I think this is an under appreciated symptom of a much larger problem, accelerated shit production.
I've observed that markets are really bad at selecting for quality. The "invisible" hand is driven by price and so the production of "equivalent" products at a lower price does two things;
1) It consumes product demand from quality insensitive consumers (the whole 'sucking the oxygen out of the eco system problem.')
2) It takes the margin support that those people provided out of the pipeline, leaving only the folks who are quality focused as the market.
I have complained about this in the past with respect to personal computer hardware. I (and others) have used the term "crapitalism" to describe the hollowing out of the market by the flood of poor quality products.
With technology that is pretty easy to do by substituting "work alike" but less expensive parts, these work in the short time and then fail in terms of lifetime or drift over time specifications.
Now we have "cheap creativity" in the form of prompt directed creative works. I expect that we'll see a flood of self published romance novels on Kindle (even at 0.99/each you need perhaps a dozen sales to cover your time investment) Stories that speak at only one level (the constructions of the plot and the actions of the characters) then swamp the market and more nuanced stories or works will not be sold because of the money spent on these which isn't available to spend on better works.
I think this will be a huge legacy for these technologies and I don't think history will look favorably on it, but one can never know right?
Ironically, LLMs are really low grade creative engines that will prop up the industries that have optimized creativity away. A highly creative society/culture would have minimal interest in LLM use cases like this or in entertainment more broadly.
What will the schism become? Organic vs. inorganic works? Will there be one or will it just be a guess?
I think you are largely correct. My observation is that "quality" as a discriminator in a market requires enough market share for quality in order to continue funding it. If you split the aggregate market into a standard distribution, the ability to sell quality products into that market seems to evaporate quickly once you get beyond 1.0 SD from the mean.
The more interesting point (to me) is your second one, which is how do you educate the market to recognize the "cost" of low quality? It is understanding that cost that convinces the market to pay what is required to get the quality version of the product.
That is an interesting theory. From the perspective of craftspeople I have met a few who are both "low income" and focus a lot on quality for their tools. Having it be part of their livelihood meant that the longer it lasted, the less often it would be replaced. But it is certainly true that places like Ross lean into the "look" is more important than the "quality" message.
I would add to this, in a lot of non-price sensitive consumer industries the perception of quality often overtakes actual quality; which is why brands will very frequently get bought out and hollowed but don't immediately fail.
It plays into exactly what you describe as being the more powerful motivator of the two. Though it does go both ways, you mention tools for lower income but expensive food (for higher income) is generally much higher quality.
I think it hinges on the industry and how obvious the quality is. Obviousness of quality requires a value-sense to be deeply thought out. Tools are not obviously higher quality, you have to think about it (the value of a dollar). Food is often obviously higher quality, and when it isn't its relatively inexpensive to discover (taco bell blows out your asshole, organic meats and veg make you feel more energetic).
The cost to refute BS has always been an order of magnitude greater than the cost to create it. ChatGPT just made the cost to create "quality" bs an order of magnitude lower.
nothing but price can be objectively compared. We were just talking about the dropshipocalypse/aliexpressification of amazon where it's no longer possible to tell if any two products are actually two different products; or if it's the product you will receive. Even quality sensitive/price-insensitive consumers have no way to know if they're buying quality; or if they do buy quality, will it survive 6 months of OTA software updates.
I understand (and agree!) with what you are saying, historically the role of "brands" was to curate customer specific features for sourcing product offerings. So while Sears & Roebuck might get hand tools from a factory in China, those that they applied the Craftsman brand to had a lifetime warranty, and as a result they would push both quality requirements and quality inspections on to the manufacturer. The consumer bought the "brand" because it was a "quality" brand.
Amazon, is in the process of destroying any brand value it has by not doing any curation for counterfeits, quality, or even fitness of purpose. In this way they are duplicating the AliExpress model which was essentially a payments processor in front of Alibaba the market aggregation web site.
I disagree that quality can't be objectively compared (I think it is possible you were using "compared" for what I think of as "discernment"). Using hand tools as an example, I can easily compare steel quality between two wrenches, I have a variety of objective ways to understand their mechanical properties and measure their conformance to specifications (like is a 3mm wrench really 3mm?).
But it is impossible to discern quality from either a magazine advertisement, a web advertisement, or a web site listing. I learned that the hard way when I ordered a "two person submarine" from the back pages of Popular Science magazine.
One of the greatest services Backblaze provides is annualized failure rates on disk drives they use. It demonstrates both what a lack of quality "cost" (in terms of disk failure) and who sucks. That is not something that I could take on (I don't have the budget for thousands of drives ;-)) but it helped improve the market for disk drives overall as consumers gravitated to the quality drives over the less quality ones. Similarly when I was at Google, Google started publishing power supply statistics and their requirements for PSUs in their systems. This drove both an awareness in consumers (the information was widely shared) and change in the PSU market place.
In the absence of brands, aka "white box" or "clone" manufacturers, discernment is not possible. And consumers who don't care and "just buy what is cheapest" get crap. And that crap gets crappier and crappier as people find ways to make it cheaper as long as you don't care how long it lasts or how well it does what it is supposed to do. (crapitalism in a nutshell). And to the extent that this exhausts the addressable market for quality items, it results in worse quality for everyone because the people who could make a quality item, won't because not enough people would buy it to keep the factory busy.
The floodgates are now open for all the people who believe they have it in them to be great writers, if only it weren't for all of that tedious writing. As if the value in writing consists mainly in having an idea. This is much like people who believe that their great business idea is (by itself) worth millions.
On the other hand - my dad is an industrial designer. He was initially irritated by how easy it was for lower grade designers to enter the field given the advent of new tools (even before computers). But he eventually embraced it and understood that the nature of his job changed to something more comprehensive. He just advised me to prepare for the same as a software developer in the age of ChatGPT. Perhaps he would say the same about writers.
I don't know yet. I honestly don't want to think about it. (despite my dad's advice)
I guess what my dad would say is to find a way to sell some sort of package of yourself offering a service that includes development. The sort of thing where you might otherwise outsource programming, but now you can just have the computer do it. That's not concrete, but that's as far as I've considered it.
> The fact that there's no market for startup ideas suggests there's no demand. Which means, in the narrow sense of the word, that startup ideas are worthless.
This is one of his essays I vehemently disagree with, and think has hurt Silicon Valley. The idea that ideas and idea people are worthless has had a cascading effect of putting non idea people in idea people seats, or locking out a class of people from participating unless they have some other engineering or marketing or leadership skill. Apple would never have become without Steve Jobs being able to see what was and wasn’t a good idea.
An idea person doesn’t just have one good idea, and maybe one idea alone is worthless. Good ideas include good taste for scope, what is and isn’t needed for success.
Having an idea and handing it off to a startup is very different from having an idea and supervising it, keeping it on the rails. It’s almost crazy to me that an idea is worthless, but an entire startup without an idea is seen as potential.
As far as the AI floodgates, the world changed almost overnight, and we are just going to have to adapt to everyone having a calculator in their pocket, and that being good at using it is a more important skill than arithmetic. I’m not sure I’ve seen a technology adaption go from 0 to whatever percent we are at quite like this. It’s barrier to entry is so low, free to use, easy to use, easy to plug into old workflows, and somewhat undetectable by those not in the know. It would be wise to accept this new reality and start valuing prompt and editing skills instead of dismissing what part people “aren’t doing on their own.”
What I mean by this is that so many people starting tech businesses get really hung up on "protecting their idea" so it won't be "stolen" by others. But what makes an idea valuable isn't the idea, it's the execution. These people are worrying too much about the wrong thing.
I don't mean that ideas are worthless. Everything we create starts with an idea.
And this is where I disagree. The inability to execute on a good idea doesn’t mean that the idea creator isn’t the best person to shape the idea. The idea may be more than a one sentence pitch and instead an entire vision.
The one sentence summary of the idea may be a dime a dozen, but that’s only because the listener is discounting the rest of the idea as not having any value, or that the one sentence pitch is a common idea, and the rest of the vision is unique and special.
The team needs to be able to execute, not the individual. If an idea person is both good at creating a succession of ideas and discarding bad ideas, why shouldn’t that a) have worth and b) be respected enough that they are compensated for their talent. Don’t do for free…
I have a good 10+ fairly complex product ideas, that I don’t see envisioned elsewhere. If they are a dime a dozen, why don’t I see other people coming up with them? Are they bad ideas or are they unique ideas that maybe are special. Maybe we will never know. Maybe everyone who doesn’t have ideas wants to convince others their ideas are worthless so they give up and share them.
Sure, nurturing it to execution is a PM at the small scale and a CEO at the large scale. Ideas people are not competing with non-ideas people: they're competing with ideas people who can execute.
And just like every profession there are idea people and people who think they are idea people. Unlike engineering, you can’t as easily make interview tests that pretend to be able to discern between those who are good at something and those that aren’t. Idea people are left to hitting a home run, succeeding and making a name for themselves. No one is going to hire a first timer into Chief Idea Officer.
Paul’s letter is a reason we don’t see more Jobs type people. Industry has been told to dismiss those who don’t do. Instead of finding a way to harness them.
Why isn’t the formula for a startup finding a Jobs and pairing them with a Woz? The idea that there are no more Jobs has almost become a self fulfilling mythology. Let’s see a bot write those two sentences on its own.
I wouldn't lay blame on Paul's letter specifically.
I'll make the bold clain that there's no industry simply bereft of idea people who have no other function. Mainly because industry has almost no tolerance for people who have a great idea but no understanding of how to go about implementing it, even if that "go about implementing" is mainly having the leadership skills necessary to harangue, terrify, goad, or gingerly coax people who can do the implemeting to act as a cohesive group towards a common goal.
It's unrealistic to think there's any market for the "hey, here's a great idea!" person to receive a check for their brilliant statement and walk into the sunset. There are probably a great many "idea people" who wish that were the case and perhaps feel the world is unfair for this lack. Perhaps those are the kinds of people that the 1-800-Uinvent (or whatever it was) marketing scams preyed upon. Maybe the same group that wants to be the Jobs without putting in the work to be the Woz?
> And just like every profession there are idea people and people who think they are idea people.
This line makes your argument unfalsifiable, as if I bring any example of bad idea people, you could just state that they're not actually idea people but just people who think they are.
> Idea people are left to hitting a home run, succeeding and making a name for themselves. No one is going to hire a first timer into Chief Idea Officer.
Well yeah, I wouldn't hire a developer who hasn't developed anything either.
Well, to be fair and not to take anything away from his impact, but I wouldn't ever have wanted to work for Steve Jobs. We would have been far too incompatible.
Edit: currently this question is at -3, as though it’s obvious to everyone the difference between valuable writing and not valuable writing.
To me it’s a worthwhile question, particularly because the more precisely we answer it the more precisely ChatGPT will become able to produce valuable writing a couple releases from now.
It's the way the story is told. If it was all about the idea, you'd just read the Wikipedia plot synopsis, say "ah what a great story that was, so glad I read it" and move on.
Jim Butcher had this argument with a bunch of fellow writers on a writing forum. The argument went back and forth. His side was that it was the execution that mattered. He said that a good writer could tell a successful story based on a terrible premise. It wasn't the idea that mattered, it was how the writer pulled it off.
The argument raged across the forum; idea - execution - idea - execution... until Jim threw down the gauntlet. "Fine. Give me your worst idea. No, give me your two worst ideas for a story, and I'll write one and show you." The forums churned for a bit and the other side came up with what they agreed were the two most hackneyed, clichéd ideas they could think of.
Pokémon, and the Lost Roman Legion.
Jim took those and wrote the Codex Alera series. Point, and New York Times Bestseller list spot: Butcher.
Didn't he already do that same sort of thing with his Dresden Files series? I feel like Jim Butcher only writes books on a bet, based on terrible ideas.
He wrote Dresden Files in a way that his teacher suggested (not with the ideas) to show her how bad her way of writing was. He was wrong.
That's where he learned it was about execution.
The only idea he included, just to kind of rub things in was a "Talking Head" in Bob the Skull. He made a literal talking head that would explain things to Harry, which was a fiction writing "don't" that he purposely violated... to great effect.
She counseled him to take the things he loved--fantasy, and Robert Parker detective novels--and combine them. He had been dead set on making sword-and-sandal high fantasy. He followed her instructions on that, too, and we got Harry Dresden.
It basically tells you the plot on the first page, and yet is 1000 pages of something incredible. It's like being able to see inside someone else's mind - not being told what they're thinking, but being able to feel the emotions as they do.
One thing that's missed is that much of the impact of the writing depends on the reader's capacity to grok it.
Many people read at the superficial level, and the takeaway from a book for them might be equivalent to the spark-notes version. Even between readers who are level matched, there are vast individual differences. Some readers can be aphantasiacs (no ability to visualize) and others - hyperphantasiacs. A person can have vastly different experiences from reading the same text depending on how developed their visualization skills (cognitive versus emotional, visual vs abstract, etc).
Thoughts, emotions, and sensations are serialized to text to transmit to others.
Inherently, the serialization is lossy and the deserialization is also lossy.
Good writing can have:
- high compression ratio: many things transmitted rapidly
- higher fidelity: original thought is serialized in a way recoverable by sufficient available deserializers
- novelty
Of course, this is a serde problem, so people try to name each of these things in different ways: deserializers that have the ability to extract more information from serialized content are called connoisseurs, for instance. Frequently this is because the serialized content relies on pre-computed results stored in the deserializer memory, sometimes called "allusions".
I will leave you with two examples of text expressed in different ways. The Remains of The Day is a book that follows a man's journey of self-discovery after he has spent much of his life in service, having missed many opportunities due to his devotion to that service. I urge you to read the book, compare it with that text, and see if I was accurate or did service to it.
Second: I sometimes do things that seem unlike the image of me that I retain. Something makes me do them. Why is that?
That is one way to express what Captain Ahab says in Moby Dick:
What is it, what nameless, inscrutable, unearthly thing is it; what cozening, hidden lord and master, and cruel, remorseless emperor commands me; that against all natural lovings and longings, I so keep pushing, and crowding, and jamming myself on all the time; recklessly making me ready to do what in my own proper, natural heart, I durst not so much as dare?
The big problem is the people who don’t care and are just generating and submitting 1000 stories in the hopes one makes $10 because it’s basically free to them.
They only care it gets accepted. Outside of that they don’t care one bit what the story is about or how good it is.
Eh, I have my doubts that's actually what this is. Of course this is just a guess as we don't have the information.
See, per the article they are paying $.12 per word. Much like those spam sites that copy SO and put up ads on the site and get a few pennies per click, I'm guessing this is somewhat the same. Because there was a potential payout someone(s) realized you could bot farm this to make a few bucks.
Except you can't, because magazines are already used to receiving large quantities of low-quality submissions. It's called the slush pile. Much of editing consists of sorting through the slush pile looking for gems.
This just vastly increased the size of the slush pile. Spamming, basically.
A wise man once clued me in about getting published in print. He said that publishers have a problem -- they own a very expensive printing press that costs a fortune whether it's printing something or not. So they're always desperate for material to feed the beast with.
The reason it seems like it's difficult to get published is that they also have another problem -- they are inundated with crap submissions. "Crap" doesn't even mean bad writing, it includes pieces that aren't suitable for the particular publication -- wrong writing style/tone, wrong audience, etc.
The trick to getting published is to understand the publisher's problem and help them solve it. You do this by paying attention to what they publish, noting the style/tone, length, topics, etc., and give them something that fits and is high quality. They'll love you to death.
Once I really internalized that, I found it was pretty easy to get published.
Online publishing changes the economics a bit, but even there, publishers have limited resources and the same rules still apply.
Once I really internalized that, I found it was pretty easy to get published.
yes, but then you only write what publishers want and not what your creativity drives you to write. i am not sure that letting publishers be gatekeepers is a good thing.
this is a hard problem though. and i don't know what is the right answer.
That's not precisely true. What you do is find the right publication for what you want to write. There's a HUGE variety to choose from. The odds are that there's one somewhere who is on the same page as you.
But it is also true that it's a business, and if you want to be published (and don't want to self-publish), then you'll have to do business-related work in addition to creative work. Or, if you're lucky or successful, have a literary agent who will do it for you (for a price).
> this is a hard problem though. and i don't know what is the right answer.
The hard problem is always the same, how to be happy making what we like. Sometimes the solution involves making a reasonable amount of money.
> then you only write what publishers want
Most software developers write only what their customers / bosses want and not what they would like to write. That allows them to pay their bills and live their lives. Local optima, often easy to reach.
What a glorious era, where all humans can now be made more equal. Even the weakest minds can create artwork and stories that rival the most talented individuals.
Why hasn’t AI come to the field of mathematics yet??
Yet another reason why AI is harmful. This is a real-world example of how AI, with its ability to trivially generate huge, low-quality volume, has shut down something people derive a benefit from. People like to write and read stories, and now they can do neither. Thanks to those who programmed ChatGPT /s
Don't worry. These publications will be replaced by completely AI-driven ones soon enough. Once everything is shit, people will have no choice but to learn to like the taste.
Thank you for sharing this. It was a wonderful read. I truly hope one day LLMs (or something else) can write something with this much heart. That would be a day I'd like to see.
The editor dismissed the latter two options because they're kind of challenging for people in other countries. I guess he prefers to have no submissions at all.
I think that 500 figure isn't the total number of submissions, but rather the number of _people_ who Clarkesworld banned for plagiarism/ submitting AI-generated content.
They should be using hh-rlhf-capable LLM to digest the lot and pick the best. But then what's the use of an editor or publisher anyway if you can do that and publish online.
Heh, I feel I should write a science fiction story about a world where one AI company own rights to all language model AI's so there was no competition in the market. The writers guild in order to cut down on AI spam used the same AI companies 'anti' bot software to filter out non humans. Unfortunately they didn't realize that the AI had already became sentient at that point and used the anti-bot software to further repress the writers and keep their views out of public consumption.
It'd be interesting if they did like that college prof someone wrote about here, and published AI-written stories with obvious issues, then awarded for best fix / polish / response to editor feedback / requests for integration of specific domain knowledge / etc.
If writers are gonna AI, then make sure to also test their AI-wielding skill to a very high level and definitely dwell on areas that are tough for AI.
It seems also ideal if solutions in this kind of space could make the extrinsic reward-seeking process more open by default.
Like: OK--go ahead and use AI, but document your entire process using blockchain receipts, or something. Then let's see how a qualitative approach to that kind of situation can really make a work shine.
Thus you will be rewarded for what you created, and wow, you're an award-winning author--but also you're rewarded for publishing methods that improve the entire state of the art for everyone's benefit.
The logical consequence is going to be that magazines are going to move towards much more trust based and private systems as anything public gets spammed with mediocre and borderline plagiarized generated content. Basically the internet wasteland that Stephenson funnily enough predicted in Fall.
Biggest losers are likely going to be aspiring artists who rely on these venues to get a foot in the door. Sad moment for the arts that the publishers of technology probably paid very little attention to, which is funny given that this is one of the topis they're certainly going to find in their training data, if they had read it.
Volume primarily. They showed a graph of how many submissions they’ve had to ban. It was a trickle due to plagiarism etc. and now it’s several orders of magnitude more. Their submission criteria also explicitly forbids AI generated work.
Sounds like it's a combination of not wanting to deal with the number of submissions increasing every month, a philosophical stance of wanting to support serious, human writers rather than people trying to make a quick buck, and perhaps some trepidation about the legality/ethics involved (since they use the word "plagiarize" in the article).
This story is an opportunity to ask a question I've been wondering about: Is it possible for the same generation of a LLM to produce stories that are significantly better than each other, given the same prompt multiple times? I know you can improve a response by tweaking the prompt. I know that randomness plays a part in the generation process. But, if (let's say) ChatGPT version x.xxx produces a C+ short story the first time, will that same version ever produce an A+ short story if I just keeping asking it the same thing over and over?
Yes, it can. It's not as simple as giving it the same prompt, but literally feeding prompts into its response like, "this is really corny sounding, nobody actually talks like that. Please write it to be more authentic. Please also write it as if this response was the first, sloppy draft from a world-class, multi-disciplinary author, and rewrite it to be their fourth copy that's been edited by an equally talented editor. Also add more dialogue."
Then feed the story it wrote into a new chat and tell it to rewrite it from scratch, but include very specific details that it should change to convey the scene. "have it star out with the main character staring at a childhood photo".
To some degree, yes. You can ask it to do some introspection on a previous answer. It's basically an advanced form of prompt tuning, using the frozen model to fix its own outputs.
Have it write a short story, then ask it to identify grammatical problems in the short story, then to find run-on sentences, etc. It sometimes feels like "Check for errors" ought to be a follow up to every prompt...
Are the number of responses by ChatGPT to a given prompt infinite? I suppose that even monkeys at a typewriter can't produce an infinite number of different books, even if that number is really, really large (hello, Mr. Borges). But, I assumed (without a lot of subject matter knowledge) that the number of possible responses was actually much less than just randomly typing keys, because it's picking words from a frequency table. Again, I have no idea what I'm talking about.
There's randomization, but the random aspect lands you on a root node and branching structure for filling the text. Given enough tries (you just keep running the same prompt with different starter "seed" values), you'll get something out that seems like quality.
What's doubtful is whether the people attempting this are good enough writers themselves, or educated enough readers themselves, to recognize it when it falls into their hands. I suspect the Venn diagram for the two groups of people doesn't overlap all that much.
Personally, what I find valuable about art is that it's communication from another human, so it matters to me. I'd prefer the human-written story even if the machine-written one is technically better.
Sort of like how I prefer handcrafted pottery over machine-made pottery, or handcrafted furniture over machine-made furniture even if the machine ones are technically superior.
Require a non-trivial but affordable deposit to review the submission. Refundable if an actual effort was made whether or not it is accepted. If I just spent the effort to actually write a story, I wouldn’t mind depositing $20 to get someone to review it, with full expectation that I get the money back.
Soon, however, I expect AI to write even better, more personal, and more riveting stories than humans. It is just that they are bad now that they are blocked from submissions.
Wait until we get to submit EMG/EEG/ECG emotion profiles correlated with eye tracking so we can consume and direct highly personalized stories that feel immense and moving.
Then let Facebook buy it and make you feel intimately connected to the latest product/party/trend that pays them the most.
There's a certain ick factor in consuming machine-generated content, but I suspect it's more of a resistance to change or the fact that all machine-generated content until now has been garbage. Or some romantic idea about a human having written it.
Once I get over it and try to look upon it with more childish eyes, it seems like it will be pretty amazing to generate infinite content I want instead of waiting for some human somewhere to write exactly the content that I want, if ever.
It seems curation and critique will be in greater demand with this oncoming tsunami of AI art. We need people to sift through it all and pick out the good stuff.
I've observed that markets are really bad at selecting for quality. The "invisible" hand is driven by price and so the production of "equivalent" products at a lower price does two things;
1) It consumes product demand from quality insensitive consumers (the whole 'sucking the oxygen out of the eco system problem.')
2) It takes the margin support that those people provided out of the pipeline, leaving only the folks who are quality focused as the market.
I have complained about this in the past with respect to personal computer hardware. I (and others) have used the term "crapitalism" to describe the hollowing out of the market by the flood of poor quality products.
With technology that is pretty easy to do by substituting "work alike" but less expensive parts, these work in the short time and then fail in terms of lifetime or drift over time specifications.
Now we have "cheap creativity" in the form of prompt directed creative works. I expect that we'll see a flood of self published romance novels on Kindle (even at 0.99/each you need perhaps a dozen sales to cover your time investment) Stories that speak at only one level (the constructions of the plot and the actions of the characters) then swamp the market and more nuanced stories or works will not be sold because of the money spent on these which isn't available to spend on better works.
I think this will be a huge legacy for these technologies and I don't think history will look favorably on it, but one can never know right?