Meet Slushy, the agent’s assistant with agency of mind: The implications of QM

[Simon Gant]

To the extent of this observer's understanding, QM’s little boxes allow the agency to organize the slush pile for a systematic approach to evaluation. There is consumer software now that examines stories for grammar, style, readability, plagiarism, etc. Publishizer and Inkitt purport to analyze reader engagement in support of market projections. I have little doubt that publishers and big agencies are developing proprietary AI models to evaluate emotional tone, pacing, and engagement potential. AI lacks the ability to judge nuance, voice, and originality as a human would, but the field is in its infancy.

 

[Sage]

Do you have research to back up this assertion? The idea of agencies/publishers putting authors' submissions through AI is a pretty terrifying one, so it's not one to just wildly throw out accusations about.

 

[Woollybear]

Related: https://janefriedman.com/like-it-or...rial-for-ai-training-and-using-ai-themselves/

 

[Maryn]

ALL discussion of AI goes in a single place here at Absolute Write. Everybody buckle in while I move this thread to that place.

 

[Simon Gant]

Do you have research to back up this assertion? The idea of agencies/publishers putting authors' submissions through AI is a pretty terrifying one, so it's not one to just wildly throw out accusations about.

It is not an accusation, only an observation of a natural progression. Isn't it better to get out in front of it? The current situation is untenable. Think of it as a pyramid. The base of the pyramid grows and grows, better writing programs, better tools, more leisure...more writers. Not only that, but one QM box I encountered asked if AI was used to generate the story, indicating that the agents are now being swamped with volumes of worthless pap. However, the top of the pyramid is static, limited to shelf space, readers, relatively fixed, if not contracting, market forces. This fixes the number of books that get published and the agents that support them. Check out Galatea, Marlow--there are many others. Is it that far fetched for an agent to run the query through a simple Grammarly spelling check. Accordingly, why should the agent do this? Why not have an automated version of Grammarly pick up the whole stack, check spelling, check for the 'tells' of AI generation. Wouldn't that be a boon for good writers? It would be great to get an actual agent's thoughts on this.

 

[Maryn]

Actually, I expect agents I'd submit to would have the integrity not to run text they did not write through any AI or LLM program, because it's A) not theirs in the first place, B) known to scrape text, and C) immoral.

The good agents do not want to encourage the use of AI, with its vast databases existing only because they steal the works of human authors.

Maryn, pretty freakin' opinionated on this topic

 

[lizmonster]

It is not an accusation, only an observation of a natural progression.

I don’t think it’s a natural progression at all. Nearly every claim about genAI is about what it might do someday, not what it’s capable of now.

Agents don’t need AI to do keyword parsing. Companies have been doing that with resumes for years. As for narrative structure? How would that work? AIs know about word frequency and relationships, not context.

 

[Simon Gant]

ALL discussion of AI goes in a single place here at Absolute Write. Everybody buckle in while I move this thread to that place.

Gotcha. According to my search, that is Critical Theory and Philosophy of language, although this is a discussion of a publishing trend.

 

[lizmonster]

The good agents do not want to encourage the use of AI, with its vast databases existing only because they steal the works of human authors.

This. Any agency that did this would be off my list, and all over social media.

 

[Sage]

Gotcha. According to my search, that is Critical Theory and Philosophy of language, although this is a discussion of a publishing trend.

An alleged publishing trend

 

[Simon Gant]

Actually, I expect agents I'd submit to would have the integrity not to run text they did not write through any AI or LLM program, because it's A) not theirs in the first place, B) known to scrape text, and C) immoral.

The good agents do not want to encourage the use of AI, with its vast databases existing only because they steal the works of human authors.

Maryn, pretty freakin' opinionated on this topic

Don't shoot the messenger. But let me clear up a misunderstanding. The LLM's are already built. A company's use of AI, the grist it passes through the machine is absolutely private. How could it be otherwise. If a pharma is using their AI to permutate chemical reactions-that data has to be secure. And, did they steal those works you mention? You know those little terms and condition boxes we all tick when we get our 'free' facebook, google, etc. The piper is being paid.

I see QM as a step to organizing the query data for automation. How can an agent possibly go through--400 queries on some day is a number I've seen kicked around? Something has to give. AI evaluation should at least be used to identify AI generated writing. The more bad writing that can be lifted from the agent's shoulders, the better in my opinion.

 

[lorna_w]

As the minority "I like AI" voice here, I say, sounds like a good idea for a genre agent. (lit fic, maybe not). 90% of stuff being illiterate or slow or changing tenses or whatever is easy to detect, and AI doesn't demand minimum wage or get migraines. I trust AI more than some upper middle class 22 year old girl in her first job out of a private university, ttytt. We've long heard 75-90% of submissions suck, and at a pretty simple and easy to see level. I've done critiquing online over the years, and I find that to be true. It's better here, but at some places, if you find 1 in 200 that is worth reading to page 3, you've had a good day of critiquing. Analyze for grammar and correct usage, do a scan for dialog (none in 10 pages? Safe to automatically reject." Use of "f***" ten times in two pages? Again, pretty simple to reject that. Current AI is perfectly capable of that. (I've also edited a literary magazine, and I didn't find the 90% rule to be true. It wasn't awful. But it also was not good, which may be a more subtle form of judgment that LLMs are not yet up to. One day, they probably will be.)

 

[Woollybear]

Some months ago there were workshops on how to use these new shiny AI gadgets to publish more and more. I think there was even a publisher proudly stating that this was their new model.

Maybe this one.

 

[Simon Gant]

An alleged publishing trend

Observed as stated, or as you wish, alleged.

 

[Woollybear]

As the minority "I like AI" voice here, I say, sounds like a good idea for a genre agent. (lit fic, maybe not).

As an anti AI but agnostic, I'm-in-this-writing-thing-for-other-reasons, you-guys-figure-it-all-out-and-flag-me-later, contributor, I'll say that agents can lose their jobs to AI too.

90% of stuff being illiterate or slow or changing tenses or whatever is easy to detect, and AI doesn't demand minimum wage or get migraines. I trust AI more than some upper middle class 22 year old girl in her first job out of a private university, ttytt. We've long heard 75-90% of submissions suck, and at a pretty simple and easy to see level. I've done critiquing online over the years, and I find that to be true. It's better here, but at some places, if you find 1 in 200 that is worth reading to page 3, you've had a good day of critiquing.

I pretty much find the same. It's tedious. I find much the same on the bookshelves at brick and mortar, just stuff I find tiresome and predictable. It's a subjective business.

Analyze for grammar and correct usage, do a scan for dialog (none in 10 pages? Safe to automatically reject." Use of "f***" ten times in two pages? Again, pretty simple to reject that. Current AI is perfectly capable of that. (I've also edited a literary magazine, and I didn't find the 90% rule to be true. It wasn't awful. But it also was not good, which may be a more subtle form of judgment that LLMs are not yet up to. One day, they probably will be.)

I don't want AI to be an arbiter of good writing or storytelling. I like to be surprised by something different. I like that there are a hundred million different ways for the human brain to be wired. I like that we can look at a brick or a windowsill and each have a different memory or association come to mind (me: Pink Floyd/must dust). AI will probably adequately keep Netflix in movies and series some day, at least for those among us seeking our nightly fix. I'm stunned how much on Netflix I have zero interest in. Last night I watched something that was offensive as all hell, but I watched it, and by the end I was glad I had done so. Any woolly-centered algorithm would have pruned that one out of consideration, but it was nice to see a different perspective on life...

 

[Simon Gant]

As the minority "I like AI" voice here, I say, sounds like a good idea for a genre agent. (lit fic, maybe not). 90% of stuff being illiterate or slow or changing tenses or whatever is easy to detect, and AI doesn't demand minimum wage or get migraines. I trust AI more than some upper middle class 22 year old girl in her first job out of a private university, ttytt. We've long heard 75-90% of submissions suck, and at a pretty simple and easy to see level. I've done critiquing online over the years, and I find that to be true. It's better here, but at some places, if you find 1 in 200 that is worth reading to page 3, you've had a good day of critiquing. Analyze for grammar and correct usage, do a scan for dialog (none in 10 pages? Safe to automatically reject." Use of "f***" ten times in two pages? Again, pretty simple to reject that. Current AI is perfectly capable of that. (I've also edited a literary magazine, and I didn't find the 90% rule to be true. It wasn't awful. But it also was not good, which may be a more subtle form of judgment that LLMs are not yet up to. One day, they probably will be.)

Sensible.

 

[Sage]

Observed

You're an author. Use your words correctly. You have not observed this, you have hypothesized this.

 

[CMBright]

Slushy appears to be an AI platform.
Blog by Jane Friedman - AI & the Slush Pile: Lots of Experimenting but No Implementation (Yet)

In my opinion, an inherently unethical tool has no ethical uses until those ethical issues are addressed. AI/LLMs scrape data that should be protected under copywrite with no permission or compensation from those who own the material. AI/LLMs have a severe environmental impact.

I am afraid I would "get ahead" of agents or agencies who use AI to manage slush piles by striking them off my list of potential agents. I would learn to self publish with a plain cover or hire a person to do the cover art (I'm not skilled enough to draw stick figures) before I would use such agents. My opinion. Until the ethical (and possibly legal) issues are addressed, I won't knowingly touch AI/LLMs. FYI: Members are not permitted to share AI text or images (such as for cover art) at AW.

 

[Simon Gant]

You're an author. Use your words correctly. You have not observed this, you have hypothesized this.

Pedantry is not my thing. These are observations that lend to my conclusion; I never stated otherwise. A noisy carriage without a horse rattled by the other day and upset Old Dan. There, there, I said and gathered the rein. It's only a passing craze. Tell it to the horse.

 

[lizmonster]

And, did they steal those works you mention?

They stole my works. There's nothing in my HarperCollins contract allowing Meta's AI to digest and regurgitate my books, and yet they have.

This is an already-litigated argument. GenAI is using work without the creator's permission. Whether people want to consider this "fair use" is their business, but so far the courts haven't gone with that.

 

[pdblake]

The slushy link

Slushy appears to be an AI platform.
Blog by Jane Friedman - AI & the Slush Pile: Lots of Experimenting but No Implementation (Yet)

In my opinion, an inherently unethical tool has no ethical uses until those ethical issues are addressed. AI/LLMs scrape data that should be protected under copywrite with no permission or compensation from those who own the material. AI/LLMs have a severe environmental impact.

I am afraid I would "get ahead" of agents or agencies who use AI to manage slush piles by striking them off my list of potential agents. I would learn to self publish with a plain cover or hire a person to do the cover art (I'm not skilled enough to draw stick figures) before I would use such agents. My opinion. Until the ethical (and possibly legal) issues are addressed, I won't knowingly touch AI/LLMs. FYI: Members are not permitted to share AI text or images (such as for cover art) at AW.

That slushy link seems to be about an AI adult entertainment thingy.

 

[lizmonster]

As the minority "I like AI" voice here, I say, sounds like a good idea for a genre agent. (lit fic, maybe not).

Feels a little insulting, tbh.

 

[lorna_w]

I'm a genre writer. I like the genres as a reader. But there is subtlety in lit fic that would be harder to parse for AI at its current state. One day, probably it will work.

 

[lizmonster]

I'm a genre writer. I like the genres as a reader. But there is subtlety in lit fic that would be harder to parse for AI at its current state. One day, probably it will work.

Litfic doesn't have a corner on subtlety.

There is subtlety in all work that AI can't parse, because it doesn't parse for meaning, it parses for commonality. Any "parsing" it does of someone's work will indicate how closely that work hews to what autocomplete would have written. That is all.

 

[kinokonoronin]

AI doesn't demand minimum wage or get migraines

I beg you to follow this idea to its logical conclusion.