So you might find AI a shortcut, or inevitable… at the same time, it’s probably terrible for your own craft, and might actually be closing doors (such as editors just turning you away). Excellent article, well worth a read.
From the article “First, she defended gen AI and its inevitability, an angle that reads like coercion.”
Reminded me of the movie “Invasion of the Body Snatchers.” Why fight it, we will all eventually be consumed.
The only problem is that no human will want to read that mind numbing dreck.
1 Like
If that’s true, what are you worried about?
Then again, have you taken a look at the content served up by social media feeds lately? “Mind numbing dreck” would appear to be EXACTLY what most humans want to consume.
How did you arrive at that conclusion, that I was worried?
I spend very little time on social media. But I would suspect that is a case where the technology is advancing but the people using it are not.
1 Like
I am worried because bandwidth is a thing: While many folks may choose to ignore AI content and (IMO wisely) read elsewhere, editors, publishers, marketers are drowning in AI slop. A real human author might just get lost in the noise.
I was working on the assumption that if you did not care you would not be trying to convince people of things. Not an infallible logic chain, but usually reliable.
I am not a fiction writer. I use AI as a research assistant etc, it is not a co-author, I have several human proof readers and editors.
Congratulations?
Claude, Siri and… that old mucker Chat Gipity, aye?
<tongue very much in cheek!>
Another problem is that fiction writers and poets have relied on day jobs doing boring corporate writing for years. That’s the kind of writing that AI actually can do “well enough” to convince the executives with the checkbooks.
4 Likes
I rarely use search engines these days except for the simplest things. Why? Because they are not much good for semantic searches. So I use AI for research, such as what I am currently investigating—the pros and cons of Hittite laryngeals as evidence that something resembling the reconstructed Proto-Indo-European system existed. Or I can ask it to go to a URL and find the information I am looking for by giving specific semantic instructions, very useful if the site has hundreds of pages. A site search engine can find key words but not ideas.
For that kind of question, I’d want to talk to a human reference librarian and probably a scholarly database. I’m not at all surprised that general search engines are useless.
2 Likes
Technical writer here, specializing in IT Security and Identity Management, that lost his job last year to a Generative AI dependent contractor.
My local public library has been more or less turned into a homeless shelter. My local university though small is the world leader in geology and environmental rehabilitation, neither of which are of interest to me in my research. It doesn’t have what I need, I already checked. The nearest university with a sizable research library is 250 miles away. I do like the feel of walking through the stacks and seredipitously coming on an important gem like the time I did at San Diego State University in the early 90s where I came across an important quote by Polyhistor preserved in Clement of Alexandria’s Stromata. But I do not feel I am limited by my remoteness because so much has now been digitized, plus having JSTOR access and other scholarly databases. Though I prefer to read physical books it is a lot easier to search a digital one. That is why I have an arrangement with a prominent scholar in my field who trusts me, that he give me a PDF copy of his book for research purposes after I show him receipt of purchasing the physical copy—he doesn’t publish digital versions.
This could be corrected from “AI-generated work” to “more competition”. Ever try to make a living in a city with a local college or technical school for your field? AI or Human, a glut of talent all trying to find their way will depress prices everywhere.
I’m not unsympathetic, and I’m not saying your wrong. What I should be saying is that AI is a tool, nothing more. How the tool is used is as important as how many get to use the tool, and at what price.
Sometimes I think people stop looking at the causes of problems after they can point the blame in convenient directions, so that’s why I bring it up here. No shade towards you @Bebop, just reminding those that need it.
Then you know – probably better than me – just what sort of changes in mental processing you have to do in order to produce a working prompt that delivers the expected results. I’m not sure its all that different than learning Assembly Language, C, Fortran, COBOL, Ada…
I could go on, but I left out one important language that is still useful today: Loglin and Lojban it’s counter-culture modern derivative.
When you restructure the language to talk to the model mathematically[1] something as archaic as Loglin suddenly has value again. Just like in Heinlein foresaw in ‘The Moon is a Harsh Mistress’, the best way to feed instructions into the ‘thinking computer[2]’ is to take out every piece of ambiguity that you can. And that has been true even before LLMs started making the process easier for the masses. Art tends to live in that ambiguity, and that is one reason AI prose or imagry tends to feel soulless.
I’ll draw another parallel here… Once upon a time, you had to have an extensive foundation in computer science knowledge to understand and operate a computer. They were expensive, cantankerous, and a really good way to destroy productivity. In the 1960’s they had a not-dissimilar effect on trained calculators[3] as we are discussing now. Over the 1970’s a lot of work went into refining and simplfying them until we got the first wave of pratical home computers in 1977. They cost more than good used cars. Over the 1980’s a lot of work by a lot of people went into making the computer understandable with less of a learning foundation, until it was practical to market the computer to everyone (OSX, Windows 3.x/95). Then a lot more work got put into things so that these days while you still get error messages they no longer mean anything at all.
Since the dumb-operator doesn’t understand hex addresses, there is no reason to include a hex address with the error messages like DOS did. Since the dumb-operator doesn’t understand IRQs there is no reason to share the conflict information. Since the dumb-operator doesn’t understand the driver-conflict, there is no reason to show them anything at all. Most technicians know to seek out the logs and look there for answers anyway.
In fifty short years we have gone from a highly technical tool of industry to a hobbyist’s passtime to inside-every-first-world-person’s-pocket. And each stage of progress has stair-stepped down how much the computers need from their operators in order to produce useful results. AND stair-stepped up how much is going on behind the cursor that is not seen by the operator let alone understood.
Those changes can be scary, but at the same time, it might just be liberating to others. I for one would rather they stop with “An error has occured. [OK]”
I don’t know what AI will bring or take away. I’m not that smart (else I’d be making money on that rather than driving a truck hauling food for a living…. Then again, I stay working through COVID so I’m not so sure I didn’t weather that storm better than some). I do know that generally speaking allowing more people express themselves generally makes art more desired and accessible.
Perhaps we should look back to the Victorian era when the combination of changeable type printing presses and high-rag content paper made books started making it so that FICTION was suddenly available to the masses. With more customers they needed more writers. With more writers we had more stories. With more stories we had more inspiration for the following generations…
Yes, I’m being optimistic. Yes, I’m being simplistic. Yes, its projecting as wish-fulfillment. But I can be honest and cynical when the time comes. I could wish for better transparency with how the models are trained. I could wish that AI makers are not destroying books enmasse for their training data. But… I enjoy being able to Google an obscure scene from a book who’s title and author I cannot remember and frequently get an answer as to what book it’s from.
If you’ve gotten down this far, thank you. The TL,DR is: Pandora’s Computing box was opened fifty years ago, and hope is still inside.
I’m over simplfying here, but since the models ONLY work on mathematical probabilities it fits close enough for this discussion ↩︎
LLM’s don’t think, regardless to what they label their ‘working narrative’ ↩︎
Mathematically trained technicians specialized in computing advanced and complicated equations. A highly respected and valued job title, not the little box in your pocket. ↩︎
3 Likes
Hey, @pigfender how much of Sturgeon’s Law do you suffer through for NiaD?
I’m curious because as we are talking about slop flooding the bandwidth, no one has really mentioned slush doing similar things.
While slush ≅ slop, it’s not the same, but it transports the same. And anyone that has ever been tasked with working through a slushpile still needs to wash their hands before they eat.
Anyone else see much difference between 90% of Human authored prose and 99+% of Gen AI?
None.
I don’t know if Sturgeon’s law (“90% of everything is rubbish”) is a good approximation for life in general. I certainly hope it isn’t, and it doesn’t really ring true to me. If you flip it round, I don’t think something needs to be in the top 10% to be considered passable. The top 10% of something in my experience is usually pretty good. I would say that at least 98% of my team are good, and the top 10% of them are absolutely flippin’ incredible. I’d also suggest that I’ve enjoyed damn near all and certainly far more than 10% of the {use your imagination} I’ve had in my life.[1]
But even if you ignore all that and take Sturgeon at face value… in NiaD’s case we are already jumping through a bunch of selection hoops that mean we’re not really dealing with a group that represents the “of everything”. People have to have been interested enough in writing to have found us, for one thing. Then they’ve got to have been confident enough to have put their hat in the ring and share a writing sample before even being accepted (an even smaller group). Then they’ve had to had the level of skill, discipline, interest, focus and talent to sit down and write a creative brief to a hard deadline – an even smaller subset. If you pass those, you’re probably already in the top 5% of people who’ve ever had the thought “I’d like to write a novel someday”, talent wise. I think the books we’ve written together support that. I’ve always been incredibly impressed (and in some cases outright blown away) by what people submit.
But, yeah, Sturgeon’s Law might be fair in the sense that writing is a rarer talent than people think it is, and if you’re able to do it to a professional standard you’re going to be a less than 1-in-a-100 talent…with even scarier numbers if that’s actually how you pay your mortgage.
To use a different example… (I kinda have to because, to the best of my knowledge I’ve not read any long-form AI-generated fiction), since the introduction of genAI tools the number of applications we get for every job advertised has gone up enormously. I mean, it’s maybe 50 times the applications now. The standard has not raised at all. We get exactly the same number of qualified credible candidates per post as we always did. We always got a good number of ‘slush’ applications, but the number of people who will now have a punt because it’s zero effort to have AI write an application is really annoying, because someone – an actual human (I still insist on this) – has to read, digest, and try to understand the person behind every application.
steak, I was talking about steak ↩︎
1 Like