so if this sub­ject doesn’t inter­est you, feel free to skip mer­rily ahead.

There is a ton of con­text about the over­all Word­craft project on the web­site; I really encour­age any­one inter­ested in the inter­sec­tion of AI and art to spend a bit of time reading. These tech­nolo­gies are now mature enough that we can begin to move from spec­u­lat­ing about their uses to … actually using them! And not only to make cool demos, but to do real work.

That last part, the “real work”, is impor­tant to me. I am done, per­sonally, with the genre of “I see what you did there”—of mak­ing things that are note­wor­thy pri­mar­ily for their appli­ca­tion of a frothy new tech­nol­ogy. That’s really say­ing some­thing, because I have made a LOT of things like that! I spent sev­eral years engaged in AI writ­ing explo­rations of my own, almost entirely in this spirit.

Approaching this project, I felt com­mit­ted to writ­ing a story that could stand on its own; a story that achieved the same things I want ANY of my sto­ries to achieve; a story to which the response might be not, “I see what you did there”, but: “I loved this!”

Whether I succeeded, you can judge for yourself. For my part, I’d hap­pily have pub­lished this story on my web­site, or dis­trib­uted it as a zine, with no men­tion of AI tools at all.

Okay, but what about those AI tools?

Over on the web­site, you can see some screenshots of the Word­craft edi­tor and read more about its affordances. It oper­ates basi­cally like a super-upgraded ver­sion of my AI text com­ple­tion plugin from way back in 2016. Every­thing about it is much more capable, but the work­flow is the same. You can, at any point, ask the AI to gen­er­ate new text, and when it does, its sug­ges­tions reflect the surrounding draft.

The Word­craft team didn’t require or even sug­gest that my story be writ­ten in any particular way, or that it con­tain any par­tic­u­lar frac­tion of AI-gen­er­ated text. My brief was: write what you want to write, the way you want to write it.

Okay, then! After noodling for a while, I decided I’d try to write some kind of locked-room mystery, because (a) it’s a genre I enjoy, that (b) I’ve never tried before, and (c) I liked the idea of giv­ing the AI the role of the detective — perhaps chal­leng­ing it to “deduce” things based on clues in the text. In my imagination, this would be really fun! I would, in a sense, play the criminal, while the AI would play the stolid but savvy detec­tive, and if its reac­tions were a lit­tle weird, well, that might be even better … 

It didn’t work at all. Absolute crash and burn. I had over­es­ti­mated the model’s pow­ers. Para­graphs into my nascent mystery, it had no idea what was happening; or at least, not enough of an idea to “participate” in any mean­ing­ful way. It knew I was writ­ing a mystery; it reached for the appro­pri­ate terms and tropes; but that wasn’t any­where close to being sufficient.

So I retreated, a bit dejected. I kept a log of my experiences, for the ben­e­fit of the Wordcraft team. A few notes from that log made their way into the team’s paper, which I think is a qui­etly path­break­ing contribution: an almost ethno­graphic study of real writ­ers doing real work with this real AI tool. Hype, this ain’t! The Word­craft team has demon­strated clear-eyed courage in syn­the­siz­ing and pre­sent­ing our expe­ri­ences, many of which con­tain some ver­sion of my own retreat and dejection.

For a cou­ple of weeks, I thought about other genres. On a medium-long drive, an open­ing line occurred to me. Beside a pool in Fresno, a story took shape. Then, I cracked it, by estab­lish­ing a fenced-in space for the AI to con­tribute what it could.

This AI, like many of its predecessors, is very good at “riffing”: given an example, it can gen­er­ate lots more, often with impres­sive fidelity to genre and vibe. This new story called for a litany of grisly deaths, so, I asked the AI to dream up grisly deaths. I asked it, also, to tell me the names of the dead: char­ac­ters from a fan­tasy saga, cliched in the best pos­si­ble way. Red meat! This assignment, it could handle.

Now, when you read the story, you’ll know that the litany of deaths is mostly the work of the AI. I prompted and prodded; I curated and edited; I smoothed and re-arranged. But “Captain Cirrus, tram­pled by a hippopotamus”? That’s 21st-century tech­nol­ogy in its fullest bloom bring­ing you that!

Three thoughts, to close.

First, I’m impressed as hell by the Word­craft team. Daphne Ippolito, Ann Yuan, Andy Coenen, Sehmon Burnam, and their col­leagues engi­neered an impres­sive, provoca­tive writing tool, but/and, more impor­tantly, they inves­ti­gated its use with sen­si­tiv­ity and courage. This puts them leagues ahead of nearly every­one else out there ped­dling AI tools. (Indeed, I wish I’d been this cir­cum­spect back when I was yakking out my own AI writ­ing explo­rations, sev­eral years ago.)

Second, I’m proud of my story! I love its George R. R. Martin-alike narrator; I love its vil­lain and his scheme. And, yes, I love the deaths, con­jured glee­fully by a com­puter that’s read it all.

But, third: I have to report that the AI did not make a use­ful or pleas­ant writing partner. Even a state-of-the-art lan­guage model can­not presently “understand” what a fic­tion writer is trying to accomplish in an evolv­ing draft. That’s not unreasonable; often, the writer doesn’t know exactly what they’re try­ing to accom­plish! Often, they are writ­ing to find out.

Will the AI get there, someday? Will I be able to recruit my detec­tive at last? I understand very well the cur­rent pace of improve­ment in this field, and even so … I don’t see it com­ing anytime soon.

There’s more than a lit­tle meta-textuality in Author’s Note; it is surely no coin­ci­dence that the assistant, in the end, gets dunked in the pool.

I don’t think I’ve ever writ­ten about my the­ory of short stories, even though I talk about it a lot. It’s very simple: I think all good short sto­ries are, in one way or another, about death. That’s not true of novels; good nov­els can be about lots of things. But a good short story is always basi­cally a memento mori.

This is one of the rea­sons Bullet in the Brain is the best short story.

I won’t defend my the­ory — it is silly, and I can­not be bothered — but, for my part, I really do believe it.

Remember, I come by my inter­est in AI honestly; I’ve been think­ing about this stuff since 1996. Here’s the title card for my “documentary” on AI, pro­duced at Troy Athens High School using a VHS editing rig:

Robin's AI doc on YouTube, mostly VHS fuzz at this point
Robin's AI doc on YouTube, mostly VHS fuzz at this point

View the whole thing on YouTube, if you dare!

The dance of death: death and time, 1817, Thomas Rowlandson
The dance of death: death and time, 1817, Thomas Rowlandson

A few notes and links — not too many this time.

Where Twit­ter’s dis­in­te­gra­tion is concerned, I stand by my pre­vi­ous assessment.

“It was ter­ri­bly cre­ative and demanding, because you’re mak­ing design deci­sions at every moment. So it was high design.” Who wouldn’t want to crawl under­neath a ta­ble and work on this? Fabulous.

Here is some very nice banjo-play­ing by Rob Stenson, who I know mainly as the cre­ator of Coldtype, a library that pow­ers cool text ani­ma­tions.

Here is the moon, “melting” behind the pres­sure waves blast­ing out the end of a rocket. Here, the effect is win­ningly explained by Scott Manley.

Here’s a beau­ti­ful paean to a Stu­dio Ghibli-adjacent movie I’d never heard about, from the great Ani­ma­tion Obses­sive newslet­ter. They are really doing some­thing unique and valu­able over there.

The movie, Princess Arete, is available to watch for free in many regions until Decem­ber 2. Here’s the clip that Ani­ma­tion Obses­sive shared:

Tap or click to unmute.

Guillermo del Toro, as ever an evan­ge­list for imperfection:

“I wanted to return the con­trols of animation to the animators, and treat [the animators] as actors,” del Toro said. He encour­aged his team to make “failed acts”—the tiny stum­bles and imper­fec­tions that make a per­for­mance by a live actor feel real — and resist what he called the “codification of ani­ma­tion into a ‘cool’ lan­guage that is almost like emojis.”

The best.

I thought I knew things about distribution

I am not really involved in the video game industry, but I’m nev­er­the­less an avid reader of the GameDis­cov­erCo newslet­ter by Simon Carless, because I find his lens so inspiring.

Simon doesn’t write about pro­gram­ming games; he doesn’t write about play­ing them. He writes about the space in between. How does a per­son FIND a video game? What makes them decide to pur­chase and play it? What buttons, levers, and warp zones influ­ence that process?

“Discoverability” is a good 21st-century word. Most of the dig­i­tal dreams of my cohort, the par­ti­sans of the open web, ran aground on this reality. It’s not enough to make some­thing and post it online; you must also inject it into some channel that will carry it to peo­ple. The web itself doesn’t do that; you need an extra layer, some reser­voir of atten­tion and/or curiosity, whether it’s Google, the blo­gosphere (RIP), Stum­ble­Upon (RIP), Twit­ter (RIP) … hmm, there seem to be a lot of dead chan­nels out here.

Back in the 2000s, I thought I knew things about dis­tri­b­u­tion, about atten­tion and networks — but I didn’t really.

It was, honestly, the expe­ri­ence of pub­lish­ing a book with FSG that showed me what dis­tri­b­u­tion really looks like, and taught me that you just can­not be start­ing from scratch every time. You need sup­ply chains — not only (or even pri­mar­ily) physical, but com­mer­cial and intellectual. Emotional, even.

There is discoverability in books, though of course it’s totally different from discoverability in games. Dif­fer­ent storefronts, different algo­rithms, dif­fer­ent tastemakers. I’d like to think this newslet­ter con­tributes to book dis­cov­er­abil­ity, in a small way, from time to time.

I’d say most peo­ple in both industries, even the professionals, mud­dle through discoverability, hop­ing for the best, some­times get­ting it. There are new kinds of pro­fes­sion­als else­where who won’t take any chances. The crass video pro­duc­ers who avidly game YouTube’s algorithm, surf­ing its sub­tlest con­tours to mil­lions of views — oof, they know dis­cov­er­abil­ity.

What I admire about Simon is that he knows it, too, but he is never crass, not for a second. His rai­son d’être is the sup­port of inter­esting, inde­pen­dent video games. He crunches the num­bers and mon­i­tors the algo­rithms on behalf of, no kidding, art.

There really ought to be a dis­cov­er­abil­ity newslet­ter of this qual­ity for every medium, in every marketplace.

The dance of death: death finds an author writing his life, 1827, Edward Hull
The dance of death: death finds an author writing his life, 1827, Edward Hull

This is my sea­son for driving, as I haul tons of olives from grove to mill for Fat Gold, the small olive oil com­pany I help operate. I cover a lot of ground — it’s a real Cal­i­for­nia odyssey, every year — and I’m grate­ful for the oppor­tu­nity. It’s good think­ing time (you should see my voice-to-text notes, dozens every day) and there’s always some­thing new to see.

This year, I have observed

In the cou­ple of days I waited to send this newslet­ter, the rain arrived. Clouds came scud­ding in from the west, won­der­ful scratchy tufts. Every­where I drove, I could see a sheet of gray out one eye and a patch of blue out another; the rain never fell where there wasn’t sun shin­ing at the same time. That’s rain­bow weather, of course, and they floated all over, not full arcs but lit­tle scraps of refracted color pinned to the air. On Monday, I expe­ri­enced one of my nicest Cal­i­for­nia days ever in my eigh­teen years liv­ing here, as I trundled along, water above and water below, gog­gling at the show.

From the San Joaquin Valley,


P.S. You’ll receive my next newslet­ter on Decem­ber 7. Before then, you’ll receive my annual gift guide, some­time in late November.

November 2022