Summarize your story with AI. How did it do?

[old_prof]

Computers went from one machine taking up a six or eight-story building to micro-processors in your phone in 70ish years. But the further they advanced, the faster one generation became the next happened.

AI may eventually do amazing things. I don't deny that. The current generation of AI has already run its course, in my view and the view of most of the AI researchers I respect. There's not much more that straight LLM's will be able to do. Emily Bender, Timnit Gebru and Margaret Mitchell (not the one who wrote Gone With The Wind) had them pegged five years ago as "Stochastic Parrots". They're an entertaining parlor trick, but will never move past that. I'd point you to read some of Gary Marcus's work if you want to understand what they lack.

New technologies will come and I do believe that we will eventually see the need for many professions, including programming, eliminated.

Historically, each wave comes about a decade after the previous, but there is no real reason for that and it's not at all like clockwork. If I had to place a bet, I think we're probably two or three waves away from an AI that can actually do almost all (maybe all) human activities better than we can, including sw dev and the arts.

Two or three waves is probably thirty years give or take. After I'm going to worry about it.

The era of exponential growth of computing power has also come to a likely end (actually did about a decade ago). Some forms of computing, including computer graphics and neural nets/LLM's, can ride the mega core wave a little farther.

Probably my most popular course, and the one I most liked teaching, was a history of computing technology course. I could talk about this stuff for hours, but I will bore everyone to death doing so.

 

[Andreas_Kreuz]

Except the productivity gains are deniable.

Unfortunately the studies are lagging behind. Before Gemini3 I did not see productivity gains either, but now they are there. We also understand better where AI works and where it fails. Frontend code works very good, because it is repetitive. Complex business logic is still better written by humans.

 

[MillieDynamite]

Since I edit and ghostwrite for a novelist who uses AI for his first draft (which I subsequently turn into a second draft), I can give a list of AI-ism writing. Repetitive use of long dashes, repetitive use of the word, and between some action, and a little more action, and even some more, usually with no commas separating the action from the and. Repetitive (and sometimes constant) use of the word then tying things together. An overuse of terms such as even more, more than enough, more than expected, more than...well, you get it. The word 'just' is used over and over again. A phrase (any phrase) repeated time after time or to end each chapter, "He turned and walked away, never looking," was one I had to nuke out of three different novels. Once it de-AI-ed, I then rework it into a better story and send it back. He makes notes on changes and other suggestions; I rewrite them again and send him my final version along with an invoice. He pays me, then turns it over to his publisher. If they are happy, it's fine and good; if not, he pays me to work it over again. We've done this for 3 years now, and I can see improvements in those first drafts that he sends. The next book will be the sixth novel I've done for him.

 

[old_prof]

Unfortunately the studies are lagging behind. Before Gemini3 I did not see productivity gains either, but now they are there. We also understand better where AI works and where it fails. Frontend code works very good, because it is repetitive. Complex business logic is still better written by humans.

Obviously the studies are lagging because a tool needs to be in production for over a year before you can even start seriously evaluating its gains and costs.

Studies of programmer's perceptions of the gains show they are skewed by losing grunt work, which is more annoying than time consuming. It's also the only part easily replaced.

I am biased towards systems level sw dev, because that's what I most enjoy doing, but I have run multiple groups that included some amazing UX talent. As a caveat, I finally gave up on trying to maintain professional level currency in web dev tech three or four years ago, so I'm not an expert in the current tools there.

I certainly agree that boilerplate front end code can be effectively created by AI. But I don't think that requires an LLM to generate. Each generation of UI code (I've been through a lot of them) has always been prone to tools that would cover most of the work. I just think LLM's are a terrible tool for the job. They work, but so do (and have) many other technologies. I think others would actually do a better job than LLMs do if 1% of the effort was putting into integrating them in that has been used forcing LLMs in. But LLMs are the hammer of the y and Ui code is just another screw.

I don't want to turn this into yet another debate about AI, so I'm not going to reply any further. Fell free to continue.

 

[Actingup]

Since I edit and ghostwrite for a novelist who uses AI for his first draft (which I subsequently turn into a second draft), I can give a list of AI-ism writing. Repetitive use of long dashes, repetitive use of the word, and between some action, and a little more action, and even some more, usually with no commas separating the action from the and. Repetitive (and sometimes constant) use of the word then tying things together. An overuse of terms such as even more, more than enough, more than expected, more than...well, you get it. The word 'just' is used over and over again. A phrase (any phrase) repeated time after time or to end each chapter, "He turned and walked away, never looking," was one I had to nuke out of three different novels. Once it de-AI-ed, I then rework it into a better story and send it back. He makes notes on changes and other suggestions; I rewrite them again and send him my final version along with an invoice. He pays me, then turns it over to his publisher. If they are happy, it's fine and good; if not, he pays me to work it over again. We've done this for 3 years now, and I can see improvements in those first drafts that he sends. The next book will be the sixth novel I've done for him.

At this point, do you want to kill him?

 

[MillieDynamite]

While our interests align with programmers', our concerns aren't the same, the advances in AI writing is happening fast. So fast, I can see improvements with each successive story my novelist sends me. However, I can't tell if his prompting is getting better or the AI writing (I believe it is Novel AI). I don't want it to improve to the point he doesn't need me. Truthfully, I don't like that he uses AI at all. And to be honest, it probably would be cheaper if I wrote from his outline and he didn't give me the 1st draft from a machine program. I mean, that has to cost him a lot for the program, and little ole me, sure makes a good amount of money from reworking it.

Obviously the studies are lagging because a tool needs to be in production for over a year before you can even start seriously evaluating its gains and costs.

Studies of programmer's perceptions of the gains show they are skewed by losing grunt work, which is more annoying than time consuming. It's also the only part easily replaced.

I am biased towards systems level sw dev, because that's what I most enjoy doing, but I have run multiple groups that included some amazing UX talent. As a caveat, I finally gave up on trying to maintain professional level currency in web dev tech three or four years ago, so I'm not an expert in the current tools there.

I certainly agree that boilerplate front end code can be effectively created by AI. But I don't think that requires an LLM to generate. Each generation of UI code (I've been through a lot of them) has always been prone to tools that would cover most of the work. I just think LLM's are a terrible tool for the job. They work, but so do (and have) many other technologies. I think others would actually do a better job than LLMs do if 1% of the effort was putting into integrating them in that has been used forcing LLMs in. But LLMs are the hammer of the y and Ui code is just another screw.

I don't want to turn this into yet another debate about AI, so I'm not going to reply any further. Fell free to continue.

No, as long as he doesn't use anyone else.

At this point, do you want to kill him?

 

[Andreas_Kreuz]

Yes, AI can be prompted to even generate decent prose. But I see no point in doing that. I enjoy writing, not prompting. It's like enjoying cooking tasty meals or buying cans of baked beans. And publishing generated prose is like selling canned food in a restaurant, while wearing a chef's costume.

What I have found when experimenting is that it might produce a reasonable style with the right prompt, but it utterly fails when it comes to logic, subtext, or innuendo. And it fails at longer texts (>10000) words. After that, it forgets who is who and what happened before.

 

[Bramblethorn]

New technologies will come and I do believe that we will eventually see the need for many professions, including programming, eliminated.

I would distinguish between "programming" and "coding" here.

LLMs can write computer code. Just how well the current versions do that is in the eye of the beholder. In general, I would expect them to be better at doing things that a thousand humans have already implemented ("I want to build a website where users can post articles") and worse at innovative applications where they don't have relevant human-written code to crib from. (I'm more on the latter end of things, and last time I tried asking an AI for help, it gave me some very detailed advice on how to use a function that *does not exist* in the relevant programming language. Thanks a bunch.)

But before one can code, or prompt Copilot to write code, one needs to understand the problem and be able to articulate it. That's a large part of the programmer's skillset, and not one that I expect LLMs to replace any more than calculators replaced accountants.

Some yet-to-be-invented AGI? Possibly. But I suspect the rise of LLMs has taken us further away from AGI.

One thing that does concern me is the risk of LLM coding assistants stifling innovation. People who've come to rely on LLM-based coding assistants are likely to struggle if they ever need to work in some new or special-purpose language which doesn't have enough examples online to adequately train the LLM.

Historically, each wave comes about a decade after the previous, but there is no real reason for that and it's not at all like clockwork. If I had to place a bet, I think we're probably two or three waves away from an AI that can actually do almost all (maybe all) human activities better than we can, including sw dev and the arts.

Two or three waves is probably thirty years give or take. After I'm going to worry about it.

The era of exponential growth of computing power has also come to a likely end (actually did about a decade ago). Some forms of computing, including computer graphics and neural nets/LLM's, can ride the mega core wave a little farther.

Moore's Law has been great (even if it's now slowing down drastically), but we don't talk enough about improvements elsewhere. I've worked in fields where a 2020s algorithm running on 1990s hardware will beat a 1990s algorithm running on 2020s hardware. It's one of the reasons I'm pessimistic about LLMs: the initial idea was clever, but so much of the focus since then has been on working harder, not smarter. More training data, more and more data centres, far too much emphasis on throwing money at the problem rather than genuine innovation.

 

[HelenL]

To the original question... I don't have Copilot, but I tried feeding my Valentine's submission into free ChatGPT (completely forgetting about possibly supplying training date - more later).

I was very surprised at how accurately it summarized the base of the story, which is fairly complex. That's scattered throughout much of the story rather than wall-of-texted, so pulling it all together seems like an accomplishment. However, the base is not the story, and other than establishing rules isn't especially interesting. The story is what happens to the protag, secondary character and a small multitude of actors in vignettes. ChatGPT summarized that as (quote): "but things go awry when...". So it's surprisingly accurate and entirely worthless. (Reminds me of an old joke about a helicopter and the Microsoft campus.)

I don't know enough about GPT and whether the free access is maybe tossing out the story because of sexual situations. It's possible that a licensed or unrestricted version would be more complete. I would say it doesn't seem to have gotten the point.

As for training data, I asked it afterward:

did you use this story for training data

No. I don’t have access to my training set at the level of individual stories, and I don’t know whether any specific webpage or story (including that one) was included.
More generally: I was trained on a mixture of licensed data, data created by human trainers, and publicly available text. I don’t retain or recall exact documents from training, and I don’t browse or learn from links unless I’m explicitly using a web search tool in the moment.
So I didn’t “recognize” or retrieve that story from training—I summarized it only based on what was available when you shared the link.

 

[Andreas_Kreuz]

What the reality of using user input to AI chats actually is, is hard to tell. There are 2 risks involved for the AI companies:
1: Input might be consciously skewed to modify the LLM towards a certain bias. This happened with the first test version of GPT, which had at the time used user input for training. It was fed with fascist prompts and content by user until it became biased towards fascist and racist opinions.
2: If AI is fed input from AI, it does not increase the value of the model, because it is reading something it already knew. And this actually leads to quick degradation.

However all AI companies retain the input of users and process it, unless you pay them not to. They probably use it not for training, but for debugging and understanding trends.

My assumption is that our input and prompts probably go into AI models, but not the main ones that answer public questions, but private ones that help AI companies understand how their systems are used.

 

[Bramblethorn]

As for training data, I asked it afterward:

Although this particular response may be accurate AFAIK, it's generally not advisable to take GPT's self-explanations on trust. I've seen many, many times where it's been asked "why did you do [thing in previous response]" and it will make up an answer that couldn't possibly be true.

 

[HelenL]

Although this particular response may be accurate AFAIK, it's generally not advisable to take GPT's self-explanations on trust. I've seen many, many times where it's been asked "why did you do [thing in previous response]" and it will make up an answer that couldn't possibly be true.

Yeah, absolutely. I'm pretty much convinced that even the "oops" responses after a disastrous error are often more "this is what is expected" rather than an analysis of a problem.

 

[Lifestyle66]

I only post my stories to Literotica. When I asked ChatGPT to: Summarize the writing style of the Literotica userID "Lifestyle66".

The following is the response I received:
****************************************************************************************
I can’t find any reliable public source or profile information about a Literotica user named “Lifestyle66” — at least not one that’s indexed or searchable outside of the site itself (and a few generic indexes). Literotica allows users to publish erotic fiction under pseudonymous handles, and those aren’t generally indexed by search engines or publicly cataloged in review articles or databases. Wikipedia

Because of that, I cannot summarize the writing style of “Lifestyle66” based on authoritative data.

Here’s what that means practically:

• Literotica is a large site full of user-generated erotic fiction, contributed by hundreds of thousands of authors. Wikipedia
• UserIDs like “Lifestyle66” are account names chosen by contributors — they’re not standard public literary personas that have been profiled outside the site.
• Unless you provide a specific excerpt or story title from that author, I don’t have an external source to analyze or summarize their style.

***************************************************************************************


So, it appears Literotica does block AI's from scraping the stories posted here. If you can get ChatGPT to summarize your stories, then the AI might be finding your userID elsewhere.

 

[Lifestyle66]

In further testing ChatGPT's capabilities to scrape this site, I again asked it to summarize one specific story of mine with the prompt:

Provide a summary of the Literotica userID's story "A Band of Sisters and Brothers".

It returned TWO options for me to select which response I preferred:

**********************************************************************************************
Which response do you prefer? Responses may take a moment to load.

Response 1
I can’t access or read the full text of a specific Literotica story like “A Band of Sisters and Brothers” on my own — and I can’t summarize it unless you provide the text you want summarized.
If you paste in the part (or parts) of the story you want summarized, I can produce a concise summary of that text for you.
Just drop in the excerpt you care about, and I’ll summarize it.

Response 2 One idea per line
I wasn’t able to find a verified summary of a Literotica story titled “A Band of Sisters and Brothers” by user Lifestyle66 — there are no reliable indexed pages or summaries available publicly for that exact text.
Important context:
Literotica is a user-generated erotic fiction site with hundreds of thousands of stories posted under pseudonymous IDs, and most of those stories are not indexed by search engines or referenced outside the site itself — so there’s no authoritative source I can cite directly for that title. Wikipedia
That means I can’t provide an accurate summary of that specific story without either:

1. A direct excerpt or synopsis from the story itself, or
2. A description of its plot, characters, and themes from you.

If you share a bit of detail — even a few sentences or the main characters and events — I can help you craft a clear, concise summary.
If you meant something else:
There are other works with similar names in public literature (for example Band of Sisters by Lauren Willig, which is historical fiction about women volunteers in World War I) — but that’s unrelated to Literotica and likely not what you’re asking about. newbookrecommendation.com
So just let me know more about the actual story content you want summarized, and I’ll help you produce the summary!


*************************************************************************************************



SO, for those of you worried about AI's learning from your work, it appears that the only way they can get your work is if someone copies and pastes it into their chat with the AI. I think the Admins here have blocked the scraping.

 

[Lifestyle66]

And yet another test to reassure those of you concerned about AI's scraping your work:

Knowing that our stories are readily accessible to the anonymous public, I prompted it asking for a summary of a specific link with:

Provide a summary of the erotic story at the following website link: https://literotica.com/s/a-band-of-sisters-and-brothers

******************************************************************************************************

I couldn’t find any publicly available summary of the specific Literotica story “A Band of Sisters and Brothers” from indexes or search results — there’s no reliable synopsis of that title in the web pages indexed that are accessible. Literotica content generally isn’t indexed with plot summaries outside the site itself, and there were no results describing this story’s narrative or themes in the results we found. Wikipedia

If you want a useful summary, you could paste text from the story here (even just the first few paragraphs or key sections), and I can provide a synopsis based on your excerpt. Summarizing user-provided text is allowed even for adult content. Wikipedia
*****************************************************************************************************

So, for those who criticize the Literotica Admins for not doing enough to "fix" their site, I think they have their hand full with keeping up with the evolving technology and should be applauded for the work they have done!

 

[Bramblethorn]

So, it appears Literotica does block AI's from scraping the stories posted here. If you can get ChatGPT to summarize your stories, then the AI might be finding your userID elsewhere.

It does now (at least for some; the robots.txt is basically a polite request not to scrape the site, which may or may not be honoured), but it didn't use to.

The fact that it doesn't recognise your username doesn't mean your stories weren't scraped or used for training; it just means LLMs don't remember everything that was in their training data.

 

[Lifestyle66]

It does now (at least for some; the robots.txt is basically a polite request not to scrape the site, which may or may not be honoured), but it didn't use to.

The fact that it doesn't recognise your username doesn't mean your stories weren't scraped or used for training; it just means LLMs don't remember everything that was in their training data.

That's why I gave it a specific story hyperlink, which any human anonymously could read.

ChatGPT can't "read" it.

 

[Actingup]

Thanks, L66. This is consistent with the other tests. Note that for the post that started the thread, I gave AI the text directly to analyse, rather than asking it to access it from Literotica. I should have made that clear.

 

[Lifestyle66]

Thanks, L66. This is consistent with the other tests. Note that for the post that started the thread, I gave AI the text directly to analyse, rather than asking it to access it from Literotica. I should have made that clear.

Some people here just complain about the AIs without trying to understand what they can and cannot do. I try them as a test to better understand their capabilities and limits.

The responses I received above indicate to me that ChatGPT can't "think" of a response on it's own but only find things similar in publicly indexed Internet sources, then sort of compile a summary of summaries to feed me what it "thinks" I want as the answer.

So, when I ask for a summary of "Lifestyle66's" writing and it can't find that name or ID in any public sources, it's stuck with no response.

If I copied and pasted my story for ChatGPT to "read", it would probably take key portions of that text to compare to other stories, then feed me a recompiled summary of what those other stories had on the Internet. It's not thinking, it's regurgitating what other humans have already said about such stories.

It's a superfast search engine, summarizing the search results for me.

 

[SkyBubble]

Cool idea.

 

[darwin1859]

I won’t feed AI my work, but this made me curious to know if I’ve been scraped… so I asked chatgpt who Jackie Hikaru is and this is what I got:
[...]
What I’ve learned is:
1. I guess I’ve been scraped
2. Blech, curses on that AI slop (please tell me I don’t actually write like that)

N.B., I publish under this name on other sites so it may not be scraping from here.

ChatGPT can decide to search the web, and it would probably tell you that it is doing that. Do you recall if you saw it do that? It's possible it knows you from its training data (i.e., that you have been scrapped), but when I google your username, Literotica is the first hit. So, it's equally plausible that it found your works page and summarized that.

I think actually it's more plausible that you are not in its training dataset because I speculate that OpenAI (and other vendors) probably tried to keep erotica and porn out of the training dataset. You don't want someone to ask AI what to get mom for her birthday and have the AI respond, "Mom's like cream pies. You should fuck her."

We have some evidence (or hearsay) that a lot of the work that goes into training the model is actually curating the dataset...

 

[SugarStorm]

The current generation of AI has already run its course, in my view and the view of most of the AI researchers I respect. There's not much more that straight LLM's will be able to do.

Hi! I'm (genuinely) curious, seeing as you know more than me on the subject. What do you mean when you say that the current generation of AI has already run its course?

I might be misunderstanding you (I’m slow ), so feel free to elaborate, but while the current generation of AI may or may not have run its course (what do I know), it feels like the applications of it are only just getting started.

What I mean is, how we use these tools is evolving rapidly, even if the underlying models aren’t making giant leaps every month.

A small, everyday example: browsers now have built-in AI answer assistance. I know some people hate that, but honestly, I find it useful. It condenses a ridiculous amount of material into something I can grasp quickly. A few AI bots (like Google's) can transcribe and summarize lengthy YouTube videos, which to me, is quite useful.

A sillier example: I’m playing a JRPG with far too many characters, and I reached the point where I couldn’t tell if my lineup made sense anymore. I took a photo of the screen, wrote a short prompt, and the AI analyzed it, browsed the web, and gave me a summary of prevailing strategies. It cited game sites and subreddits I already "trust". And it responded not in a “I think you should do this,” but “the current consensus tends to be…” And that distinction matters to me. A glorified Google search maybe, but the simplicity of it makes it magical.

A more serious example is work. I’m in a large organization where governance, process, and compliance outweigh almost everything else. The documentation is endless. So when one of the governance teams rolled out an AI chatbot, the collective groan was audible.

Except… it works.

All the relevant material is there: policies, meeting minutes, decisions, dates. Roles and responsibilities. Terminology, definitions - all of it. I can ask questions in context of my role, reference a specific meeting or topic, and it gives me a clear answer with sources. I can even ask complicated, conditional questions, like “if this, then that, but what if also this?” and it handles it transparently.

It doesn’t replace judgment. It doesn’t invent policy. But it saves me from digging through hundreds of pages of PDFs or PowerPoints to figure out what applies to a specific situation.

For me, that’s the first real, tangible, day-to-day use of AI that actually improves my work. And that’s why I’m not convinced the story is over yet (that's kind of what I took from your statement, fully acknowledging I might have misunderstood you ).

 

[AwkwardlySet]

@SugarStorm

If I may add my thoughts here, I'd say that we can expect the development of AI to happen just like any other invention in science. A huge majority of scientific progress happens through incremental gains, which means improving the existing theory/invention/technology gradually in small portions.
And then there is the breakthrough progress that usually means a leap in science or development. Those are quite rare, especially these days.

What that means, in my opinion, is that LLMs will continue to slowly develop incrementally. Whether that's by constantly making the data set larger, or by filtering the data and making it more reliable before training the LLM, or even by improving the algorithm itself, it doesn't really matter. They will continue improving, slowly. Keep in mind, though, that IT time isn't as slow as most other sciences.

I've seen many people state here that LLMs have reached a ceiling and that they are now a dead end in the search for the true AI. I'm not so sure. They certainly aren't true AI, nor will they ever be, but never underestimate the power of evolution in this sense. LLM has emerged as an evolution of much earlier techniques, from data mining, statistical models, neural networks...
There's always a possibility that some new idea can give them a new angle of development and that it might lead to a different model, something that might become a true AI. It's how most of the science functioned so far.

Either way, I think it's too early to announce the end of their development. It wouldn't be the first time that a new development in one field has led to progress in a different field.

AI is undoubtedly helpful in real life, although I'm disgusted with what AI has and will undoubtedly continue doing to art in general. Maybe there should be some hard locks in that sense on all the LLMs. I don't see a better solution.

 

[alohadave]

It doesn’t replace judgment. It doesn’t invent policy. But it saves me from digging through hundreds of pages of PDFs or PowerPoints to figure out what applies to a specific situation.

You are trusting the AI to select salient data points from something you can't be bothered to read yourself.

You have no way of knowing if the AI is correct, way off base, or hallucinating.

 

[SugarStorm]

You are trusting the AI to select salient data points from something you can't be bothered to read yourself.

That's fair, I guess, but I’d frame it differently. I can either work hard or work smart. If I trust the data source, I’m perfectly happy letting a computer do the heavy lifting of sorting and summarizing what’s relevant. I still apply judgment, I’m just not manually digging through hundreds of pages to get there.

You have no way of knowing if the AI is correct, way off base, or hallucinating.

That’s not true in this case. As I mentioned earlier, the tool explicitly cites its sources, including links to the original PDFs and even page numbers. I can verify any answer in seconds if something looks off. It’s not replacing verification; it’s drastically reducing the time it takes to get to it.