AIs gunning for our precious freelancers

[Raphael]

Has everyone forgotten that the real problem is capitalists cutting costs by exploiting machines that don't need to be paid instead of exploiting people that do need to be paid? If we weren't living under capitalism, the invention of sophisticated generative networks would not lead to this kind of discussion. Instead we'd be talking about how cool it is that we can achieve the same amount with less effort, which leads to more leisure time.

Well, there was this post of mine from late July:

I don't think that abolishing capitalism would necessarily mean the end of AI. It might as well mean the switch from a state of affairs where new labor-saving devices mean "Oh no, a lot of us will lose our jobs!" to a state of affairs where new labor-saving devices mean "Yay, we all have to work less!"

[KathTheDragon]

Oh, good, at least one other person has sense.

[malloc]

Has everyone forgotten that the real problem is capitalists cutting costs by exploiting machines that don't need to be paid instead of exploiting people that do need to be paid? If we weren't living under capitalism, the invention of sophisticated generative networks would not lead to this kind of discussion. Instead we'd be talking about how cool it is that we can achieve the same amount with less effort, which leads to more leisure time.

In principle, but currently we don't live in that world so it makes little sense to evaluate the technology along those lines. Before speculating about such possibilities, we need to consider the practicalities of its use under our current system. More importantly, the jobs eliminated by generative AI are not crap that people consider intolerable burdens. Nobody is complaining about the onerous chore of writing novels or drawing pretty girls and wishing a machine would take the burden of creativity from them. If the jobs on the chopping block were flipping burgers and mining, the sure, I could accept that as no great cultural loss (even if a practical one for people who currently need those jobs). Indeed the resulting loss of human control over our own culture would surely be a problem even under socialism.

[zompist]

More importantly, the jobs eliminated by generative AI are not crap that people consider intolerable burdens.

Yes, they are, that's the real danger. Look at the point Moose has been making. No one is going to destroy Art. What the CEOs might do is reduce the amount of human-made craft.

What LLMs are good at is, in fact, writing crap. Probably the vast majority of written materials, and jobs involving writing, deal with producing crap. We normally call it things like "marketing materials, corporate guidebooks, job descriptions, brochures, menus, fundraising, grant proposals, SEO pages, spam, propaganda, articles for minor news outlets, etc."

That doesn't mean it's a good thing to eliminate crap-writing jobs. They are not necessary crap jobs-- some of them are skilled labor, and most of them are more desirable than, say, retail. And there's hundreds of thousands of such jobs (though AI is not going to eliminate them all).

Nobody is complaining about the onerous chore of writing novels

CEOs don't give a shit about novels; they are not going to eliminate novelists.

or drawing pretty girls

CEOs don't really care about that either.

The art-related jobs threatened by AI are creating commercial illustrations, the sort of thing use to illustrate magazine articles. These jobs do not, sad to say, consist solely of drawing pretty girls. A lot of them are fairly mindless work: "make me an illo to go with this listicle on finding hotels."

I doubt we know the economic impact of AI illustration yet. I do worry that a lot of CEOs think that they can fire their artists/freelancers and use Midjourney prompts instead. On the other hand, people may get tired of the AI style. Plus, the cheapness may mean that we get ten times as much art of this sort as when only humans could produce it. That may mean that most AI art would not be putting a human artist out of work; it's adding art to things that didn't use to have any.

[KathTheDragon]

so it makes little sense to evaluate the technology along those lines

I mean, it does, for the whole reason I made the post in the first place. To point out that the mere existence of generative networks is not the problem, the problem is how the technology will be used for cost-cutting. A subtle distinction, perhaps, but I think it's rather significant.

Indeed the resulting loss of human control over our own culture would surely be a problem even under socialism

Apparently you've neglected to read what's been said, yet again, because both I and Raphael explained why this would not happen under socialism. Go back and try again.

[Travis B.]

Has everyone forgotten that the real problem is capitalists cutting costs by exploiting machines that don't need to be paid instead of exploiting people that do need to be paid? If we weren't living under capitalism, the invention of sophisticated generative networks would not lead to this kind of discussion. Instead we'd be talking about how cool it is that we can achieve the same amount with less effort, which leads to more leisure time.

Agreed completely - the real problem here is capitalism, not generative AI. Under socialism, one would not be concerned about one's job being taken away, because in worker self-owned and self-managed workplaces generative AI would be seen as a plus to do away with the drudgery and free workers up to do the kinds of things they really want to do, whether as part of the work they do or as leisure outside of work, rather than as a way to cut costs by putting workers out of work.

[malloc]

CEOs don't give a shit about novels; they are not going to eliminate novelists.

But they do care about paying writers. Why wouldn't they choose to fire up an LLM to write whatever is trending rather than hiring writers? Why would a publisher deal with expensive, unreliable, and potentially controversial novelists over AI that can write? Imagine churning out popular fantasy novels without the baggage of JK Rowling. The same point applies to pretty much any creative job. Human artists are expensive and have an annoying habit of getting into scandals. It seems obvious that corporations would love to avoid all that.

Everyone insists that LLMs can only write crap, but nobody has explained why they can't get better or why newer iterations of AI couldn't build on them. The tech companies are aggressively researching AI and trying everything to boost its power. How do you another breakthrough won't occur in the coming decades?

[Travis B.]

CEOs don't give a shit about novels; they are not going to eliminate novelists.

But they do care about paying writers. Why wouldn't they choose to fire up an LLM to write whatever is trending rather than hiring writers? Why would a publisher deal with expensive, unreliable, and potentially controversial novelists over AI that can write? Imagine churning out popular fantasy novels without the baggage of JK Rowling. The same point applies to pretty much any creative job. Human artists are expensive and have an annoying habit of getting into scandals. It seems obvious that corporations would love to avoid all that.

Everyone insists that LLMs can only write crap, but nobody has explained why they can't get better or why newer iterations of AI couldn't build on them. The tech companies are aggressively researching AI and trying everything to boost its power. How do you another breakthrough won't occur in the coming decades?

Again, the real problem is not AI - the real problem is capitalism.

[WeepingElf]

CEOs don't give a shit about novels; they are not going to eliminate novelists.

But they do care about paying writers. Why wouldn't they choose to fire up an LLM to write whatever is trending rather than hiring writers? Why would a publisher deal with expensive, unreliable, and potentially controversial novelists over AI that can write? Imagine churning out popular fantasy novels without the baggage of JK Rowling. The same point applies to pretty much any creative job. Human artists are expensive and have an annoying habit of getting into scandals. It seems obvious that corporations would love to avoid all that.

No, because people want such things done by real humans. Most people won't buy a novel (or a music album, or whatever) which they know to have been generated by an AI, and the revelation that a popular work has been done by one would cause a major scandal. Ever heard of Milli Vanilli? If not, here's the story: Milli Vanilli was a pop duo that was highly successful at the end of the 1980s - until it turned out that the two performers hadn't sung a single note, it was all recordings of a professional session vocalist. The result was one of the greatest scandals in the history of pop music. There was no AI involved here, but I hope you get the point.

[Linguoboy]

Industrial processes can create flawless diamonds for a fraction of the costs of mining them. Yet people still pay through the nose for natural diamonds. Why?

It is already possible to use AI to generate pop songs which require not one single human participant, not even a singer. Computer-generated pop singers have been a feature of the Japanese pop music scene for other three decades, but they've never really caught on elsewhere. Why not?

[malloc]

Then why are so many artists worried about AI putting them out of work? Why are Hollywood writers striking in part against being replaced with AI if it cannot write a script or people won't go for AI generated film? Are all these people simply paranoid without good reason?

[Raphael]

Then why are so many artists worried about AI putting them out of work? Why are Hollywood writers striking in part against being replaced with AI if it cannot write a script or people won't go for AI generated film? Are all these people simply paranoid without good reason?

There are good reasons to worry. What you are apparently psychologically incapable of understanding is that "good reasons to worry" does not mean the same as "good reasons to be certain of impending doom".

[zompist]

CEOs don't give a shit about novels; they are not going to eliminate novelists.

But they do care about paying writers.

Why should they start now?

I think you have a fantasy, pretty widespread among non-writers, that novelists are big rich celebrities like Rowling. The median income of full-time authors is something like $20,000. (Numbers vary, depending on how the category is defined.) 96% of books sell less than a thousand copies. The median advance for a fantasy/sf book (one of the most popular genres) is $5000 from a major publisher, $3000 from a minor one.

One article I read noted that novel-writing was much more lucrative in the 1800s. No big surprise: much smaller number of writers, no competition from TV and movies.

Why wouldn't they choose to fire up an LLM to write whatever is trending rather than hiring writers? Why would a publisher deal with expensive, unreliable, and potentially controversial novelists over AI that can write? Imagine churning out popular fantasy novels without the baggage of JK Rowling.

Because they wouldn't make any money. When you flood the market with crap, the market becomes unusable, and new methods for decrapping will be invented. (Already people are complaining that Google has gone to hell, as search results are dominated by SEO crap and now AI-generated SEO crap. Google has forgotten what made it successful in the first place, namely, a search engine whose first few results were high-quality.)

You and some others really seem to devalue low-end art, think it can be "churned out", and think its consumers will be satisfied with AI-produced swill. It's like saying that people who eat at McDonald's hate food and love actual shit. No, they like cheap, good-enough food. I don't think anyone's going to get rich marketing romance novels written by machines that aren't quite sure how many fingers humans have.

[Travis B.]

Consider what I do both for a living and as a hobby, computer programming. While I have heard things about how Microsoft's Copilot makes coding much easier, I find that in a relatively obscure language, Forth, ChatGPT at least cannot even write a workable factorial routine (first time I asked it to write one it wrote one which would explode the stack, and the second time I asked it to write a non-recursive factorial routine it wrote one which simply did not work). And this was something as simple as a factorial function! I cannot imagine that generative AI would actually accomplish anything non-trivial (i.e. beyond doing what people used to do manually, i.e. searching StackOverflow) as software development goes any time soon if it cannot write a working factorial function on the first or second try. So for doing the very non-trivial things I do both for my day job (i.e. working on MR image reconstruction) or on my own (i.e. working on a Forth implementation cum RTOS for embedded systems) I do not feel that I will be replaced by a generative AI any time soon. However, it may just put out of work the kind of code monkey who relies heavily upon scraping StackOverflow and other people's GitHub repositories to find answers to their programming questions.

[malloc]

I think you have a fantasy, pretty widespread among non-writers, that novelists are big rich celebrities like Rowling. The median income of full-time authors is something like $20,000. (Numbers vary, depending on how the category is defined.) 96% of books sell less than a thousand copies. The median advance for a fantasy/sf book (one of the most popular genres) is $5000 from a major publisher, $3000 from a minor one.

But the AI costs even less than that, literally nothing. Nobody who writes for a living can compete with something that works for free. Even the most impoverished writer still earns more than any computer. Given the choice between spending $3000 and spending nothing, why would you choose the former?

Because they wouldn't make any money. When you flood the market with crap, the market becomes unusable, and new methods for decrapping will be invented. (Already people are complaining that Google has gone to hell, as search results are dominated by SEO crap and now AI-generated SEO crap. Google has forgotten what made it successful in the first place, namely, a search engine whose first few results were high-quality.)

You and some others really seem to devalue low-end art, think it can be "churned out", and think its consumers will be satisfied with AI-produced swill. It's like saying that people who eat at McDonald's hate food and love actual shit. No, they like cheap, good-enough food. I don't think anyone's going to get rich marketing romance novels written by machines that aren't quite sure how many fingers humans have.

That assumes the products remain crap. As the technology improves, the problems you cite will recede until generated novels and images match or even exceed human-made ones. Airplanes once struggled the leaves the ground. Now even the cheapest airplane can outfly any bird. Even a beater car can goes faster and longer than the best racehorses could ever hope to achieve.

[Travis B.]

I think you have a fantasy, pretty widespread among non-writers, that novelists are big rich celebrities like Rowling. The median income of full-time authors is something like $20,000. (Numbers vary, depending on how the category is defined.) 96% of books sell less than a thousand copies. The median advance for a fantasy/sf book (one of the most popular genres) is $5000 from a major publisher, $3000 from a minor one.

But the AI costs even less than that, literally nothing. Nobody who writes for a living can compete with something that works for free. Even the most impoverished writer still earns more than any computer. Given the choice between spending $3000 and spending nothing, why would you choose the former?

Because they wouldn't make any money. When you flood the market with crap, the market becomes unusable, and new methods for decrapping will be invented. (Already people are complaining that Google has gone to hell, as search results are dominated by SEO crap and now AI-generated SEO crap. Google has forgotten what made it successful in the first place, namely, a search engine whose first few results were high-quality.)

You and some others really seem to devalue low-end art, think it can be "churned out", and think its consumers will be satisfied with AI-produced swill. It's like saying that people who eat at McDonald's hate food and love actual shit. No, they like cheap, good-enough food. I don't think anyone's going to get rich marketing romance novels written by machines that aren't quite sure how many fingers humans have.

That assumes the products remain crap. As the technology improves, the problems you cite will recede until generated novels and images match or even exceed human-made ones. Airplanes once struggled the leaves the ground. Now even the cheapest airplane can outfly any bird. Even a beater car can goes faster and longer than the best racehorses could ever hope to achieve.

The mistake here is to conceive that AI is free. To do anything with generative AI, you need to carry out extensive training and testing and all the powering and upkeep of datacenters, and those things themselves cost money. It likely is cheaper to simply throw authors a few thousand bucks here or there than to maintain anything capable of generating books that humans will actually bother to read.

[zompist]

I think you have a fantasy, pretty widespread among non-writers, that novelists are big rich celebrities like Rowling. The median income of full-time authors is something like $20,000. (Numbers vary, depending on how the category is defined.) 96% of books sell less than a thousand copies. The median advance for a fantasy/sf book (one of the most popular genres) is $5000 from a major publisher, $3000 from a minor one.

But the AI costs even less than that, literally nothing.

Er, no it doesn't. LLMs are made by big business... do you really think that their business model is to provide their product for free? This article might open your eyes: a small developer is paying close to $100,000 a month on LLMs.

And, current pricing is based on the old drug-dealer practice: start by charging nothing, till the customer is addicted.

Nobody who writes for a living can compete with something that works for free.

You're talking to someone who writes for a living and competes with websites that are free.

[malloc]

Er, no it doesn't. LLMs are made by big business... do you really think that their business model is to provide their product for free? This article might open your eyes: a small developer is paying close to $100,000 a month on LLMs.

And, current pricing is based on the old drug-dealer practice: start by charging nothing, till the customer is addicted.

Fair point, but technology gets cheaper over time. Computers in general went from something that only large organizations could afford to something the average American has in their pocket. Eventually generative AI will become cheap enough to compete with writers unless they accept ever shrinking salaries. That business may sound like they're spending a lot, but hiring human writers must cost even more or they would have simply done that.

For that matter, given all the examples of AI generated text and images I have seen, they can hardly cost that much. Is everyone who fills their DeviantArt account with midjourney images really paying huge amounts for them?

[Raphael]

But the AI costs even less than that, literally nothing.

A few rounds of argument ago, you complained about how much money was being spent on AI, and how many good things could supposedly have been paid for by that money.

[alice]

I cannot imagine that generative AI would actually accomplish anything non-trivial (i.e. beyond doing what people used to do manually, i.e. searching StackOverflow) as software development goes any time soon if it cannot write a working factorial function on the first or second try.

A few months ago I was at a demo of some AI which, I was confidently told, could write Java code in the style of someone who had only ever written Python, or it might have been the other way around. Unfortunately other things took over the demo and I never got to see this. Which was a pity; it would have been most instructive.