Quality non-fiction books are the antithesis of AI slop
resobscura.substack.com
resobscura.substack.com
I think good fiction is actually the antithesis of AI, fundamentally. LLMs are really only capable of combining different things into output. This can give you some creative results, but it is ultimately kind of limited. I cannot imagine that prompting an AI a thousand times would give you something truly original in the way many high-quality fictional stories are. The magic and absurdity of life is a necessary thing to create such works.
Most good fiction authors are trying to communicate a story that’s forming in their minds. Even when they get stuck, they are usually not stuck because they just don’t know anyone to ask. It would’t be their voice anymore. Some do seek input and those might experiment with llms for ideas. Using llms for research, technical questions, grammar, synonym sentences, word selection, etc makes sense too.
I’d agree with OP for people generating full stories through llms. While I say that, I know a very young person who told a while back (like early ChatGPT 4 time) they use it to write fan fiction for books or characters they like.
Do I reflect on the objectively correct structure of the story? Do I "agree" with everything Dostoevsky is saying? Not really.
I don't know if this offends literature people or not. But this sure as hell is incompatible with the AI bro paradigm in which all thought can be objectively reduced to some shared standard to benchmark on.
My position will be read into as deep irrationality from many opposing positions, but it's fine.
Now, none of this doesn't "preclude* the ability for AI to have the same kinds of network effects. Of course it can - that's what I've been saying for the longest time. But don't confuse, becoming integrated with society, with this ideal of superhuman X.
Now of course, you could argue, "there is no such thing as anything else besides becoming integrated with society. There is no other standard that works." Very good! Darwinian-adjacnet positions are interesting. This is a position - but then don't claim it's equivalent, or god forbid, the exact same, as the "super AGI" ontology. These require fundamentally different assumptions about what justified what, and an attempt to try and pretend otherwise is sophistry.
LLMs are really only capable of combining different things into output
it's a real question. I've seen a lot of people stating that the human mind "is more than this" as if it was obvious
Like a human can reach into the chaos of their own emotions and pull something unique out. There is some layer of the human experience where ideas can ferment and change based on all your memories is weird ways, and LLMs completely lack this, which makes things LLMs create very blunt and obvious. Like there is no nuance to anything they do.
> The magic and absurdity of life is a necessary thing to create such works.
These two points seem slightly conflicting. Why if a human is truly more creative than AI, must it experience "the magic and absurdity of life" if it can be creative without "combining things into output"? Is a creative human then just not splicing and recombining their experiences into their creative output? And therefore, if an AI had more data that mirrors that experience in some way would it be sufficiently creative to create something "truly original"?
My view is that intelligence and creativity is all just purely recombination. AI can do that well just as humans can. The gap is that the data humans have access to is much more analogue, emotional, and grounded in the real world compared to the data that AI is currently trained on.
An important distinction seems to lie in a human having original experiences to draw from, while an AI has only "data that mirrors" experiences.
True for a lot of current AI. With more robotic bodies, loads more access to cameras/audio/satellite/science measurements, an AI can have original experience on a level that can only be described as god-like compared to humans.
Even then, it’s not really clear to me that an android with the exact appearance of a person will have the same experiences. They aren’t biological creatures with millions of years of evolution behind them.
The part that AI can't (yet) truly replicate is 'feelings/emotions'. Experiences such feeling dread, loneliness, happiness. Or being anxious, nervous or having goosebumps. You can describe all these states but they are differently experienced for each individual and also dependent on each unique situation.
Ofcourse this also then comes down to how you exactly define emotions.
Though I think there's at least one more important thing missing, namely the ability to form memories - to incorporate recent context into the model's weights. I don't think it's a coincidence that this is a very complex process in humans as well (involving sleep and dreams as neccessary mechanisms).
If you read a thousand books about living in New York City, is that the same thing as the experiential knowledge that comes from living there in person? I certainly don’t think so.
You cannot claim that descriptive knowledge of something is equivalent to experiential knowledge of it, unless you have a theory of knowledge explaining why this is the case.
I am sure I Fable (or GPT-5.6-Sol) could answer this question very well. It can probably even tell us what it feels like to eat a hamburger for the first time when you spent the first 18 years of your life as a vegetarian.
Asking another person or Fable what it feels like to eat a hamburger will give you a description of what it feels like. It will not give you the actual feeling. You are merely repeating words that someone else has told you. The description of something is not equivalent to the experience that comes from doing it yourself, as an experiencing subject. Reading a war journal about being in battle is not equivalent to the actual experience of being in battle.
To illustrate the point again: if we gathered ten people that all spoke different languages and gave them a hamburger, they would all gain the knowledge of what it feels like to eat a hamburger, even if none of them are able to communicate this to the others in language.
I can't keep repeating myself.
If I've understood you correctly, your objection is that this description is second hand, rather than being a description of the model's own feelings.
Like if I write a description of what it's like being a soldier in a war zone, despite never having been a soldier and never having been in a war zone. Perhaps I can write something plausible if I've read enough and use my imagination.
snakeboy (whose commend started this subthread) seems to claim that having real experiences (not just having learned of others' experiences through reading) is a prerequisite to creating something 'truly original'.
Is that your position also?
If so, is there any possible demonstration that could change your mind?
Like if I write a description of what it's like being a soldier in a war zone, despite never having been a soldier and never having been in a war zone. Perhaps I can write something plausible if I've read enough and use my imagination.
This is not an answer of what it feels like to be a soldier in a war zone. It is your guess. You do not know, you are imagining. It may be "plausible" for a novel, but that is not what we are talking about.
snakeboy (whose commend started this subthread) seems to claim that having real experiences (not just having learned of others' experiences through reading) is a prerequisite to creating something 'truly original'.
It would seem to imply that if you are creating something merely by reading others' work and synthesizing it, and then rearranging it into something else, all without actually using your own novel experience – then yes, it is not truly original. It is like writing a book about living in New York City by reading books about living in New York City. Your work is second-hand by definition.
Nobody did any original research, you didn't discover gravity, and you didn't participate in any historically significant events.
Yet we say that people who have been through education know about such things.
You may know every detail of the American Civil War, from every book ever written on the topic. You may read every journal from every person that wrote one about it.
You do not have firsthand knowledge of what it was like to participate in the war. Any assumption of that knowledge is a (false) generalization of what you think a person “would have thought or felt” based on contemporary ideas of the personality, psychology, etc. Reading a journal entry about a soldier seeing his comrade killed is not equivalent to actually having that experience yourself.
Of course not all works need to be this way, and many, many books are written about war by people that haven't been in it.
The broader point is that LLMs cannot have original experiences.
Stick a camera and a touch sensor on the machine, now it is experiencing the world.
Besides that, there is a lot of value in redigested experience. Quite a lot of history is only really examined in a larger context, where the person describing it cannot have experienced all the original events.
Implication is that virtually none of the novels ever written which include wars are actually original works, since vanishingly few novelists have been in active wars.
Maybe originality just doesn't mean much then?
Does reading an infinite number of cookbooks mean you know what the ingredients in each taste like?
In other words, when a real human needs to translate an experience into words to share it, A LOT of information is lost. The LLM might actually be better-suited for the the task in your example, even if the point you were driving at stands.
It's basically the Turing test.
Definitely agreed, but at least when talking about experience we've been stuck in this position for most of human history. How do I know what it's like to eat a hamburger? Perhaps, like the cute visual in Ratatouille there is a deep inner experience but (also like in Ratatouille) when I need to communicate that experience with another person I'm stuck relying on language, and the language is found lacking.
I know it's just a kid's cartoon, but interestingly Ratatouille was making this exact same point -- he really lacks any meaningful way to convey his inner experience. It's locked behind language. The film tackles this by showing an Remy's complex and interesting visual, but his friend only gets the fuzziest, dullest visual when he attempts to communicate it.
Remy's immediate rat & family peers actually _never_ get to understand his experience, and instead later help him out of love an allegiance.
The difference is huge. An LLM conversation can contain within it a batshit crazy model of the world and still produce a plausible stream of tokens. No skin in the game. If you don't test your ideas against reality, you don't have a model of reality; you have a model of written human thought. It's critical to understand the difference.
I still doubt that we have the knowledge to describe all of our human desires as mere results of our brain interacting with the nervous system either. There are many processes going on besides analysis and synthesis. If I plan on doing something extraordinary, I don't only sit there, think, and try to rearrange some pieces from memory. I'll go look around, touch, taste, re-explore and expand the perception my environment has to offer. Further, my perception can vary by the minute and I could even try to bring in a touch of chance whenever I feel like.
Wittgenstein's "whereof one cannot speak, thereof one must be silent" might be fitting here, yet language still continues to evolve.
I have a gamer friend that believes, having played 100's of 1st-person games, that he has had rich experiences because of it.
I reflect back on a day my family and I set out to hike through Arches National Park (still a Monument at the time, I believe) and the heat was such that my oldest daughter seemed in danger of going into runaway heat exhaustion. I began to panic as we were so far along on the hike already that even returning to where we parked seemed to pose danger for her.
There was little to no "coverage" (shade) along the trail and the heat and sun were intense that day—we had passed almost nothing. Finding the smallest of bushes to provide minimal shade for her, I scouted ahead on the trail to look for some kind of larger area where there would be enough shade, perhaps a breeze even, to allow her to bring her core body temperature back down.
Fortunately I found just that. (And it turned out that, in the news later, two hikers within a few hundred miles of us had in fact died that day as it was a record heat wave for the area.)
I think somehow that me and my daughters experiences that day, when actual death was perhaps on the table, was something a "virtual environment" can never really offer.
I agree, yet I also think AI is extremely limited in terms of new creativity.
My guess is that humans operate on some deeper level. Much like the grokking model example where it easily trained on the training data, but only later did it really start to understand the deeper problem space and 'grok' the problem, I think LLMs are doing something similar with human thought. LLMs can throw more compute, but they are still lacking some introspection/drive that leads to the better creativity of humans. Grokking this from the training data might not be possible, or it might take some fundamentally different approach, or even an imperfection in humans that LLMs don't have (ego/drive?).
This isn't to say that LLMs could never do it, but the cost to do so might require magnitudes more scale in training times and size. Or it could be in the next model released. Or maybe they already can, but that gets trained out in favor of overall more general correctness accross all domains.
No, they aren't. I'm not sure why it's such a popular narrative that all art is just some kind of recombination of existing things; it's not. Actual lived experiences are unique and impact what is created. Novels written by people in post-Revolutionary France are not the same as novels written by post-WW1 France. Events in history shape culture and introduce new ideas and experiences that weren't there before.
The gap is that the data humans have access to is much more analogue, emotional, and grounded in the real world compared to the data that AI is currently trained on.
That's the problem with your argument, right here. Emotions are not merely portable "data," and framing this way is a problem from the start.
I don't think you get human experiences without actually being human. The most perfect LLM copy of a mind is just the copy of externalized output.
I would like to see art that was "created in a vacuum" so to speak. I am not aware of any. A first time poet who had never read any poetry or prose or heard a tale told? A painter who had been blind since birth and just handed a canvas, oils?
(To be pedantic, the person you quoted said art was recombining their experiences which you took as existing things.)
I do think so. The difference is that, even apart from synthesis, the creative human knows how to filter out the shit ideas from the clever, subtle ones—LLMs seem to just run with the first…um, thought that… um, comes out of their head.
I suspect knowing that you've on to something special probably does require "human experience", woe, melancholy, "the human condition", reflecting on the brevity of life…
Yes!
Aside from the more fundamental questions about the nature of cognition and what constitutes intelligence, there's a practical matter: humans have life to learn from. We have a lifetime of persistent memories, an identity, a social life, physical interactions with the world of many qualitatively different kinds, and we're taken on a lifelong journey into different studies, workplaces, and living environments by the biological need to survive.
Even if human creativity is in some sense also recombinative at its core, we have a very different set of inputs from LLMs. Even under the (almost certainly false) assumption that human condition is basically similar in structure to an LLM, we should expect some differences in outputs based on these profound differences in inputs.
you can make LLMs do that to
In terms of how writers think about creativity this doesn’t reflect reality- there’s the old writer’s saying that there are only 7 stories (and other variations of this idea) and that writing is about creatively remixing these well-worn ideas.
The LLM’s capability to write acceptable fiction and nonfiction is coming soon if it’s not already here. (I think right now it still needs some high level input from humans, depending on length and topic)
It’s clear to me that, especially as LLMs get better and better, there’s only one real difference- it’s that I don’t want to hear what an LLM has to say. I’m only interested in what other humans have to say. I feel the same way about AI music- even if it sounds ok, even as good as the kind of unoriginal pop music that’s not AI made (but is a kind of it’s own slop) or a bad committee made hollywood movie, it’s still fundamentally more interesting than something that’s been generated. Even if it’s 100% fiction it’s still based on a real human’s life experiences.
Maybe, but that has basically zero effect on the experience of actually eating food. Just because the form may be the same doesn’t mean an LLM will be able to supply original content.
I would say human made slop writing isn’t any better or worse quality than LLM writing.
Knowing if it was made by a real person makes a difference to me, but I also know it won’t to everyone.
I could write you a story right now about being an expat living in Poland in 2026. It’s unclear to me how an LLM would have the cultural knowledge or experience to write such a story that is actually accurate and up to date, without already having something like my story already in its dataset.
Shakespeare did the same, so I wouldn't blame you for it.
Superhero Wars Trek 42 - the latest franchise installment, billion dollar Hollywood budget, financed and produced by billionaires, audience tested committee-made script, lots of people involved, designed primarily to push merchandise.
Personal Story - a movie about someone's personal reality or fantasy, something typically not made by studios because the story isn't profitable to a wide audience, painstakingly generated and edited by one person using AI, designed primarily to tell a story.
I put up a different strawman because your strawman is too beatable. As humans, of course we're all interested in what other humans have to say. But you haven't explained why you feel people can't use AI to say things, or why AI output can't be based on a real human's life experiences.
Personally, it seems like movie studios are the ones with only 7 stories to tell, and I've seen the movie and its sequel, the prequel, the requel, the remake, the gritty reboot, the musical, and the animated series. I'm excited that AI will let people skip the studio gatekeepers to tell some new stories for a change.
From what I can see, AI is absolutely able to create new work. Things that are not in the dataset, but look like they could be. This is pretty much what generalising means, they can make a wide variety of things that aren't actually in the original set but can hide in that set pretty well.
When you are an original piece of culture, what do you mean by it?
It seems pretty obvious to me that there are cultural trends which are happening instantaneously and aren’t merely some sort of consequence of past historical data.
I don't see why a computer (or a person) couldn't make up a plausible sounding story about a NYC artist of the present day? You would just reference things that don't change that much (artists having a boheme type existence, wild inequality in who can live off their art, NYC being a sort of place where people go to become successful).
The older I get, the more "nothing new under the sun" returns again and again.
- St John of Damascus
I once tried telling it to write a fantasy novel for me because I didn't know what to read next. Gave it some examples of books I have enjoyed in the past so it would have an idea of my taste...God was that awful. Absolutely terrible idea.
AI is not coming for fantasy authors jobs any time soon!
Everything else is just not that interesting to a human.
But I want to say that non-fiction books are a lot more resistant to LLMs than you might think. I'm reading two non-fiction books currently. And both (https://www.goodreads.com/en/book/show/17801.Underground, https://www.goodreads.com/en/book/show/44824581-a-k-pop-live) could not have been written by an LLM. One is based on in-person interviews that were hard to set up, and the is based on a log of sources a number of which are going to live events.
Not every book fits this pattern of course. But even a more straight forward book (https://www.goodreads.com/series/139929-designers-dragons) will often rely on sources that are not easily accessed by an LLM, they may not be digitized, or they may not be accessible in your common digital collections.
I 100% think LLMs can do good synthesis work, don't really think we have the tooling/checking in place for them to write a competent non-fiction book yet, but that will certainly happen, but I still think the best and most interesting non-fiction remains firmly in the realm of humans.
(Also, a lot of non-fiction is memoirs, travelogues, etc. which LLM can't really replace unless you're dictating to it.)
I do think book prizes are a better-than-average signal but I previously volunteered with a book award. I will caution that pretty much every publisher mass-submits these books for consideration in every remotely relevant prize. It is a cost of doing business (similar to how photographers pay to enter photography competitions to try and win the "award winning photographer" title, or businesses submit dossiers with consideration fees on why they're one of Michigan's top 100 places to work).
There are often so many books and so few willing qualified readers that which books get an award can either be completely arbitrary, or comically easy.
For example, the NCR Book Award in your book corpus faced a big scandal when it was revealed that the judges did not read the books themselves. [1] The PROSE award is so comically large that an ordinary category finalist or win is usually overstated in prestige and value.
[1] https://www.theguardian.com/news/2013/may/19/literary-prize-...
The beer and wine I enjoy the best my wife finds completely unpalatable. For wine, I enjoy a dry Grenache; Mas de Gourgonnier is a good one. For beer, I enjoy a beer with body, a malty profile, and balanced hops; there are many mass-produced styles that fit this, but I like Sam Smith Organic Lager, and Leffe Brune. I’ve also brewed many beers myself and one of my all-time favorites was an all-grain formulation of a honey ale [1].
Such is the nature of beer and wine (and HN) that someone is going to “disagree” with me. That’s fine. Just go try some things and see what you like.
I think there's a lot to learn on how we really process information.
^1 In the early days of AI, the concept of Hebbian learning was popular („neurons that fire together, wire together“). Its implementation is even simpler than gradient descent used today, but never caught on https://en.wikipedia.org/wiki/Hebbian_theory
A lot of my friends can't do same trick. But the fascinating part is that it still works for them! Usually they can recall whatever I forgotten after I described the place we were in.
This sounds like it's more down to how the individual uses the tool. I am not someone who has even been particularly good at reading > learning the thing. I've only ever been capable of learning by doing and with capability to interrogate on the points I don't get. LLMs allow that in a way that is just not feasible with any human being whose tolerance of me would diminish rapidly.
In most cases, I will be writing down my understanding as I would in isolation from a primary source, building flash cards and actively practicing what I have learned, only with more capability to interrogate on the points I have difficulty understanding or need clarity on. Effectively, I am doing the following in a capacity I personally never had via any other means:
> I seem to be doing a lot of integrating and reorganizing my thoughts.
If you are using it as a slot machine of knowledge, and going from receiving > doing with no intermediate step I see how outcomes could differ.
I genuinely agree with a lot of your comment but lines like this drive me crazy. Anytime somebody has a critique of AI, so many people hand wave away with “well you’re doing it wrong.” If we can just chalk up every negative thing with AI to “incorrect tool usage” then we’ll never honestly assess it.
Their assessment was:
> When interacting with LLMs it feels a lot more like I'm just receiving knowledge passively and I don't think it gets integrated as well.
My point is that the passivity is not necessarily an inherent property of the tool. It depends heavily on an individual chooses to engage with the output.
Saying LLMs make learning passive is a bit like someone else saying books make learning passive because you can skim them without thinking. When we all know instead of reading a book passively you could in addition annotate it, question its claims, summarise it, test yourself, and ultimately apply what you learned. The medium permits both. The same is true of the output of an LLM. You'll get out what you put in and nothing more.
You can use it like a slot machine, repeatedly pulling the lever for another answer without reflecting on any of them. But the fact that a tool permits shallow engagement does not mean shallow engagement is the only option available to you.
That also doesn’t mean every criticism of AI should be dismissed as “incorrect tool usage.” AI has real limitations and deserves serious criticism, and I assure you, I am not popular with the AI hype bros for this reason. However we still need to distinguish between limitations of the tool and consequences of how someone uses it. Not every potentially negative experience is rooted entirely in the technology itself. Sometimes user behaviour is a significant part of the cause.
It's... almost an oxymoron, and very different from my experience.
Your subjective experience matters in this case. Almost completely.
For example, a child learns that "foreigner" means someone from outside their country. Then, when they're 11, they go on their first holiday abroad and realise "Wait! _I_ am a foreigner here!"
So, maybe one way to frame what you're saying, is that LLM output tends towards being easily assimilable.
The recent "society & culture" books gave me some good book club ideas.
Bug report: filtering by "award" appears to be broken for some awards. If I select Pulitzer or National Book Award, no books show up, but I can find books with these awards by browsing.
Of course, having created a cool thing generates no obligation for further work on your part! I'm grateful that you did this work and made it available for free.
Did you build your semantic search index based on the full text of the books, or just the reviews and descriptions?
Also I just had to upgrade my Vercel plan due to traffic, which is very welcome and appreciated, but if anyone wants to donate, that would be really helpful! There's a Stripe link at the site: https://book-prize-index.vercel.app
Thank you for building this!
The JSON downloads quickly and had all the info I needed.
Oxymoron. S3 and its clones are one of the most expensive ways to host files.
Here’s what I did: 1. Install Jomo on the phone. There are other apps, but basically it makes me wait 5s before I can open a brainrot app, and then I can only use it for 5min, and every time I do this I have to wait an additional 5s. It resets at midnight. This introduces the necessary friction.
2. Paper books. The house is now full of paper books. Book seems interesting? Buy it. Books everywhere. Real books.
3. The actual habit. I stack habits. Right now I have a daily workout routine. First thing I leave the house and either run or go to the gym. I bring my book, and afterwards go to the cafe across the street for a coffee. At night, I read before going to bed. Once you finish a few books this way, you’re in. TV now is actually hard to pay attention to, and I feel kind of… dirty doomscrolling now (like eating junk food, it tastes good but you feel gross during the experience).
If you want something to start with: https://bookshop.org/p/books/true-grit-charles-portis/5ac454...
I tried attaching the book to the phone as their marketing suggests, but that didn't click somehow. Having both out means I would check the phone and then not look at the book.
Working so far, but it's only been a couple of weeks, so let's see.
[0] after reading this HN post: https://news.ycombinator.com/item?id=48662381
Having a nice library of paper books, for whatever reason, has worked a lot better for me. I think it’s passively seeing them around the house and in the background always considering what is next.
This last round I couldn’t decide what to dig in to after I’d finished True Grit, so I spent a few minutes browsing what we had on hand and it was just… a very pleasant experience.
Also, unless you pirate ebooks or have access to a library that offers them, used paperbacks are actually much cheaper than buying ebooks.
I used to have paper books. But I found they were just a massive pain in the arse to carry around, and move. They became very expensive, very inconvenient wallpaper. I gave them all away to a charity shop and bought an e-reader instead. Now I carry around 250 books in my pocket wherever I go. It's liberating.
And any other ideas that HN readers have for awards to add would be welcome. Currently it's probably too history-slanted since I'm a historian and knew those awards better.
I can point out that the UN agencies' respective reading lists for their staffs' professional development are 99% internal UN papers of very limited interest to external audiences but I haven't had a professional reason to look at any EU agencies to confirm or deny value.
Additionally, the Financial Times has a second set of book awards along their main awards which you've not listed, the FT reader's best books list.
Historians who are not trying to sell mass market books have a much broader set of books. You won't see them in a normal bookstore.
The toggle between "traditional" text and "semantic" search is cool and new, but maybe the toggle could stay visible (it's there on the home tab, but then not anymore on the books tab).
- "The Spanish Civil War" by Paul Preston
- "Homage to Catalonia" by George Orwell
Books on the same topic on my reading list (and quite frankly can't wait to get hold of them):
- "The Spanish Civil War" by Hugh Thomas
- "The Fight for Spain" by Antony Beevor
Regarding the other two authors in my previous comment: Hugh Thomas was also a leading Hispanist historian [2], and Antony Beevor is a prominent military historian [3]. I expect both books to be in the same vein as Preston's.
(Quick aside for the astute reader that noticed all of the recommended authors being non-Spanish: I find Spanish politics extremely polarized; the Spanish Civil War being perhaps the most polarizing topic. As such, I find outsider's perspectives to be more balanced and objective.)
References:
1. https://en.wikipedia.org/wiki/Paul_Preston
2. https://en.wikipedia.org/wiki/Hugh_Thomas,_Baron_Thomas_of_S...
Makes sense to me, even as someone who live in Spain. It's a hard subject to talk to people about, especially because it never really was addressed as a country, everyone (politicians) just agreed to pretend it never happened, sweep a bunch of stuff under the rug and hope for the best. Then everyone act surprised when things start to bubble again...
I think the approach makes sense, also why you probably don't want to read about US politics from people actively inside of US politics, and it's similarly polarized today. Seemingly, at some point, people just lose all head and reason, and just start backing into their side, regardless of what that means.
So I'd read The Collapse of the Spanish Republic, by Stanley Payne. He was a journalist in Spain through the entire period, so most of what he does is provide a first hand account. He never claims to try to be impartial, but he's pretty nouanced, as he knew a lot of the players personally.
We all know the "AI-tics" that give away a sloppily AI-written piece, but even if you steer them, they still struggle to write consistently high-quality prose.
Somehow I feel that the work of a good copywriter has never been more noticeable.
Diversity of writing styles was part of that, but I’d point to vernacular exposure as the larger component. We’re going to go through a period where we try to adapt to a form of “Universal English” for those of us who read primarily English writing.
Other languages’ readers may be experiencing the same dissonance when they come across AI-generated prose in their native language (but I’ll let others validate /reject my hypothesis).
But when I submit a novel with em-dashes, will sloppy agents and sloppy editors be able to tell that em-dash was deliberately put there by me?
I think there's also a lot of recency bias in it. The same thing that makes you suddenly notice how many people are driving the same model car you just looked at, or how many ads there are for Turbo Encabulators after you read an article about them. A lot of the "tells" people picked out in early AI are tells because they're also really common in the material that the AI was trained on and the styles it was made to emulate. But until everyone wanted "one quick trick" to pick out AI writings, people didn't have any particular reason to need to notice those tells and so they slipped under the radar.
All these HN posts with "stop with the LLM generated garbage" will slowly fade away because both people and LLMs will talk in a similar format and will be inured to the the distinguishing feature.
Certainly "humans" will keep carving out distinguishing characteristics, but just like "corporate speak" is a thing, AIsm will be a thing.
That's how language spreads.
I have education/experience in both literature and coding, I have a pragmatic starting point when approaching text while also being able to recognize stylistic oddities, so I get to be the guy editing out AIsms sometimes.
But I've helped other people copywrite where their environment was all org-speak and academic writing, and AIsms don't really stand out in that case. AI is effectively "doing the right thing" writing the way it does for those tasks. Even tho the right thing is often a bad thing.
(I may be projecting real life experience onto the LLM.)
I am genuinely fascinated as to how Claude acquired its utterly aggravating way of writing. It’s so much more irritating than ChatGPT, which is already not good.
Though this might be me as a British reader, simply preferring a rather less American turn of phrase.
I reckon the more transatlantic, english-as-international language DeepMind team have had a subliminal (or maybe deliberate) impact on the way it chooses to write.
Or perhaps small open weights models simply aren't under the same commercial pressure to be engaging and sycophantic and are therefore less likely to adopt the samey overly casual, upbeat, Californian sales assistant manner. (Don't get me wrong, I like this from real human Californians just fine!)
Either way, the default tone is much less showy. I would be interested to find out if you agree.
I am very much an LLM cynic. I am engaging because I must, and trying to learn fundamentals, but I would not say I am overly excited by any of this, just glad that small open weights models exist as a counterpoint.
I loathe the way ChatGPT writes, and the Claude-isms that are everywhere; it is actually quite enraging, especially when you start seeing it in internet comments from people who used to try to write out their own thoughts.
But in my experiments with open weights models I have found I am much less aggravated by summaries and outlines written by Gemma 4, so much that I am happy enough to read them, because they have fewer irritants that take me out of the reading flow.
Though this evening it told me very kindly that my photography is a bit "safe". How very dare it… understand me that well.
This seems to be a hard problem for LLMs, as passing would probably require good self-perception ("oh no, I am writing like an AI!") and fine-grained control over its own output ("let's write like a human instead!").
There's a lot of work in the humanities about different aspects of good writing, but that's not quite the same thing. And anyway they tend to assume a pre-existing level of writing ability. Students are supposed to learn good writing through practice; there are rules and exercises but they're incomplete.
It’s much easier to understand this once you think about other generative forms. MidJourney never just sits down and draws for fun, so fun never informs its art (only the outward appearance of others’ fun, separate from the fun itself). Suno doesn’t waste hours trying to find riffs on a guitar, so its output is never informed by the direct joy of getting it right. Its music is never optimised for playability on a particular guitar with a scratchy seventh fret and a too-high action. Neither Midjourney nor Suno have evolved their styles due to short-sightedness or carpal tunnel.
If you had a human writer who over a long career only ever wrote articles from an outline given to them by someone else, and you had all the outlines and all the resulting articles from those outlines, and you could train an LLM to generate an article from an outline, it still would not be kicking itself frustrated by an inelegant phrase in a prior article, it would not avoid certain phrases out of a passive aggressive reaction to some editor’s note, it would not ever just rush an article because everyone is gathering at the pub, and it would not choose an analogy just to rub the author of a bitchy critical letter to the editor the wrong way. An LLM could not “subtweet”. It could not write a series of articles hoping one important person will spot that they are auditioning for a job.
Creators have unseen, undocumented influences and motivations that inform their work over a long period. I don’t mean to say that these individual influences can be reliably detected in individual pieces of work. I do mean to say that I think their broad absence tends to be felt in LLM writing. As readers we develop an affinity for writers as much as for their writing, and we do this in part because we deduce things about them.
However, general purpose LLMs like Fable have been trained on huge amounts of all kinds of data, and therefore find it exceedingly hard to break out of the grooves carved by that data. They can’t avoid defaulting to centroids and averages, even when they are trying not to. This makes it possible for classifiers like Pangram to discriminate their writing.
A plausible way to work around this limitation would be to train a LLM on a limited and cohesive subset of writing materials, so it would absorb their specific writing style.
One example might be Talkie, a LLM trained on pre-1930’s English text. Talkie is a far smaller and less powerful model than Fable.
And yet, Talkie’s writing is so distinctive that it is often classified as human by Pangram.
Of course, the amount of self published work has also helped make the lack of a good copy editor noticeable. I can excuse self published blogs though. But the stuff released "professionally" has really become farcical.
That said, I had high hopes for the catalog that the author had put together, but for my technical fields, I found it lacking. Trying to find books that would be close to van Soest's Nutritional Ecology of the Ruminant or van der Werf's Animal Breeding: Use of New Technologies, yields a whole bunch of pop-sci titles. The joy of finding these titles at the library in the past was how remarkably understandable and engaging they were while remaining scientifically rigorous. I immediately feel that the pop-sci titles that I was recommended are going to try to sell me something or some idea, rather than letting me explore ideas myself. I guess that means I need to go back to browsing library shelves myself.
> There is really nothing “AI” about this aside from the tool that collected the data and coded it, and, crucially, semantic search […]
So really, everything about it is AI. And that’s not a bad thing! It’s okay to simultaneously preach the superiority of award winning books over AI-generated garbage while also acknowledging the same AI as a valuable tool for other uses.
I have a stupid AI app I use now for some bookkeeping which was previously, just an excel sheet. except it has a UI, it validates inputs, and does a bunch of other housekeeping stuff that makes it nice. Did the excel work? yes totally.
Libraries are certainly declining in their traditional form. I find it odd that everyone has a digital resources in their pocket yet libraries are squeezing out physical books to make way for more and more computers. Try to find paper copy of the Sony founder's book.... that will be £50 on amazon. Prohibitive. No library within a 10 mile radius has a copy.
I remember joyfully discovering Tony Royce's book in my library a few years ago. That enlightenment will never happen now. Primary knowledge is being lost, churned crude will forever lubricate the delusions of those who have no facility to collate the basis of our understanding.
This is incredible!
back in the reasonably early days of the web (I would guess ~2000) I stumbled across a webpage that said "I used to collect random interesting snippets in a shoebox, here is the internet version of it". I was delighted enough that I actually wrote to the author to compliment them on helping keep the web interesting, and though I haven't thought about it in ages it clearly made enough of an impression on me that I remembered the author's name. and it's still up! https://www-users.york.ac.uk/~ss44/cyc/index.htm
You can get quite specific. You can find texts you wouldn't discover unless you spent years studying the topic. Often these are completely approachable and give interesting perspectives they just get buried behind a wall of syllabi and listicles.
This is how I ended up reading Thompson's 'The Making of the English Working Class' and Graves' 'Goodbye to all That' among others.
Also, this is about the lost joy of "browsing".
I wonder how much serendipity played when wandering through a library (or other things in life—even window shopping). But now, everything on the internet, you more or less have to know what it is you're looking for—are rarely "surprised or delighted".
I’ve found plenty of interesting things by randomly wandering around on Gutenberg.org, completely undirected.
School textbooks are the kind of books that have the very precise purpose of giving you a foundation and roadmap to explore and understand a whole field of knowledge. And they're an industry that has been improving them on successive iterations to produce highly organized, comprehensive and accessible reading material through years.
IE: Was the US Civil War fought over slavery or states' rights? Someone's answer may come from the biases of the textbooks they grew up with.
For me the keyword in your comment is "also". What this word means to me is that, whatever faults textbooks might have, there is no alternative solution. They replicate a problem that all books have.
However, within the domain of textbooks, you still might find good solutions for these bias. These bias are mostly within history and social sciences and good and reputable Universities are renowned for publishing books in these domains that are quite balanced, comprehensive, deep and rigorous. Oxford and Cambridge are the 2 best examples that come to my mind. The U.S. has good Universities too, but after their government declared a war on science and knowledge I'd be carefull before trusting these universities.
I'm always cautious when reading nonfiction because it's hard for me to tell if the author knows what they're talking about when I'm not an expert myself. Using awards is a good metric. Maybe.
I am curious what the "score" for each book means. Is it calculated by giving each award a certain weight and adding them up?
I understand your passion is non-fiction, but would be wonderful to have the equivalent for fiction, of which there is a great deal of excellent literature. Perhaps the same site could incorporate both fiction and non-fiction, with a filter?
I just recently read a really mediocre book with a 4.7/5 rating on Goodreads. Shows being sloppy is not exclusively a privilege for neither the AIs or producers.
Now, can this interface with Libby? The worldcat interface is, well, confusing. I can't figure out if it knows libraries near me exist.
To work on something truly original in non-fiction in the future against AI pushing everything to the mean of its training set and fighting against that when doing research, as opposed to generating a whole book of slop from familiar ideas, is going to be quite the unrecognized effort of writing those books.
[1]: https://store.steampowered.com/app/570940/DARK_SOULS_REMASTE...
Why does it feel wrong (or meaningless) to you?
It's like calling a CRT monitor a really big flatscreen.
Since AI output is based on the most likely next word/token given the context, let's consider how humans do this, i.e. choose the next word based on context.
Is one of the parameters the degree of "randomization"?
Are these other parameters?
- Specific experience?
- Recent exposure to relevant materials, i.e. paraphrasing?
- Borrowed knowledge from a specific resource?
- Writing style, i.e. choice of words or sentence structure?
I'd like to hear what others think.
- almost every book try to stretch core idea into book size format
- unique ideas are rare people attack them under different angels
- a later phenomena : their believes almost predict entire book, outcomes etc, brainwash impact is real
so not sure, how ai slop is better vs book slop, at least with ai you can distill the idea, with the book, you have to spend 10-40 hours to digest average, absolutely non fresh ideas, that author brought in just to sell that book, otherwise it would be magazine article worth.
Why don't you just use whatever LLM you have on hand to summarize every book you could read instead of reading it? The idea that ai slop is 'better' because you can distill the idea is an insane point to me because it treats books as a pure consumption-based concept, where the goal is strictly to finish a book and move on.
I think even the worst book is more valuable than whatever garbage an LLM writes because you can at least learn from the intent. Where and why the author failed, where they got stuck, the points they failed to make. LLM writing is a void, there's nothing to be learned.
You have access to every single wicked problem in the world and you choose to do nothing.
Shame
I've been really exercising my non-fiction reading muscles of late. I'm trying weaponise it as an antidote to AI slop and doom-scrolling. My last read was Designing Data-Intensive Applications by Martin Klepmann (the newly released second edition), it was a wonderful read!
> And before you wonder, yes this is actually free. I am paying for the hosting and the API costs entirely because I just want people to find and read more good non-fiction books.
Just to note, you do have affiliate links so it isn't 100% out of the goodness of your heart ;) Not blaming you or anything, just saying...
It does not do the most important function well, giving me a list of award winning books that is easy to sort through. It is not clear what is a link and what is not.
It looks fine though somewhat generic. But it is not actually functional in a real sense.
Less relatable AI, maybe, but not sloppy AI.
AI trained on other AI (i.e. distilling), can be better than the AI they were trained on, likely because of this reason. Each iteration can raise the bar.
As apposed to the deep spiral of generating shiny attention crystals for humans, and then training on that.
Obviously, you can handicap any model with an indiscriminate dataset. Are you training on slop? Why?
More data is not automatically better data.
Just three (of many) advantages of quality data selection: (1) It gives a clearer signal. (2) It, ironically, reduces the complexity of what needs to be learned, since slop adds its own complexity. (3) It reduces dataset size which means more learning per watt (and per just about any other cost). The advantages compound.
Humans are no different. Children surprise with their ability to absorb sophisticated relationships and skills, while people who had bad examples struggle to recover.
99% of human effective intelligence is higher quality cultural knowledge learned as children. None of us had to spend centuries deciding whether zero, negative numbers, the square root of -1 are numbers. Slop + quality is not quality.
> Uses AI to produce website
Perhaps I am just old-fashioned, and vibe-coding is something that can be done with full focus and care for high quality, but I remain unconvinced. This is a project that needs to be cared about sincerely to be trustable/meaningful/useful, and the approach taken casts doubt on that.