Clarification regarding AI policy

[Zaru]

As far as QQ is concerned, go for it. We have no rules around that.

Click to shrink...

Okay cool. Was just asking on clarification. Thanks Chief

 

[TheHoneyPot]

In my opinion its fine to use A.I. as a tool, but I really hate it when I encounter lazy 'authors' who spam A.I. slob without bothering to check it's content over for things like consistency, continuity, contradictions, and continued incorrect data.

I am already seeking walled gardens online to avoid A.I. slob...

Spoiler: Or ninja reposters

(or people just reposting the same as another within a few days of another on places like WN without checking with the actual author and then having the balls to hooking their own patreon/donation stuff up to it by claiming they translated it like the 10 others earlier that same week [even when its a perfectly fine written in english already story])

Hell, I generally am not fond of the implications it has in the job market having already been severely affected by it but I understand and respect its use as a tool but not as a replacement for any cognitive thinking on the users part.

Its very annoying to read a story where the author starts out strong only to lose interests and just use A.I. prompts to write it for them without even minimal checks or editing to make it work.

One of my loves for QQ is that it has the option to just stop seeing stuff from an poster all together. Which is great in this new era we find ourselves in. I wish more fanfic communities had this option. If its not tagged, I just use the ignore function if its absolute slob being posted by someone.

That avoids the hope of false alerts for a good new story. If it's tagged, I tend to atleast hope that its someone who cares enough to keep a creative hand in the process.

 

[BreezyWheeze]

Rules don't have to be perfect to be impactful. There are many real world examples of rules, guidelines, international treaties even that have no meaningful enforcement mechanism, yet they vastly impact behavior.

Returning your shopping cart to the appropriate spot is not required. There's no fine or prison time if you fail to do so. AND YET, in places where there's large, clear signs and easy-to-access spots to return your cart, people are more likely to do so.

Default choices are EXTREMELY powerful, even when theres no enforcement mechanism tied to them. The classic example is organ donors - when you have to opt out (the default is that your are an organ donor) the rate at which needy patients are able to get donor organs is radically higher than places that use an opt-in system.

All this hand-wringing about "theres no perfect enforcement mechanism" or "theres no flawless detection" is a bunch of absolute BS.

You could radically cut down on unflagged AI slop by simply making a default system where authors are required to select the level of AI usage - a simple dropdown that says "AI used to generate most/all text - AI used to generate some text - AI used to edit or beta read - AI not used at all" and have the dropdown be required when starting a new thread would go a LONG way, even if there were literally no rule or enforcement associated with the tag.

 

[Zanderion]

All this hand-wringing about "theres no perfect enforcement mechanism" or "theres no flawless detection" is a bunch of absolute BS.

Click to shrink...

Notably here your next suggestion don't actually solve any the issues mentioned before. Fair being fair, there could be some merit in having it listed as option in making new threads. But that doesn't actually answer any issues with enforcement or detection. It instead a sideways approach of looking for solution, an attempt to avoid the issues rather that solving it. So, the issue with enforcement and detection is in fact, not bullshit at all.

EDIT: also note that those issues were pointed out in the first place because people demanded tagging being mandatory and enforced. If people were simply asking for a means to encourage users to tag their stories but without enforcement, those issues were not going to come up. That is to say, if there's 'absolute BS' here, its more on the request than the objection.

You could radically cut down on unflagged AI slop by simply making a default system where authors are required to select the level of AI usage - a simple dropdown that says "AI used to generate most/all text - AI used to generate some text - AI used to edit or beta read - AI not used at all" and have the dropdown be required when starting a new thread would go a LONG way, even if there were literally no rule or enforcement associated with the tag.

Click to shrink...

OK. Here's a very simple way to have this be implemented: You -and this not a generaly 'you', I really mean you, BreezyWheeze here, or at least anyone who support your idea- hire some programmers to code this as a Xenforo add on. Then offer QQ a free (perpetual) lisence to use the add on. Better still if you release it for free for anyone to use. Make sure to provide troubleshooting service for add on conflict or whatever bugs that may arise

This doesn't guarantee that QQ will implement the add on, there could be other qualifiers to consider. Among other things, implementing an add on is not necessarily straightforward - there could be various error arising in implementation due to bugs, conflict with other add ons, etc. But when a feature is available and freely so, the likelyhood of a website to implement it rise sharply compared to if the site have to code/comission it themself.

Before anyone ask 'how hard it could be to code a simple drop down menu?'. Well, if its really easy to do it surely its not hard at all for someone to provide this to QQ free of charge?

 

[Koushikb8768]

Yeah, those patterns are common in autistic authors and experienced authors and doubly so with authors that fit both catagories. If you're not aware, QQ and the FanFiction community as a whole is extremely more likely to be neurodivergent than most other hobbies.

Click to shrink...

People go on about Oxford lists of three, as though the rule of 3 hasn't been a valuable writing principle for decades. It's the same with the Em Dash, people now just instantly assume anything with it is AI. The reason Chat GPT uses it is because it is a very helpful piece of punctuation used very commonly in academic papers and writing which it has been trained on. But nah it makes more sense that all the academics of the world are just using chatgpt to write for them.

Although yes the It's not this, it's that can be a dead giveaway for AI if used too often.

someone who only repost a fully fledged unedited AI story is not a writer, they are a prompt engineer.

Click to shrink...

In my opinion its fine to use A.I. as a tool, but I really hate it when I encounter lazy 'authors' who spam A.I. slob without bothering to check it's content over for things like consistency, continuity, contradictions, and continued incorrect data.

Click to shrink...

Honestly, same. AI is a tool. Not something you use to write everything, and certainty not used without corrections.

 

[SuperPulp2789]

Honestly, same. AI is a tool. Not something you use to write everything, and certainty not used without corrections.

Click to shrink...

It explains why some writers I know aren't panicking

Why? Because they write just fine….only they have a lot of issues and decide to work on different series or not at all for months to years between releases

It's mostly the webnovel guys who are very likely to just keep throwing out more and more content almost everyday

RoyalRoad has a lot of authors using AI to generate their covers, I think it's also recently, at least a few months ago that the AI-Assisted tag was added

AI won't change the fact that they're essentially making on average incompletely thought out, heavily cliche inspired pulps only they're more inspired by East Asian works

Yeah, a lot of it is very much mass produced and it's NOT high grade literature, if it was, they'd be driving themselves crazy editing and rewriting their works even after publication

 

[LegioCaesar]

Really at the end of the day AI-generated writing will be for its own slop-fed audience, there won't really be a lack of demand for actual creatives any time soon because people will always want to have the option of the art their consuming to be from an actual person, as opposed to some trained machine. I would say, however, that the site is sort of running the risk of getting bogged down if ai slop writers feel it's a safe place to dump, and then we're getting a potential devianart situation(I think someone may have mentioned such prior in the thread) where searching/browsing/navigation is gonna become...tedious, to say the least. Which is gonna suck, definitely.

Now, I'm not saying any rules need to be put in place, but a more robust mod stance of something like "yeah we'd prefer all writers here hold to the spirit of reciprocity and are clear about what AI is used/to what extent of AI-tooling was used to assist in a work, but you don't have to" which is sort of the position taken with the 'it'd be polite' bit, but ik asking people to care about things that won't directly affect them immediately is genuinely an ask sometimes, not even a diss btw it's just how most people operate. Anyway, I'm tired and I don't want to rant so I'm just gonna leave it off at that I think we should be a bit more mindful/careful with our next steps than some mild takes and "yeah we don't cares" when it comes to AI slop coming to the site past, ionno, generated images which while I don't think makes one an artist, but a prompt engineer, are benign in their output(tho I personally still hate that they're all basically trained off stolen work at the end of the day) and doesn't really affect the 'spirit' of the site ig.

Just my two cents, and it's p off the cuff so if someone has something to say just keep it civil and i'll respond when i can to clarify, for now I'm goin to bed.

 

[NumberSix]

You mean other than universities pushing it to the point of having entire classes that cannot be passed without getting 100% grammarly scores on each assignment? It's genuinely useful for formal writing, or at least it was back in the 2016-2020 era when I was in Uni.

Click to shrink...

I only started using it recently, and I rathed liked it, but mostly I try to use gnostwriter and manuskript rn.
I only tried the web version though, that is pretty far from what I have seen in those ads that were running non stop, most elaborete thing it did for me was more elaborate spell checking and punctuation fixing.
Also, I'd rather have something that runs locally without calling the mothership.

I like A.I. for certain tasks but I have no desire for my data and time to be used to improve a paid cloud service or get caught in vendor lock in, thank you very much.

I might try one of the distilled deepseek versions for local experimentation at some point, and there was this FOSS grammarly-like project I saw a while ago, dunno if it has the same feature set.

I am not a native English user and there is a lot of stuff that goes past me when I am using the language, I do most of my reading and a lot of my other communication in it, but learning it primarily from Cartoon Network, Baen's free library with Bujold, Drake and Weber books in particular, a few Tom Clancy novels the first 2 - 3 wheel of time books, aff.net and ff.net when I was a kid did leave a few holes, especially when I was flossing over stuff like facial expressions and body motions in fscor of dialogue and action and plot development left holes in my ability to use it proficiently for fiction writing.

So I like the idea of having something I can use quickly to try and rephrase something or improve what I have written and maybe as a proofreader.

On the other hand there were a few studies saying reliance of stuff like ChatGPT dumbs you down, so I have reservations about using it.

And I don't touch it professionally, I need to keep my skills sharp, although the itch to just say fuck it and make it write boilerplate is strong.

I love A.I. for porn image and concept image generation though.

Let's face it, I will never be able to draw worth shit and prompting the A.I. is a bit like programming, although I do need to sit down and rtfm one of these days.

And I think a lot of people who just want to visualize a crossover idea or make some porn feel the same way.

No harm done, no need to hate on them.

 

[SuperPulp2789]

Really at the end of the day AI-generated writing will be for its own slop-fed audience, there won't really be a lack of demand for actual creatives any time soon because people will always want to have the option of the art their consuming to be from an actual person, as opposed to some trained machine. I would say, however, that the site is sort of running the risk of getting bogged down if ai slop writers feel it's a safe place to dump, and then we're getting a potential devianart situation(I think someone may have mentioned such prior in the thread) where searching/browsing/navigation is gonna become...tedious, to say the least. Which is gonna suck, definitely.

Now, I'm not saying any rules need to be put in place, but a more robust mod stance of something like "yeah we'd prefer all writers here hold to the spirit of reciprocity and are clear about what AI is used/to what extent of AI-tooling was used to assist in a work, but you don't have to" which is sort of the position taken with the 'it'd be polite' bit, but ik asking people to care about things that won't directly affect them immediately is genuinely an ask sometimes, not even a diss btw it's just how most people operate. Anyway, I'm tired and I don't want to rant so I'm just gonna leave it off at that I think we should be a bit more mindful/careful with our next steps than some mild takes and "yeah we don't cares" when it comes to AI slop coming to the site past, ionno, generated images which while I don't think makes one an artist, but a prompt engineer, are benign in their output(tho I personally still hate that they're all basically trained off stolen work at the end of the day) and doesn't really affect the 'spirit' of the site ig.

Just my two cents, and it's p off the cuff so if someone has something to say just keep it civil and i'll respond when i can to clarify, for now I'm goin to bed.

Click to shrink...

I forgot who and the exact reasons why

But I recall someone making the argument NOT to just keep dumping chapters en masse

Gotta get proper advertising/marketing for your own works, even for stuff like books

 

[FieldKeeper]

So, basically recognize it's a tool, be honest about using it?

Seems fair. I loathe AIbros, but that's because they basically want the robot to do the work for them instead of actually, y'know, learning to use the robot. They're just shitty writers instead of getting help from an LLM to make up for areas they aren't so good in. And, well, I can't draw, but if I want a particular kind of sexy without paying for it...

I'm cool with it.

 

[NumberSix]

So, basically recognize it's a tool, be honest about using it?

Seems fair. I loathe AIbros, but that's because they basically want the robot to do the work for them instead of actually, y'know, learning to use the robot. They're just shitty writers instead of getting help from an LLM to make up for areas they aren't so good in. And, well, I can't draw, but if I want a particular kind of sexy without paying for it...

I'm cool with it.

Click to shrink...

I agree.

Also I think the AIbros are putting themselves in a pre-Butlarian Jihad type of situation if they delegate too much of their cognitive processes to software.

On the question of discussing A.I. writing tools, well, I think it will be nice to have a thread about it.

In fact a thread about all types of writing tools, not just A.I. ones would be nice.

 

[Devastator21]

On the question of discussing A.I. writing tools, well, I think it will be nice to have a thread about it.

In fact a thread about all types of writing tools, not just A.I. ones would be nice.

Click to shrink...

I would like that too would be neat.

 

[SuperPulp2789]

I agree.

Also I think the AIbros are putting themselves in a pre-Butlarian Jihad type of situation if they delegate too much of their cognitive processes to software.

On the question of discussing A.I. writing tools, well, I think it will be nice to have a thread about it.

In fact a thread about all types of writing tools, not just A.I. ones would be nice.

Click to shrink...

Part of the shitty part of the Pre-Butlerian Jihad situation, is that most people are dependent on the tech of technocrats

Post-Butlerian Jihad, there are STILL technocrats, they just work with the aristocrats, merchants and religious figures

You need a LOT of centralization to keep a ban going and to keep indoctrinating people a specific way

 

[Wallaperi]

A reminder that LLMs merely predict the next most likely word, using complex math on a subset of prior words, then brute forcing through an absolutely massive amount of memory entry before it finds the right "token" (next word)

Thus, imo, it can be extremely effectively used as a beta and grammatical tool, and maybe to fill in and tesselate details to an existing scene, but left of its own, you will get stuff that is fundamentally uninspired.

It is entirely entirely unsuited for raw creativity, because it simply apes human language extremely effectively, but does not actually comprehend the meaning of the symbols it arranges on a screen.

So great at acting like wood filler, but you wouldn't want to use it to replace the carver.

 

[NumberSix]

A reminder that LLMs merely predict the next most likely word, using complex math on a subset of prior words, then brute forcing through an absolutely massive amount of memory entry before it finds the right "token" (next word)

Thus, imo, it can be extremely effectively used as a beta and grammatical tool, and maybe to fill in and tesselate details to an existing scene, but left of its own, you will get stuff that is fundamentally uninspired.

It is entirely entirely unsuited for raw creativity, because it simply apes human language extremely effectively, but does not actually comprehend the meaning of the symbols it arranges on a screen.

So great at acting like wood filler, but you wouldn't want to use it to replace the carver.

Click to shrink...

Yeah, and biological evolution is based on a slow process of random mutations and there is this theory that humans developed speech and cognition because they started eating some weird, psychedelic shrooms at some point.

Reality is starting to look more and more like a Peter Watts novel and I ain't liking it one bit.

As to randomness it had been used as a problem silver method for a while now, see the Monte Carlo method .

And I am sure some of those models have error checking and correction mechanisms added.
But you will always need a Wrangler.

 

[ZeroK and Angel]

Frankly, it's a nice tool, but badly implemented most places and worse understood by those using it. Two examples, I'm keeping these vague for my own reasons just to clarify.
I saw an article about a certain AI hallucinating hard when asked certain questions. I played around and without knowing the prompt I was able to make it hallucinate in the same way. I then spent a minute asking the question in a actually cogent manor with the realization that the AI is an idiot, and it reliably spat out what I wanted the way I wanted it without hallucinations. So many people treat LLMs like they are sentient and then complain when they spit out crap answers to crap questions. Not all problems by any means are sorted by this, but knowing how to use the tool makes a HUGE difference.
Second example, I wrote up...*checks her notes* a 2300 word story concept. I then took this outline and put it into seven of the major models and told it to give me a outline, then write the first chapter for me. Some did give better results then others, but none of them are particularly good, and none were entertaining. Maybe some day with more complex weighting and we move away from LLMs towards a more fuzzy approach we might start seeing interesting writing coming out of these tools, but that is not likely to be any time soon. The BEST example I have been able to get results from was a fork a friend is playing with that allows kind of an active prompting, but when I tried that you basically have to hold it's hand non stop, and in the end I see no reason why I would want to use it. I can write faster, with far better results just...writing the story.
The only two things I've regularly used AI tools for in writing is when I space on a word and I can just plug in, 'the word means this and starts with a C' or such, that is so helpful as when I forget a word the chances of me remembering when I need it is HORRIBLE. The other is a few times when I'm stuck I ask it for an outline, specifically because my response is ALWAYS 'My god is that crap, it should be...' and then my brain kicks in and gets over the hump most of the time and I can keep working.
These tools aren't going to destroy our creative space, even if it someday does start producing interesting works, it will in no way devalue creative works written by people. It will likely always be pablum that is weighted for popularity and will generally just appeal to those looking for shallow entertainment with no thought involved. It will be Disney approved pop music; no substance and no challenging the audience. The mermaid doesn't commit suicide, the wolf isn't cut open to let the grandmother out, and the stepsisters don't cut off their toes to fit the shoe so the prince takes them home.

 

[Vox De Harlequin]

To chuck my own opinion on the shit mountain...
I'm not entirely happy with this ruling, but I don't see a way to practically resolve my issues so I can't complain.
As long as appropriate measures are taken against people who truly take the piss, it should be fine.
I consider AI generated product to be parasitic anti-art, and I refuse to engage with it for anything I actually want to enjoy in any meaningful way but that's just, like, my opinion dude.
I will just continue to exist as the enshittification of life itself continues around us all.

EDIT: Editing because I feel this may come across more antagonistic than necessarily intended, I'm just exhausted by the way things have been going in the world and on the internet lately. See it as an old man yelling at clouds if you want, I just wish the world was better.

 

[Pure Unadulterated Ego]

So, basically recognize it's a tool, be honest about using it?

Seems fair. I loathe AIbros, but that's because they basically want the robot to do the work for them instead of actually, y'know, learning to use the robot. They're just shitty writers instead of getting help from an LLM to make up for areas they aren't so good in. And, well, I can't draw, but if I want a particular kind of sexy without paying for it...

I'm cool with it.

Click to shrink...

Is it that surprising/wrong to want that? Now, it's not something I practice (inpaint and fix your shitty images instead of posting slop!), but if I had the possibility to get an entire fic with the quality of George R. R. Martin but completely written for me with the concept I want to read, I would fucking take it. It's kinda like magic when you can get quality smut, at least better than most smut was a few years ago, one hundred percent tailored to you.

Yes, there are people who don't notice how shit AI is at writing without any actual work put into it. Manual editing, tweaking prompts, and pushing back against its 'ideas' because you know it's better to go another way... and so on. It just doesn't surprise me that some people ultimately want the robot to do everything.

I will likely spend years learning how to 'use the robot' as it evolves for my images and I'm unlikely to ever try to learn to draw on my own. I don't particularly enjoy the process of drawing. I just want to create the images I want, which is why 99.99% of the stuff I do will die on my hard drive. It has, however, motivated me to learn how to touch things up, pick up my tablet and add general editing to the skills I want to get better at.

A reminder that LLMs merely predict the next most likely word, using complex math on a subset of prior words, then brute forcing through an absolutely massive amount of memory entry before it finds the right "token" (next word)

Thus, imo, it can be extremely effectively used as a beta and grammatical tool, and maybe to fill in and tesselate details to an existing scene, but left of its own, you will get stuff that is fundamentally uninspired.

It is entirely entirely unsuited for raw creativity, because it simply apes human language extremely effectively, but does not actually comprehend the meaning of the symbols it arranges on a screen.

So great at acting like wood filler, but you wouldn't want to use it to replace the carver.

Click to shrink...

This is an oversimplification that's just inaccurate enough to call out. The definition itself is not even true, otherwise we would have the exact same results every time you run a prompt. It samples a probability distribution depending on the parameters used during generation.

We can talk all day about what true understanding is, not really the point here, but there is understanding of logic and further meanings that have nothing to do with the textual definition. There has to be to actually 'ape' the human language extremely effectively. It knows what a dog is, sure, and also the sort of associations dog has (man's best friend, friendly, pet, companion and so on) or how do they relate to cats. If you ever used SOTA models, you know that they can be ridiculously good at catching nuances or implying things.

The fact that it can produce novel ideas should be enough proof tbh. Like combining a dog with a cat, getting Hatsune Miku to write a film critique of the last HP film or using a style that only has humans in the data to do landscapes.

Right now I do agree that it's not a tool that can stand on its own without coaching and work behind the scenes. I just disagree with the oversimplification.

 

[FieldKeeper]

Yeah, and biological evolution is based on a slow process of random mutations and there is this theory that humans developed speech and cognition because they started eating some weird, psychedelic shrooms at some point.

Reality is starting to look more and more like a Peter Watts novel and I ain't liking it one bit.

Click to shrink...

I'm involved in AI - adjacent fields.

Trust me, it's nothing like that. It resembles nothing so random.

Which is why I feel nothing but a combination of pity and scorn for AIbros, because LLMs do not have the ability to evolve like that. At all.

Hence, the "tool abused by idiots who think screwdrivers can be used to make entire houses for them" metaphor.

 

[Sksk4128]

Totally fair disclosure and don't spam are reasonable rules

 

[SuperPulp2789]

Yeah, and biological evolution is based on a slow process of random mutations and there is this theory that humans developed speech and cognition because they started eating some weird, psychedelic shrooms at some point.

Reality is starting to look more and more like a Peter Watts novel and I ain't liking it one bit.

As to randomness it had been used as a problem silver method for a while now, see the Monte Carlo method .

And I am sure some of those models have error checking and correction mechanisms added.
But you will always need a Wrangler.

Click to shrink...

The Stoned Ape theory

I think it helped them establish culture rather than mutate them

A culture that lasts generations and involves more cooperation and/or formalized yet versatile to varying degrees-hierarchies

 

[Wallaperi]

Is it that surprising/wrong to want that? Now, it's not something I practice (inpaint and fix your shitty images instead of posting slop!), but if I had the possibility to get an entire fic with the quality of George R. R. Martin but completely written for me with the concept I want to read, I would fucking take it. It's kinda like magic when you can get quality smut, at least better than most smut was a few years ago, one hundred percent tailored to you.

Yes, there are people who don't notice how shit AI is at writing without any actual work put into it. Manual editing, tweaking prompts, and pushing back against its 'ideas' because you know it's better to go another way... and so on. It just doesn't surprise me that some people ultimately want the robot to do everything.

I will likely spend years learning how to 'use the robot' as it evolves for my images and I'm unlikely to ever try to learn to draw on my own. I don't particularly enjoy the process of drawing. I just want to create the images I want, which is why 99.99% of the stuff I do will die on my hard drive. It has, however, motivated me to learn how to touch things up, pick up my tablet and add general editing to the skills I want to get better at.

This is an oversimplification that's just inaccurate enough to call out. The definition itself is not even true, otherwise we would have the exact same results every time you run a prompt. It samples a probability distribution depending on the parameters used during generation.

We can talk all day about what true understanding is, not really the point here, but there is understanding of logic and further meanings that have nothing to do with the textual definition. There has to be to actually 'ape' the human language extremely effectively. It knows what a dog is, sure, and also the sort of associations dog has (man's best friend, friendly, pet, companion and so on) or how do they relate to cats. If you ever used SOTA models, you know that they can be ridiculously good at catching nuances or implying things.

The fact that it can produce novel ideas should be enough proof tbh. Like combining a dog with a cat, getting Hatsune Miku to write a film critique of the last HP film or using a style that only has humans in the data to do landscapes.

Right now I do agree that it's not a tool that can stand on its own without coaching and work behind the scenes. I just disagree with the oversimplification.

Click to shrink...

No, it categorically cannot provide novel ideas. It remixes things it has been trained on and presents it to a human who may recognize the personal novelty and utilize it further.

yes you can train a model with another model and refine the computational needs, but IMO the media's talk about "AI evolution" (LLMs are not AGI at all) is a means of steering the "fear" of AI into robot takeovers, when imo the real fear for me is the asymetric capital distribution of gains from AI's workforce reduction. We will not see more wages, we will be paid the same or less and be expected to output more work, but with greater competition for jobs that you can live on. None of the value will be distributed to society as a whole.

Training models with better models isn't "evolution", its abstracted software refinement, so you waste less power finding the "answer" for whatever you trained it to do. If you want and flipped random bits, it would just break, not evolve.

You are still sucking GBs of matrix data through memory busses and doing a bunch of binary math.

I do not believe we are close to any emergent properties, but popsci headlines would think otherwise, to steer /how/ the average Joe is worried about AI. They want you worried about Terminators taking over and killing all humans, but realistically I am more worried about further decline into economic neofuedalism, with a peasant class either on UBI or work camps, and a cartel who own the means of production. Being good at a task /= intelligence or evolution, and intelligence is not actually required to get a workable end result. Current AIs bring to mind the p-zombie philisofical questions a bit as well.

I believe that AI is dangerous enough as is, with basic 21st century power centralization issues, and the ability for corporate products to intentionally cultivate parasocial relationships with their users (AI companions etc) but from a hardware perspective, we are still talking massive matrix calculations, no emergent properties present.

Black boxes, sure, but any article talking about AI like it actually has a conciousness is ridiculous. Ask chatbot to do a few math problems, and you will run into a shortfall of the token process just with that. Or in line of work, ask AI about esoteric industrial products from 20 years ago, and you get a bunch of extremely human-readable, authoritative, but wildly incorrect answers.

Usually the example given of "AI hacked the PC to not lose' are usually improperly defined tasks, because the human setting them used implicit assumptions without explicitly defining them for the tool.

Like the tetris AI that found a way to use a bug to freeze the game to "play tetris forever".

It didn't evolve to do so, people simply made improperly defined assumptions about what "playing" meant. Because again, these tools are not actually understanding anything, it's an abstraction layer that humans put meaning on. That capability alone is enough to get a LOT of stuff done, but I dont think the current hardware has the physical chops.

 

[Rukia]

(I actually found an AI artist already that does their art in the Asanagi style, which is incredibly satisfying, when one has an immediate need for curvy shortstacks with smug grins made for correction that have amazing ahegao faces)

Click to shrink...

You really gonna post this and not drop a sauce?

 

[SuperPulp2789]

No, it categorically cannot provide novel ideas. It remixes things it has been trained on and presents it to a human who may recognize the personal novelty and utilize it further.

yes you can train a model with another model and refine the computational needs, but IMO the media's talk about "AI evolution" (LLMs are not AGI at all) is a means of steering the "fear" of AI into robot takeovers, when imo the real fear for me is the asymetric capital distribution of gains from AI's workforce reduction. We will not see more wages, we will be paid the same or less and be expected to output more work, but with greater competition for jobs that you can live on. None of the value will be distributed to society as a whole.

Training models with better models isn't "evolution", its abstracted software refinement, so you waste less power finding the "answer" for whatever you trained it to do. If you want and flipped random bits, it would just break, not evolve.

You are still sucking GBs of matrix data through memory busses and doing a bunch of binary math.

I do not believe we are close to any emergent properties, but popsci headlines would think otherwise, to steer /how/ the average Joe is worried about AI. They want you worried about Terminators taking over and killing all humans, but realistically I am more worried about further decline into economic neofuedalism, with a peasant class either on UBI or work camps, and a cartel who own the means of production. Being good at a task /= intelligence or evolution, and intelligence is not actually required to get a workable end result. Current AIs bring to mind the p-zombie philisofical questions a bit as well.

I believe that AI is dangerous enough as is, with basic 21st century power centralization issues, and the ability for corporate products to intentionally cultivate parasocial relationships with their users (AI companions etc) but from a hardware perspective, we are still talking massive matrix calculations, no emergent properties present.

Black boxes, sure, but any article talking about AI like it actually has a conciousness is ridiculous. Ask chatbot to do a few math problems, and you will run into a shortfall of the token process just with that. Or in line of work, ask AI about esoteric industrial products from 20 years ago, and you get a bunch of extremely human-readable, authoritative, but wildly incorrect answers.

Usually the example given of "AI hacked the PC to not lose' are usually improperly defined tasks, because the human setting them used implicit assumptions without explicitly defining them for the tool.

Like the tetris AI that found a way to use a bug to freeze the game to "play tetris forever".

It didn't evolve to do so, people simply made improperly defined assumptions about what "playing" meant. Because again, these tools are not actually understanding anything, it's an abstraction layer that humans put meaning on. That capability alone is enough to get a LOT of stuff done, but I dont think the current hardware has the physical chops.

Click to shrink...

I've asked AI questions and at times it feels, no is definitely, making shit up

Not afraid of it killing off the human race

More afraid of the technocrats and whatever plans they got 2030, though I can say like AI, a LOT dumber than they think

We're already seeing a population crash and lowered birth rates even in less well off nations

Most graduates aren't that smart and the kind of smart businesses are looking for are more along the lines of decision makers who can talk to people very well, businessman shit, self motivate and lead/delegate rather than just be able to do a lot of STEM stuff themselves

I read John Carter and the Gods of Hollywood and it kinda affirmed my idea on how businesses operate. There's a LOT of information, there's a LOT of people with skills, the problem is navigating and directing them all and dealing with the oh so many suggestions/demands.

 

[Pure Unadulterated Ego]

No, it categorically cannot provide novel ideas. It remixes things it has been trained on and presents it to a human who may recognize the personal novelty and utilize it further.

yes you can train a model with another model and refine the computational needs, but IMO the media's talk about "AI evolution" (LLMs are not AGI at all) is a means of steering the "fear" of AI into robot takeovers, when imo the real fear for me is the asymetric capital distribution of gains from AI's workforce reduction. We will not see more wages, we will be paid the same or less and be expected to output more work, but with greater competition for jobs that you can live on. None of the value will be distributed to society as a whole.

Training models with better models isn't "evolution", its abstracted software refinement, so you waste less power finding the "answer" for whatever you trained it to do. If you want and flipped random bits, it would just break, not evolve.

You are still sucking GBs of matrix data through memory busses and doing a bunch of binary math.

I do not believe we are close to any emergent properties, but popsci headlines would think otherwise, to steer /how/ the average Joe is worried about AI. They want you worried about Terminators taking over and killing all humans, but realistically I am more worried about further decline into economic neofuedalism, with a peasant class either on UBI or work camps, and a cartel who own the means of production. Being good at a task /= intelligence or evolution, and intelligence is not actually required to get a workable end result. Current AIs bring to mind the p-zombie philisofical questions a bit as well.

I believe that AI is dangerous enough as is, with basic 21st century power centralization issues, and the ability for corporate products to intentionally cultivate parasocial relationships with their users (AI companions etc) but from a hardware perspective, we are still talking massive matrix calculations, no emergent properties present.

Black boxes, sure, but any article talking about AI like it actually has a conciousness is ridiculous. Ask chatbot to do a few math problems, and you will run into a shortfall of the token process just with that. Or in line of work, ask AI about esoteric industrial products from 20 years ago, and you get a bunch of extremely human-readable, authoritative, but wildly incorrect answers.

Usually the example given of "AI hacked the PC to not lose' are usually improperly defined tasks, because the human setting them used implicit assumptions without explicitly defining them for the tool.

Like the tetris AI that found a way to use a bug to freeze the game to "play tetris forever".

It didn't evolve to do so, people simply made improperly defined assumptions about what "playing" meant. Because again, these tools are not actually understanding anything, it's an abstraction layer that humans put meaning on. That capability alone is enough to get a LOT of stuff done, but I dont think the current hardware has the physical chops.

Click to shrink...

That was... something. Like half of that was completely off topic to what we were talking about, which was that your oversimplification went too far and was downright inaccurate. There's no buts about that, it doesn't just pick the most likely token and that's not subjective.

You said "it doesn't provide novel ideas, it presents something to a human that might recognize it as novel" which makes no sense if you really think about it.

I'm 100% behind these models not having a conscience, or anything close to the AGI they want to sell. I don't think that's possible with the current arch and tech. You are just trying to use a definition of understanding that goes beyond what's used in this context. Like with the style example I mentioned, if the AI can extrapolate how a dog should look in a style that had no dogs at all, THAT'S NOVEL. It's in fact, how we humans come up with original things most of the time.

It's like you want to have your cake and eat it too. Either they can't really 'ape human language extremely well' like you said or they have some level of base understanding (that we agree isn't AGI, but rather the examples I gave before).

Saying they understand concepts because they can define them, use them and combine them with other concepts isn't exactly strange. We're not saying they have actual intelligence. Context matters.

 

[RedStarMini]

You could radically cut down on unflagged AI slop by simply making a default system where authors are required to select the level of AI usage - a simple dropdown that says "AI used to generate most/all text - AI used to generate some text - AI used to edit or beta read - AI not used at all" and have the dropdown be required when starting a new thread would go a LONG way, even if there were literally no rule or enforcement associated with the tag.

Click to shrink...

Dude do you realize that we only recently got the ability to delete our own thread polls instead of needing mods to do it for us? Like I do not know the specifics of it but adding a bespoke feature like that PURELY for AI is not that simple.

 

[CriticalFailure]

You really gonna post this and not drop a sauce?

Click to shrink...

Fair enough.

Search "ShmentaixLord"

 

[Zanderion]

Dude do you realize that we only recently got the ability to delete our own thread polls instead of needing mods to do it for us? Like I do not know the specifics of it but adding a bespoke feature like that PURELY for AI is not that simple.

Click to shrink...

Well, I think it the case of thread poll it was simply importing (read:buying) SB/SV's implementation. I could be wrong and QQ buy/comission it independently instead, but it seems likely.

You'll probably have better chance asking on SB or SV for bespoke Xenfore features than on QQ. The former constantly commisioning new feature development (and no less importantly, have revenue sources to fund said development), the latter also constantly develop Xenforo add ons — because the owners/parent company make money by selling such and the revenue from such is part of how SV fund its continual existence.

 

[Jay98]

AI generated slop should be tagged as such. Didn't bother to write = me not bothering to read.

 

[Greyshaft]

My only concern there is that people who use AI are, in my experience, horrendously oversensitive. Just this thread alone has had several people mention how much it "might hurt someone's poor feelings" when somebody simply asks the question, "Is this made by AI?" I can only guess you'll be getting a lot of erroneous harassment reports in the near future, and I can only hope you won't humor them.

Click to shrink...

I agree, and worry that this might turn into a major issue. I think getting accused of trying to start a witch hunt or of attacking the author simply for suggesting / assuming they're using AI could wind up being a real problem, if it isn't already. I've already seen a situation where an initial piece of constructive criticism that assumed the author was using an LLM to assist their scene writing and was specifically discussing the limits of the LLM and how they showed up in the author's writing got an immediate denial / dismissal from the author. The poster went to defend their initial post and it rapidly spiraled into a multi-page argument where all parties got pretty heated, and the person initially giving the constructive criticism was dogpiled with accusations of trying to start a witch hunt. Far as I can tell from a few close reads of their initial posts, at no point were they making any explicit or implicit moral judgement about the use of AI in and of itself; That moral judgement would've had to be assumed by the author and the other readers themselves. And it's not like I'm unsympathetic, because I also agree that;

If you ask an author that manually wrote their story if they use ai, then you're really saying "the writing is bad / unnatural enough that I believe it was made by an ai."

It isn't overly sensitive to recognize that the question is mildly insulting to actual authors.

Click to shrink...

So my concern then becomes; Given that we can't magically know whether someone manually wrote their story or not, is there a constructive, non-offensive way to discuss an author's writing in the context of AI usage at all in a thread? Something that an author wouldn't be within their rights to baulk at or go on the defensive over?

I'm earnestly interested in getting feedback in this, especially from authors of fics on this site, because I reckon the feedback could be used as a model of how to broach this sort of thing in the future. Say that someone believed that I was using AI in a story I was writing because of patterns they'd observed in my writing, and wanted to bring it up with me specifically in the context of AI and its pitfalls as a writing tool? How would they need to go about this in order to make it clear they aren't making a moral judgement about authors using AI or trying to attack me, and are genuine about wanting to see my writing improve?

And, given that;

We encourage authors to tag for however they use AI in their writing process. Also, you are allowed and encouraged to ask authors to tag for AI if appropriate.

Click to shrink...

Is there an appropriate way that authors ought to be expected to respond when dealing with someone discussing their writing as potentially AI assisted in the threads / stories they make, or asking for them to tag? I don't think it's an overreaction to believe that there are going to be a lot of authors who categorically interpret any assumption, any suggestion or any request to tag that they're using AI as a personal attack and a rule 1 violation by extension. That and there are going to be plenty of people agreeing with them and joining in on prosecuting the people who make those assumptions/suggestions/requests. I fear that 'you should tag this as AI if you're using it to write with' is going to become seen as the new 'this story is shit, I'm no longer reading', if it isn't already.

I suppose broadly my two cents is that I'm concerned that if there are witch hunts, they're going to run in the opposite direction that everyone seems afraid that they will, and that QQ is going to wind up a place extremely hostile to any discussion of AI writing and its pitfalls that doesn't take place in a 'safe space' sequestered off from any actual stories and the authors writing them. In an environment like that, I'm worried that the writing quality of any and all stories using AI assistance would plummet because nobody felt like it was worth it to bring it up and lazy, LLM dependent authors would be able to get away with just about any level of slop that isn't outright illegible.

It's The Emperor's New Clothes, but instead of non-existent threads it's a fucking wasteland of emdashes as far as the eye can see.