Clarksworld closes submissions due to GPT spam
twitter.com
twitter.com
Anyways, thinking more about the topic at hand, things like ChatGPT are just pushing us further towards a world of nepotism and social back-channel connections. Like in the modern art world where it isn't about talent, it's about who you are and who you know. AI means it's not about what you can write or create, it will be about your personal connections. IMO this is a bad thing for society as it leads to entrenched stratification and eventual ossification.
Addendum: Drowned out amongst millions of AI augmented voices.
See e.g. https://wtfhappenedin1971.com/
Only if you view in black/white absolutist terms where if every single person isn't on 100% merit then it doesn't count at all.
But the only thing this argument achieves is making the people actively causing or cheerleading the damage feel better about what they are doing. For everyone else, it's a massive regression.
Nepotism doesn't have a defense, because it's defense is essentially "tradition". I've genuinely not heard a compelling argument for that either - it always boils down to something anachronistic (i.e. racism, classism, etc).
Wouldn't that be an ideal application for ChatGPT etc?
But if ChatGPT can be trained to remain civil and avoid prejudice and conspiracy theory, the quality of Hacker News would objectively improve even if everything ChatGPT says is qualitative nonsense. And it won't be, and as software improves, it will be less and less so. You can't say that about people, unfortunately. The quality of people tends to degrade over time, the more people the more degradation. That's how you get Eternal September. ChatGPT, however, can maintain a high degree of quality over time, unimpeded by irrationality or emotion.
Maybe we could split the difference and have all comments filtered through ChatGPT. Run sentiment analysis to catch and remove or edit low quality comments where feasible.
There are probably many more that don't get caught. What anti-botting measures does HN have deployed?
That's a good question I'd like to know the answer to.
Look im talking to you and we dont know each other, and Im not sure why you think it's different using smtp or http ?
Or is it like wanna-be writers who can't tell the difference between ChatGPT output and really good writing, and figure that if they can just put their brilliant idea for a story into a prompt and have ChatGPT clean it up, their talent will finally be recognized?
What does proof-of-work look like for fiction? Uploading drafts, ensuring "adequate" time between them and "adequate" revisions? What else would work?
Proof of work could just mean "do your work here".
Shouldn't tech be about making things easier not harder.
The reason is the title of this forum post. Proof of work is precisely what we are seeking here. There is no magic wand solution to this problem.
For my own entertainment, I've tried to have ChatGPT write a science fiction story a few times, but its stories weren't entertaining.
The main problem is that it sounds like it is summarizing. For example, here is a typical start to one of its stories:
" In a world much like our own, a team of scientists were working on an experimental device known as the Gravitational Wave Interferometer. They had been working on the project for years, but had been unable to make it work. The device was designed to detect the subtle ripples in space-time caused by gravitational waves, but it required an incredible amount of precision and sensitivity."
That's fine, but it sounds like the summary of a story rather than the story itself. It's telling rather than showing. For comparison here is the intro from the popular book The Hunt for Red October:
"Captain First Rank Marko Ramius of the Soviet Navy was dressed for the Arctic conditions normal to the Northern Fleet submarine base at Polyamyy. Five layers of wool and oilskin enclosed him. A dirty harbor tug pushed his submarine’s bow around to the north, facing down the channel. The dock that had held his Red October for two interminable months was now a water-filled concrete box, one of the many specially built to shelter strategic missile submarines from the harsh elements. On its edge a collection of sailors and dockyard workers watched his ship sail in stolid Russian fashion, without a wave or a cheer."
This sets the stage in a more interesting way than ChatGPT does.
What kinds of prompt strategies can I use to make ChatGPT's stories sound more like storytelling and less like a synopsis or summary? Can someone share a specific prompt?
Stories are entertaining when they raise the expectations of the reader or viewer and make them wonder how a difficult situation could possibly resolve. Entertaining means you keep on reading because you are curious about what will happen next.
But language-models just produce an average over the many texts they have been trained on. While each story they read had their own "expectation loop", when you average over many such stories the expectation-raising disappears, like multiple waves interfering and disrupting each other.
The output of ChatGPT is bland, because there is no single author behind its output, who could "feel" whether the story was entertaining enough to be published. ChatGPT of course can not "feel" anything. It can not evaluate its own output as to will it raise curiosity in its readers. ChaptGPT has its model of language and texts, but not a model of what its texts MEAN and thus no model of how readers might react to it.
“Set the scene introducing a protagonist in a Soviet missile depot in the 1980s Soviet Union in the style of Tom Clancy. Make it ominous, descriptive and imaginative.”
And got:
The Soviet missile depot was a forbidding place, set deep in the heart of the country's vast and unforgiving landscape. It was a place where men spoke in hushed tones and shadows seemed to lurk around every corner, where the hum of electricity and the beeping of computer terminals were the only sounds to be heard.
At the heart of the depot stood a solitary figure, a man with a weathered face and piercing eyes that seemed to see straight through you. He was dressed in a drab olive uniform that was stained with grease and sweat, and he wore a cap with a red star on it, the symbol of the Soviet Union.
His name was Ivan Petrov, and he was a veteran missile technician with over twenty years of experience. He had spent his entire adult life in the service of the Soviet state, and he knew every inch of this place like the back of his hand.
As he walked through the maze of tunnels and corridors, he could feel the weight of the world on his shoulders. He knew that the missiles stored here were capable of destroying entire cities, and that the fate of the world might one day rest on his shoulders.
The air was thick with tension and uncertainty, and Ivan felt a sense of foreboding that he could not shake. He knew that the Soviet Union was on the brink of collapse, and that the future was uncertain.
But he also knew that he had a duty to his country, and that he would do whatever it took to protect it, even if it meant sacrificing his own life.
As he stood before the missile control panel, his fingers poised over the launch button, he could feel the weight of history bearing down on him. He knew that this moment would define the rest of his life, and the fate of the world.
In that moment, Ivan Petrov was the most powerful man in the world, and he would do whatever it took to protect his country and its people, even if it meant unleashing the most destructive force in human history.
I've had success with some promopts but for others it just falls into vague nonsense like "In a flashback a hero fights a villain." Telling it to come up with some more specific than "a bad guy" often doesn't work.
I think the problem with ChatGPT's output in this case is that it's not really a story, or even a summary of a story. It's possible the summary of the start of a story, but where is the ending? What happened with the Gravitational Wave Interferometer? Did the team of scientists make it work, after all? Was it eaten by space sharks? Was it engulfed by the sun going nova? Did Nurgle hold a wedding feast on the planet? How does the story end? How does it even continue?
But, if the story had an ending, it would make decent flash fiction, in my humble. In fact I like to write very short stories like that, that sound like summaries. It can work- but it has to be a summary of something interesting to read. A story with a beginning but no middle or end doesn't cut it.
The best solution will be for ChatGPT to have a way to verify that certain creations are not generated. Just like identifying duplicate content using tools like Copyscape. Of course, there are issues like user privacy which will need to be taken into consideration, but that seems to be the only fool-proof way.
Clarksworld is probably where you find discussions about the latest forklift model :P
But also I expect AI-written novels to be pretty crap, so I'd prefer humans keep doing it.
I can’t see why. It could very well, on its own, use other genre’s narrative structures, stories or styles that are uncommon or inexistant in SF.
It could also be used by a human writer to flesh out the story and concepts the writer has created. There are a few authors I find brilliant but whose writing and characters were sorely lacking.
It’s really more about what you do with it than wether or not you use it. I mean, computers have enabled people to simply copy-paste stuff and not produce anything meaningful. Some do just that. Others have thrived.
"...a last ironic thought drifted through Hendricks’ mind...They were already beginning to design weapons to use against each other."
There’s too many things to list that you “can” do that society collectively believes you shouldn’t because the negatives outweigh the positives.
2) The volume of submissions has grown massively because spammers are using LLMs to generate spam submissions in the hope of getting paid.
3) The current generation of LLMs are not particularly good at generating quality literature on their own, so this massive deluge of submissions is mostly garbage.
On other hand GPT might technically be good enough. But there isn't any actual substance. The reason why we read literature. This is varying bar, but still it does exist.
We currently don't have any machine-level filter that would identify whether or not a piece of prose is actually a coherent work of narrative fiction or not. And annoyingly, if we did, AFAIK we'd also have everything we needed to train an AI (specifically, a GAN) to generate narrative fiction. So we're going to be dealing with this right up until the point where it doesn't matter any more, rather than being able to do any kind of quick hack to banish it "early."
https://thegradient.pub/an-introduction-to-ai-story-generati...
Unfortunately those are a) virtually unknown outside academic circles and b) about to die a death. The latter, the death, they're about to die it because their space is now being taken over by a much more prolific technology that generates low-grade bullshit cheaply.
Life, innit. It goes in circles.
If all the money is in generative models that give "good looking" results, who is going to spend money investing in the techniques that don't produce as "good looking" results right now but who have a potential beyond recombining it's training data?
Sufficiently advanced AI is indistinguishable from people who don't understand the core issue here. It doesn't feel pity, or remorse, or fear. And it absolutely will not stop... ever...
If the stories were great, or even if .001% were great, then I think the magazine would be happy as they’d be able to put out more great stories.
Now you do not have to pay attention to anything. This is utopia. It was a waste of time for my father to tell me to work hard.