- This topic has 10 replies, 3 voices, and was last updated 1 year, 2 months ago by
.
-
Posts
-
It was only a question of time. I’m surprised it’s taken this long. Perhaps there are other ‘offers’ that haven’t hit the newspapers.
Writers condemn startup’s plans to publish 8,000 books next year using AI | Books | The Guardian
Yes, it’s rather chilling, isn’t it?
One doesn’t tend to think of Roald Dahl as a sci-fi writer, but he foresaw something like this. Over seventy years ago one of his early adult short stories, The Great Automatic Grammartizator, told the tale of a nerd who invents a machine that generates stories at such a prodigious rate that he takes over the entire publishing world, so that real authors are faced with the choice of either signing up with him and lending their names to artificially generated fiction, or going to the wall. It ends with one of the last authors to hold out crying in despair, ‘Oh Lord, give us the strength to let our children starve.’
At least he wasn’t engaged in the grubby business of exploiting people’s dearest dreams to con them out of large sums of money – which is what vanity publishing is all about.
If you haven’t seen it before this report from April 2022 on paid-for publishing is interesting and depressing. I can’t remember if it mentions AI. A questionnaire was sent to authors who’d paid for an all-in editorial and publishing package. These weren’t indie authors paying for their own editors, cover designers etc and doing the hard work of self-publishing.
Summary:
94% of respondents lost money, typically in the thousands.
• The average loss was £1,861 with some writers reporting losses as high as £9,900.
• The median cost of publication was £2,000.
• A median of only 67 books were sold per deal, resulting in royalties of only £68.
• 59% of writers said their book was not available to buy in retail outlets.Hi @janette
I agree it could bring self-pub into disrepute though the large number of self-pub books on Amazon that seem not to have had human editing, regardless of the availability of AI, is already doing that.
The publishing industry can seem very off-putting. I’ve never tried to be published by it so I haven’t been through that particular mill. There are small indie publishers that inspire hope such as Galley Beggar, CB Editions, Influx Press. Stinging Fly.
I’m not surprised by this. For some time there have been AI editing features available with Prowritingaid, Grammarly and so on. I have the former, and although I seldom use the AI features, some of which are behind an additional paywall, I have given them a spin in the past. What you get from one of the reports does appear to be a reasonably detailed assessment of the text together with suggestions that seem reasonable and, dare I say, useful.
Let’s be clear though. The report is, in essence, a sophisticated “checklist” of things to consider. It isn’t a response to the text. If there is intelligence in the process, then it is the intelligence of the programmers, the critics who have shaped its responses (willingly or otherwise). You will not get anything beyond this generic output. You won’t get the kind of insight and commitment that a real critic will offer. But, and this is where profiteers make their profit, it is quite enough to convince the beginner that they have value for money.
Yes, it’s the old vanity approach of taking advantage of unaware writers’ naivety. There’s a common belief that editing more or less equals proofreading without so much as a beta read along the way. I imagine all vanity publishers will be using AI in the same fashion.
I think what’s notable about Spines is the number of books they plan to produce: 8,000 in the first year. It’s also a short step to using generative AI to generate stories, from plot to sentence level. The Guardian article doesn’t suggest that will happen but it seems inevitable.
I’m hardly knowledgeable enough to comment but the distinction between assistive AI and generative AI seems to be disappearing. Even if all you want is a spellcheck, it will be done by gen AI. I imagine it doesn’t make sense from the tech point of view to have two different systems when one can do everything incorporate machine learning to boot.
On a day-to-day level my concern is environmental – the vast amount of electricity and water involved in powering gen AI.
As we are in imminent peril of being swamped with things-AI if Mr Starmer and our present government are to be believed, I thought I’d bite the bullet and give ChatGPT a whirl. As an experiment, I thought I’d try for a summary of my WIP. Since that’s something I will have to wrestle with when I come to submit, I wondered whether it might be useful. I first checked carefully and was assured that any files submitted are deleted at the conclusion of the session. I thought this would be the case, since even with the relative cheapness of storage these days, it would be hugely expensive to store the hundreds of thousands, even millions of files uploaded.
After loading my file and typing ‘please summarise this’ (note the automatic politeness), it took ChatGPT about 15 seconds to start delivering a summary to my screen. I have to say I was impressed. Although it was relatively light and didn’t delve into subtle nuances, it was quite comprehensive, even detailing character development and so on. For 85K words it produced a summary of around 2.5 A4 pages. I did notice the occasional oddity: very slight deviations from the actual story almost exactly what one might imagine if a reader had skimmed the text and guessed how some parts panned out from how similar plots progressed, rather than how this one went. Curious.
So far, so good. I thought I press on and requested a detailed synopsis. In some ways this was more revealing. Although it was 75% a decent synopsis, it was 25% plain wrong. Some key plotlines were way out. A confrontation between the MC and a criminal gang was outlined as a major event, but never happened and isn’t described anywhere in the text. I suspect, although I haven’t got evidence, that ChatGPT will have difficulty with plotlines that contain deception, or are ambiguous. I suspect that it has been designed to look for certainty.
Interestingly, both summary and synopsis completely omitted what was very obviously the climactic event. The scale of the omission is a little like describing Trafalgar Square but overlooking the lions and the pointy thing in the middle. Not sure what to make of that.
Useful? Maybe. Reliable? Not yet.
This is very interesting, Ath. Have you tried giving it a word count, e.g. to produce a 300-word synopsis?
I tried a 500 word synopsis. It was curiously like a “bad” synopsis that an inexperienced writer might try, too much detail at the outset and then broad generalisations for the middle and end. It always concludes by asking whether you want any refinements, so I asked for more detail about the conclusion and it invented a character and several plot lines! I asked why it had embellished the story and invented parts and it and the reply was that it had done this because the end was missing.
I don’t know whether this was a result of the cache for an individual session being limited, i.e. it literally didn’t have access to the end, or whether it couldn’t deal with an “open” conclusion. I suspect the former, because it’s summary on a previous occasion was more complete. It may depend on how many users it has.
I asked it about limits and it recommended breaking text into chunks, so I divided the input into blocks of chapters, but as I was about to do the final block I hit an error, which I think was the limit of my free access: there was a message about that soon afterwards. Annoyingly inconclusive! Some chapters were summarised in full, others were missing nearly all the action. No idea why. When my free access starts again, later today, I’ll finish the synopsis and see if the individual chapters with missing detail can be improved.
I asked it for a 200-word synopsis for my 2,200-word Den Challenge story (which I haven’t posted on the Den yet). The result was mostly a true reflection of the story – about 90% right I’d say. It had the minor character’s motivations wrong; they went in the opposite direction to what I’d written, unless of course I haven’t described them clearly enough. The story’s ending was correct in a general way: factually recognisable but lacking the emotional connections which I hope a real reader would make. But successful short story endings are often subtle and potentially open, and my ending may not be up to scratch, in which case I can’t blame ChatGPT for not picking up my draft’s wayward clues.
- You must be logged in to reply to this topic.