I wrote about AI and its writing prowess over a year ago in this column. Many of those of you who commented then felt that ChatGPT’s writing was soulless and repetitive and unlikely to usurp that of us humans.
I thought it would be interesting to share an update and see if your feelings have changed.
I asked ChatGPT to write me a ten paragraph book review of Lisa Kleypas’ classic Devil in Winter. Here’s what it produced:
Lisa Kleypas’ “Devil in Winter,” the third installment in her Wallflowers series, offers a refreshing take on the Regency romance genre. The novel follows the unconventional union of Evangeline Jenner, a stammering wallflower, and Sebastian, Lord St. Vincent, a notorious rake. Kleypas’ nuanced character development and richly detailed historical setting make this a standout in the crowded field of historical romance.
Evangeline, or Evie, is not your typical romantic heroine. Desperate to escape her abusive relatives, she boldly proposes a marriage of convenience to Sebastian. This unexpected move sets the stage for a story that is as much about personal growth as it is about romance. Kleypas handles Evie’s transformation from a timid girl to a confident woman with great sensitivity, ensuring that her evolution feels both realistic and compelling.
Sebastian begins the novel as a charming yet jaded aristocrat, more interested in pleasure than in commitment. However, as the story progresses, Kleypas peels back the layers of his character, revealing a man capable of deep affection and loyalty. This gradual revelation is handled with a deft touch, making his journey from rake to devoted husband believable and satisfying.
The chemistry between Evie and Sebastian is expertly crafted. Kleypas excels at creating tension and intimacy between her protagonists, making their developing relationship feel authentic. Their interactions are marked by witty exchanges and genuine emotional connection, ensuring that the romance is engaging without being overly saccharine.
The historical setting of “Devil in Winter” is another of the novel’s strengths. Kleypas paints a vivid picture of Regency England, from its grand ballrooms to its seedy gambling halls. This rich backdrop not only enhances the story’s authenticity but also immerses the reader in the period, providing a compelling context for the romance.
Secondary characters play a significant role in the novel, adding depth and complexity to the narrative. The loyal Wallflowers—Evie’s close friends—offer support and camaraderie, while figures like Ivo Jenner, Evie’s father, bring an element of danger and intrigue. These well-drawn characters enrich the story, creating a fuller and more immersive world.
Themes of redemption and personal growth are central to “Devil in Winter.” Both Evie and Sebastian must confront their pasts and insecurities to forge a future together. Kleypas explores these themes thoughtfully, adding a layer of depth to the romance that makes the characters’ journeys feel meaningful and rewarding.
Kleypas’ prose is both elegant and accessible, striking a balance between vivid description and engaging dialogue. Her writing flows smoothly, making “Devil in Winter” a compelling read from start to finish. The narrative is well-paced, with a mix of action, dialogue, and introspection that keeps the reader engaged.
The emotional depth of the novel is one of its standout features. Kleypas does not shy away from exploring the darker aspects of her characters’ lives, which adds a sense of gravitas to the story. This depth makes the characters’ eventual happiness all the more poignant and satisfying.
“Devil in Winter” is a prime example of how historical romance can be both entertaining and substantive. Lisa Kleypas has crafted a novel that not only delivers a satisfying love story but also explores themes of resilience and transformation. Fans of the genre will find much to admire in this well-crafted tale.
In a genre often criticized for formulaic plots and flat characters, “Devil in Winter” stands out for its rich character development and historical detail. Lisa Kleypas has proven once again that she is a master of her craft, delivering a novel that is as emotionally resonant as it is enjoyable. This book is a must-read for anyone seeking a romance that goes beyond the ordinary.
Is it as good as Lea’s review? No. Absolutely not. But, if I hadn’t told you, would you have thought it was by a bot? I’m not so sure.
What do you think? Have your feelings about AI changed since last January?
Still a big nope from me. I’m with whoever said – I want AI to do the stuff like the washing up and the ironing so I can write that novel or symphony – not for AI to do the good stuff while I get stuck with the chores!
Well, I that whatever we wish AI would limit itself to is unlikely to define what it will actually end up doing.
True, I was just saying I enjoy thinking for myself and don’t plan on using it if I can avoid it.
I’m not worried about us wanting to use it. I’m worried about it displacing most human work!
AI needs humans to do the writing, creating so that it can take all of that and produce a composite work. I think what AI does in the creative realm is totally derivative. If there had not been a single review of Devil in Winter, AI would not have been able to write a single sentence about it let alone ten paragraphs.
I could not agree more. It’s theft all the way.
And it’s getting to the stage where we ALL have to be incredibly careful about what we’re signing up to. Authors are having long drawn out battles with publishers to get anti-AI clauses into their contracts (I remember KJ Charles said she’d been arguing with one for months) – which you’d assume they’d be bending over backwards to do in order to protect their intellectual property- but no.
To be honest, this one reads like it was written by a AI machine. Not a single original thought. I am not sure it actually read the book. It is a collation of all the reviews written by humans that can be trolled on the internet. I was almost tempted to call it out for plagiarism!
You’re right. It’s a collection of stock phrases that reminds me of the new AI reviews at Amazon. “Customers say…”
Maybe because I – and AAR readers – look for something more nuanced and personal we notice something that is the written equivalent of a plank of wood where others might not.
Exactly, Caz. The review is smooth, but it doesn’t have the personality I enjoy in the reviews I trust. There are some reviews here that have made me laugh uncontrollably, like the one about how the hero must need to weigh down the hem of his kilt. When AI does something like that, I’ll be interested.
I agree AND I think, with more prompts, I could have made AI write this with more personality.
Then it is not AI doing any original work. It is just a machine being pushed into doing what you want. If you want blah, AI produces blah! If you want humorous and teach it how with prompts, it will definitely produce a humorous review. But then, what do we need AI for?
I suspect that businesses will want it because it’s a cheaper way to do most everything.
Modern AIs by definitions don’t have original thought. They answer any questions by combining and transforming texts written by humans. They are capable of producing something “original” by analogy – say, combining a couple of existing recipes in a new one.
But a review is an opinion. There is no analogy for it – the book is original, there’s no recipe for reviews. The only way an AI can produce a review in this situation is to reach into its database and combine existing reviews. They have been retransformed several times and so what comes out is not exact phrases as we would define “plagiarism” but it’s definitely a copyright issue because the AI model takes content produced by others and doesn’t pay them for it.
Sorry to intrude in the discussion, but I have no other mean to notify that the power search is not working properly (i.e. missing authors, missing books, missing reviews …) again I apologize for intruding
Yes. I know. I am working on it. Thank you. It’s making me crazy!
I think this reads like AI. If you are looking for a generic overview, then maybe it would be fine. But if I’m looking for a real review with information that would help me decide whether or not to read this book, this is all but useless. You’ll notice it never delves into any actual aspects of the story – doesn’t mention characters beyond the surface or discuss any plot points either good or bad. At this point in time, at least, I’m not too concerned about AI replacing me as a snarky reviewer!
On a side note but AI related, I’m going to share this link to a NYT Daily report about a problem in schools with kids (mostly boys) creating and sharing deepfake nudes of their female classmates, created by AI. It’s a real problem that we need to press our local, state and national lawmakers and schools to address. AI is an amazing tool, but we need to be on top of how it can and is abused and make sure we are taking action to control it.
Unfortunately, I think it’s too late to close the stable door. A year or so ago, I saw quite a few articles and discussions about this aspect of AI in a number of national newspapers, but it seems to have gone quiet lately. Possibly because no-one, from political leaders to tech-gurus – has a clue what to do about it. As is typical these days, nobody seems to actually think anything through before we’re hit with it. My favourite saying when this kind of thing happens is from Jurassic Park (the movie), when Jeff Goldblum’s character says that the scientists were so busy working out if they could do something, that nobody stopped to think whether they should.
I know people whose livelihoods are already threatened by AI – and I’m vehemently against it for any creative endeavour.
I felt that way myself – that AI should NEVER be used for any creative endeavour because 1) it is generated by using existing creative work created by humans (IP theft!) without compensation or credit and 2) it could and will effect peoples’ livelihoods and 3) there are a lot of ethical concerns on how and when it is used. And I have said to my husband many, many times that very thing – they are so focused on whether or not they CAN do something that they don’t ever ask if they SHOULD do it with no satisfactory answer. I blame this on most of these AI tech people being men and being greedy. Always the worst combination for humanity.
That said, I’ve come around a bit. First, AI is not going anywhere, so we need to figure out how to work with it. To regulate it and use it ethically AND to compensate those whose original creativity is used to train AI. Second, there are things that AI can do that are useful tools, and we shouldn’t throw out the baby with the bathwater. And I’ve come to realize that using AI effectively is actually an art unto itself. It can serve as a starting point or a base from which to begin. For example, the sample review Dabney created is pretty bad, but a person looking for angles could use it to give themselves some ideas of things to discuss when writing their own review. Kind of like a prompt.
The real problem is that our governments are not incentivized to keep up with the new technologies as far as regulating and controlling them, which in the case of AI is imperative. We can’t afford to wait until there are crisis situations (like massive loss of jobs to AI when greedy men don’t want to pay human workers) before they finally act.
I think AI will create UBI unless governments step in now.
When my brother and I were kids, we received a big container of plasticine (modeling clay of the pre-PlayDoh era) in different colors. We decided we would mix up all the clay to have a bigger amount to play with—but, of course, that meant all the colors got mixed, and what we had was a big ball of grayish-brown clay. After we finished playing, my brother turned to me and said, “Can we get the different colors back now?” That’s my feeling about AI: it takes all the color from creative enterprises and turns it into an enormous ball of colorless blah—and we’re going to end up without the brightness of true creative endeavors. How this gets resolved I don’t know, but I loved this comment on the copyright page of Garrett Leigh’s HOUSE OF CARDS: “The author grants ZERO rights for any and all AI reproduction/training/bot use of this material. Write your own shit.”
As the mom of four millennials, I worry hugely about AI producing mediocre results that are good enough for companies to use AI rather than people. Most consumers will take cheap over quality any day. It terrifies me that we will have shit content and no work.
I think a lot of people are saying the same thing I’m thinking. It says all the words but there’s no passion there. In simpler terms there’s no real personal opinion, just a lot of stock phrases from elsewhere.
It’s something I’ve been noticing a lot online lately, especially on recommended feeds. I’ll see an article title that looks interesting but when I actually start reading my eyes glaze over. Not because it doesn’t have information about the topic but because it’s nothing new that I don’t already know. It’s all canned phrases and generic blah, blah, blah. No “this is horrible” or “this is greatest thing I’ve ever come across”. Nothing to say that a human being has “feelings” about the topic.
And just to be extra fair, I went back and reread it again to so see if I could figure out exactly what type of romance plot was in the book, i.e. friends to lovers, rivals, etc. The only thing I could see was that it might be a marriage of convenience but I’m still not sure. So yeah, it sound AI.
I just got this article in my inbox.
Whoa.
Great article – thank you for sharing. I, too, have a gut-deep fear of what AI may do to all of us in the future. And I tend to be a doom-and-gloom kind of person. But my husband is quite the opposite and takes the view that humanity has adjusted through so many major changes that appeared at the beginning to be extinction-level (metaphorically) but didn’t end life as we know it. AI is very scary, but it’s also really cool. If you scroll down the article to the last section – starting with “To conclude…” I’ve decided to adopt that attitude. AI is not going away. The genie is completely out of the bottle. We are going to have to figure it out and adapt.
I agree AND I also think the way governments currently work, AI may cause huge economic pain for most in the short run at the very least.
A friend of mine’s father-in-law died last week and apparently the obituary was written by AI. I guess I don’t know how I feel about that; most obituaries are pretty workmanlike writing anyway, so maybe that’s somewhere AI can play a role. And I’ve noticed that Amazon is using AI to get the gist of reviews and, I guess? But yeah, this review above feels like something 10th graders write to cover all their bases. It doesn’t have a lot of nuance, or style in the writing, or any real opinions.
Also, I just recently heard about that Apple ad for their new ipad. It’s shocking that with how people feel right now about AI’s impact on creative endeavors that they thought a ad called “Crush!” was a good idea. If you haven’t seen it, it’s actually kind of a hard watch: https://www.youtube.com/watch?v=ntjkwIXWtrc
Apple clearly lost their mind there. So. Dumb.
I just watched the ad. It’s really pretty frightening.
To me, the 2nd half of the review read like: depth, depth, layers of depth, depth. Reviews are interesting when they contain examples or specific details from the book and when there is a sense of the reviewer’s personality. These were particularly lacking here.
This was very boring – I couldn’t get past a few paragraphs. I think written AI content will be used by cheapo places, but I don’t think it will save significant money because: A) eventually no one will want to read a site if all the content is like this; B) there are already legal battles for the sites that have been scraped (the New York Times is suing an AI company for using its stories, etc); and C) the internet is becoming less and less valuable as more sites are filled with ads and garbage AI content. Eventually AI will be scraping other AI articles – and it will be like what happens if you keep making copies of copies (unusable). From what I’ve read ‘creative’ AI content has been wildly overblown by tech companies heavily invested in it, and also by tech reporting that tends to make anything new sound inevitable and innovative. Anything can happen, but I think soon it will more likely be used by bad actors (people wanting to influence social media, elections, deep fake nudes, etc) than legitimate companies.
We’ll see. I worry that the way the global financial system works that anything that leads to fewer employees will be eagerly embraced by almost all employers whether the content it good or not. Think the airlines!
This is not a review. At best, it’s pieced together from real reviews. It’s regurgitated, offering absolutely nothing new. I don’t care how much they can or can’t make it sounds like a real person writing it. Of all the places “AI” has so business being, pretending to convey actual human reactions has to be at the top of the list.
P.S. You could tell me that this was a trick and the review was actually written by a person and my opinion would still be exactly the same.
My biggest issue (in addition to the IP theft) is that AI uses so much energy: https://www.scientificamerican.com/article/the-ai-boom-could-use-a-shocking-amount-of-electricity/
Plus its not really intelligent at all and people seem to believe it is. To quote an astrophysicist friend, its just basic regression analysis on a massive scale. Its the latest hype like “smart tech” and “21st Century” and “digital”
Thanks for that article. Fascinating!
I recently read an article that said AI was actually just “predictive text & autocorrect on steroids.” Sounds about right.
I have the depressing feeling that there could be a healthy market for AI books. I have picked up (and DNF’d) so many books that feature a heroine who is widowed/divorced with a child and leaves her high-powered job in the big city for a small, scenic town where she finds love/solves crimes/does both. Or a series in which a heroine with an odd job solves crimes while being torn between two suitors, one of whom is a cop.
Yes, there are good books using those set-ups, but there are heaps of mediocre or worse ones that garner hundreds of good reviews on Amazon. Will those readers notice that a book was written by AI? Will they care?
And the electricity issue worries me too.
I think you’ve spotlighted the main issue:
(If it’s cheap) Will readers care?
Here’s another example.
My prompt: Write me the first chapter of a contemporary romance where a man, age 39, who has two kids and was recently widowed–it was a tough death, the meds she took made her hateful–meets a woman, age 35, when she, by mistake, runs into his minivan in the grocery store parking lot.
Chapter One: Collision Course
The morning sun cast a golden glow over the sleepy town of Crestwood, where the streets were lined with charming, albeit slightly worn, Victorian houses and the scent of freshly brewed coffee lingered in the air. In the grocery store parking lot, Michael Harris was loading groceries into the back of his minivan, trying to ignore the gnawing emptiness that had become all too familiar over the past year. He was 39, a father of two, and a widower navigating the murky waters of grief and single parenthood.
His late wife, Laura, had fought a long and brutal battle with cancer. The medications had taken a toll, turning the loving woman he had married into someone bitter and angry. Her passing was a mixture of heartbreak and relief, a complexity of emotions that left Michael feeling guilty and alone.
“Dad, can we get ice cream?” his daughter, Lily, asked, her big brown eyes hopeful. She was seven, full of questions and innocence, unaware of the weight her father carried.
“Sure, sweetie,” he replied, managing a smile. “Let’s finish packing these groceries first.”
As Michael turned to grab another bag, a loud crash echoed through the parking lot, followed by the sound of metal scraping against metal. He spun around to see a small sedan embedded into the side of his minivan. A woman, looking mortified, stepped out of the driver’s seat, her hands trembling.
“Oh my God, I am so sorry!” she exclaimed, rushing over. She was about his age, maybe a bit younger, with dark hair pulled into a messy bun and wide, hazel eyes filled with panic.
Michael took a deep breath, trying to keep his frustration in check. The last thing he needed was more complications. “Are you okay?” he asked, his voice strained but calm.
“Yes, yes, I’m fine,” she stammered. “I just… I didn’t see your car. I was distracted and… I’m really sorry.”
“Let’s take a look at the damage,” he said, walking over to inspect the vehicles. The side of his minivan was dented, but it seemed mostly superficial.
“I’ll pay for the repairs, of course,” she said quickly. “I have insurance.”
“It’s not about the money,” Michael said, shaking his head. “It’s just… it’s been a rough year.”
“I’m really sorry,” she repeated, her voice softer now. “I didn’t mean to make it worse.”
Michael looked at her, seeing the genuine remorse in her eyes. “It’s okay. Accidents happen. I’m Michael, by the way.”
“Rachel,” she said, offering a tentative smile. “Let me at least buy you and your kids something from the store. It’s the least I can do.”
He hesitated, glancing at Lily and his nine-year-old son, Jake, who had been silently watching the scene unfold. They both looked curious and a little excited at the prospect of an unexpected treat. “Alright,” he finally said. “But only because you offered.”
As they walked back into the store, Rachel tried to make small talk. “So, are you from around here?”
“Yeah, lived here most of my life,” Michael replied. “How about you?”
“I moved here about six months ago,” she said. “Needed a fresh start.”
Michael nodded, understanding the sentiment more than she could know. They picked out a few items, including the promised ice cream, and as they stood in line, he felt a strange sense of calm wash over him. It was the first time in a long while that he’d felt anything close to normal.
When they got back to the parking lot, Rachel handed him her business card. “For the repairs,” she said. “And, if you ever need anything else… I’m a good listener.”
“Thanks, Rachel,” Michael said, pocketing the card. “I’ll be in touch about the van.”
As she drove away, Michael watched her car disappear around the corner. He couldn’t help but feel that, despite the dented metal and the chaos of his life, something good might have just collided into his world.
“Collided into”??
I basically came to say that looks like something I might have written when I was 12 – then noticed the glaringly incorrect syntax at the end.
I know when I’m involved in an accident, the first thing I check is the age, hair colour and eye colour of the other person. The damage to my car can wait!
Paging Sir Basil Exposition: AI is trying to take your job!
Another thing to remember about LLMs is that prompting matters. It’s rarely just a one and done thing. You have to be precise to get what you want. And it ends up being a conversation as you refine. So, this (Dabney’s prompt for chapter 1) may be a starting point and then you continue asking for refinements until you have what you want (or close enough). I think we’re a ways off from humans being sidelined by GPTs, but they can create efficiencies that make work easier.
I’ve been thinking about this quite a bit. It’s possible that the New York Times lawsuit will change things substantially, depending on how deeply it cuts into available data for training, and there’s some speculation now that AI development will hit a ceiling just by running out of usable data (see here for a discussion of that issue, which I got to from Zvi Mowshowitz’s Substack — that’s the blog you want to read if you want really, really thorough discussion of AI developments and their implications, by the way).
Current regular-people-facing AI does seem to have a garbage-in-garbage-out problem: I ask Perplexity questions here and there, and it’s frequently wrong, not because it’s hallucinating but because its sources, helpfully linked, are also wrong. This is kind of illustrated by Dabney’s prompts — if you train an AI on desultorily written, unimaginative stuff, it’s going to put out desultorily written, unimaginative stuff. A few years ago I had a side gig writing for a site that wanted the occasional article to be available alongside apartment search listings — “How to Draft a Roommate Agreement,” “A Guide to Portland Neighborhoods,” that kind of thing (I’ve never been to Portland). AI was not available when I was writing them, every word was 100% mine, and if you went back and read them you’d find them indistinguishable from AI writing today.
I think the answer to Dabney’s question — “If it’s cheap, will readers care?” — is “Some won’t, because they don’t need a lot from their books, they just need a few tropes and descriptions to stimulate their own imaginations.” There’s an algorithmic quality to romance writing anyway — look at stuff like Gwen Hayes’s Romancing the Beat (which I’ve read, and like!), that breaks down exactly how much story time you should pass before the first kiss, the dark moment, et cetera. A bot that can hit those beats will satisfy a lot of readers. (Also I imagine that pretty soon the story engines will be good enough to meet readers’ microtrope requests — i.e. “Write me a story in which a paramedic with visible scarring from a previous fire has three different men fall in love with her: a chef who insists on cooking for her, a professional hockey player who calls her ‘Sugar’, and a lonely billionaire who helps her restore her dad’s 1969 Karmann Ghia.”)
I think–I hope–that writers willing and able to do truly original stuff are less threatened by AI. There’s not an LLM yet that can hallucinate anything as insane as the plot and character behavior of Colleen Hoover’s Verity, or use a calligraphy/graphic design metaphor as fully as Kate Clayborn does in Love Lettering, or come up with a voice as compelling, as, say, this one pseudonymous Tumblr writer who recaps Junji Ito books. (Spoilers for the books, but seriously, if you have any stomach for horror at all, it’s worth reading Uzumaki and Black Paradox just to enjoy the reviews.) It will be hard for those writers to stand out from the more run-of-the-mill stuff, but then, that’s already true.
Fascinating. Thanks!
I think the other key piece is money. AI generated stuff is so much cheaper to make than the stuff we humans churn out. It offers, I worry, the ultimate outsourcing.
I think (hope?) that AI will not impact for long romance novels or even book reviews. The reason I think this is the paranormal market. When a few really great paranormal novels came out, people were buying them in droves. Every other book published seem to be a paranormal. The books were being written by fans of the genre and even the bad ones at least got some of the ambience and world building right. Then the authors looking to write to the market showed up. Their books were the very definition of mediocre. They didn’t understand what the readers wanted and delivered subpar material. People became increasingly dissatisfied and only those authors with an existing fanbase survived. The market moved on and so did the writers who were writing just to cash in on the trend. Doubtless destroying another trend as they did so. I think AI will be like those writers. It will put out content that can’t compete with the good stuff and someone will self publish a book that has people talking and that cycle will repeat till publishers get the point.
It may well impact the job market but I am hopeful the same thing will happen there. But I could definitely be wrong.
I thought the AI generated review was anodyne and full of purple-ish prose. I expect that AI stuff will be out there in the market and as the romance genre, and its readers, is so often looked down on there is a presumption that we don’t need the “real deal”, we just need a continuous supply because artistic quality and originality go over our poor tiny little girlie heads. For me this is like halal chicken in the supermarket: it’s fine but make sure it’s correctly and visibly labelled so that I can make my own informed choices.
That is not a review; it’s an extended blurb. And it’s full of flat-out untruths. “Gradual revelation” of Sebastian’s true character? He went from villain to saint so fast I got whiplash! It’s the reason this book was one of the biggest disappointments of my romance reading career. Did the AI not sample any negative reviews? Maybe other genres have something to worry about, but until AI can summarise half the plot, throw in some references to the previous books in the series and perhaps sneak in a complaint about some classic romance novel failing (e.g. historical inaccuracy, too much or not enough hockey/football detail, an unnecessary Big Mis or a stupid epilogue, irritating plot moppets, the demonisation of ex-wife, etc.) then it can’t replicate a real romance reviewer.