Comments on: Large Language Muddle

Large Language Muddle

sickos haha yes dot jpg — Tue, 16 Sep 2025 17:59:06 -0800

Call the genre "A.I.-and-I": Writers are obsessed with writing about what A.I. will do to their profession. Strange that they all sound the same.

"How can intellectuals chronicle such a shift — one that threatens their ability to chronicle anything at all? O'Rourke is among a growing group of literary writers who have tried to answer the question in the first person, scoping out artificial intelligence's encroachments from within the domains it most imperils. Their writing asks what AI-generated writing can or (much less often) can't do, and how human writers can or (much more often) can't respond. Call the genre the AI-and-I essay. Between April and July, the New Yorker published more than a dozen such pieces: essays about generative AI and the dangers it poses to literacy, education, and human cognition. Each had a searching, plaintive web headline. 'Will the Humanities Survive Artificial Intelligence?' asked Princeton historian D. Graham Burnett. 'What's Happening to Reading?' mused the magazine's prolific pop-psych writer Joshua Rothman, a couple months after also wondering, with rather more dismay, 'Why Even Try If You Have AI?' 'AI Is Homogenizing Our Thoughts,' declared Kyle Chayka, with the irony of a columnist whose job is to write more or less the same thing every week using his own human mind. An article by Hua Hsu put it most starkly, seeing the artificial-versus-human intelligence war as all but lost: 'What Happens After AI Destroys College Writing?' ... The final notes of the AI-and-I essay are cowed resignation, awed acquiescence, and what Trotsky called the 'terrifying helplessness' of cultural production at 'the beginning of a great epoch"'— all from the very writers best placed to condemn AI's creep into literary life. 'What if we take seriously the idea that AI assistance can accelerate learning — that students today are arriving at their destinations faster?' Hsu asks. But what if we don't?"

By: mittens

mittens — Tue, 16 Sep 2025 18:14:58 -0800

I haven't finished this one yet but I've been reading bits of it on breaks all day. My favorite bit thus far: 'While spoken in the voice of an individual author, each piece in this emergent corpus stages a more collective drama. To read these writers writing about AI writing is to witness, almost in real time, intellectual laborers assimilating a threat to their own existence. The threat looms more distantly for some than for others. But whether or not one enjoys the near-extinct security and legacy prestige of a New Yorker staff job (to spend one's days "focussing" on the rapid erosion of the life of the mind — the dream!), this work paints a persuasive picture of a world hollowed by machines — a world, the writers suggest, we will all have to learn to live in.' It's kind of exciting to frame the thinkpieces in this jaundiced way.

By: clavdivs

clavdivs — Tue, 16 Sep 2025 18:53:22 -0800

"that almost sentient mechanism" (1901)

By: smelendez

smelendez — Tue, 16 Sep 2025 19:13:56 -0800

according to a recent Financial Times op-ed by Salesforce CEO Marc Benioff, "the real magic lies in partnership: people and AI working together, achieving more than either could alone." I see this kind of framing more and more, as if LLMs were actually living beings and coworkers, not work tools. ChatGPT can achieve nothing on its own, any more than Excel or the Salesforce CRM can.

By: flabdablet

flabdablet — Tue, 16 Sep 2025 19:22:45 -0800

AI against I

By: jonnay

jonnay — Tue, 16 Sep 2025 19:25:38 -0800

ss if LLMs were actually living beings and coworkers Imagine thinking this, and being 100% okay with these entities being intellectual property that is owned and controlled.

By: storybored

storybored — Tue, 16 Sep 2025 20:15:08 -0800

Essay starts out well, with a meta examination of writings about AI, but then swerves into a typical anti-AI screed which is pretty familiar at this point.

By: Deminime

Deminime — Tue, 16 Sep 2025 20:28:05 -0800

flabdablet, I think it's pronounced I against I

By: flabdablet

flabdablet — Tue, 16 Sep 2025 20:47:29 -0800

a typical anti-AI screed which is pretty familiar at this point As a practising Luddite I am of course familiar with having all arguments in support of the Luddite position casually dismissed as "screeds" so as to avoid any temptation to engage with the actual points raised. I'm not angry about that, just disappointed.

By: mayoarchitect

mayoarchitect — Tue, 16 Sep 2025 22:37:17 -0800

I'm not angry about that, just disappointed. I'm mostly amazed that I somehow still never saw any use for a digital brownnoser, and somehow I still manage to do my job perfectly well without one. Screed indeed

By: chavenet

chavenet — Tue, 16 Sep 2025 23:33:49 -0800

I'm not in the business... I *am* the business. --Blade Runner

By: TheophileEscargot

TheophileEscargot — Wed, 17 Sep 2025 01:03:20 -0800

Essay starts out well, with a meta examination of writings about AI, but then swerves into a typical anti-AI screed which is pretty familiar at this point. I wondered if this was intentional irony. Spend the first half of the essay complaining about the clichés of the "AI and I" essay, then the second half going through those clichés.

By: sohalt

sohalt — Wed, 17 Sep 2025 01:46:48 -0800

We also have expert consensus when it comes to climate change. I tend to take this as a clue that the experts might be on to something.

By: DreamerFi

DreamerFi — Wed, 17 Sep 2025 02:39:15 -0800

I see this kind of framing more and more, as if LLMs were actually living beings and coworkers, not work tools. It helps to realize these CEO's see their workers as tools too, not as human beings.

By: mittens

mittens — Wed, 17 Sep 2025 03:09:43 -0800

Our final defenses are more diffuse, working at a level of norms and attitudes. Stigmatization is a powerful force, and disgust and shame are among our greatest tools. Put plainly, you should feel bad for using AI. [...] there's still time to disenchant AI, provincialize it, make it uncompelling and uncool. I don't believe this, at all. Tastemakers and influencers are not going to lead your kids or your boss to an AI-free future. Nobody cares if it's cool, any more than they care whether Excel or Gmail are cool. All anyone cares about is that the tools are ubiquitous, free, and get the job done. (The fact that it's really only one out of those three turns out not to matter much!) Anecdotally, someone told me about an informal class survey--just an average state college class--where the kids were asked who used AI, and every single one raised their hands. That's amazing market penetration of what is essentially a very boring product. So I think stigmatization will actually be the opposite here: Who wants to be the one to say don't use the miracle homework answer machine? Who wants to be the nerd who pushes up his glasses and says well actually? It's all well and good for us here on this site to come to an understanding that the way we're using this technology is dangerous and dumb. But are you really going to make someone feel 'disgust and shame' for looking up how to do a formula in Excel?

By: Smedly, Butlerian jihadi

Smedly, Butlerian jihadi — Wed, 17 Sep 2025 04:30:51 -0800

a typical anti-AI screed which is pretty familiar at this point I think you're misreading this article. The question is a practical one. In a world where there is real value in human cognition and the ability to formulate meaning via language, and also young people have been provided a ubiquitous, easy to use mimicry machine, how do we preserve the ability to read and write in the future? Unless of course you dispute the premise, and would insist that this archaic use of language is best left to the dusty past. But then, why are you posting here, on a text-only web forum?

By: rikschell

rikschell — Wed, 17 Sep 2025 04:43:36 -0800

Who wants to be the nerd who pushes up his glasses and says well actually? ... But are you really going to make someone feel 'disgust and shame' for looking up how to do a formula in Excel? When I was growing up, being a nerd was bad and you would get ridiculed and punched. For a while it seemed like nerd culture was ascendent and celebrated, but the jocks want their hegemony back (an underexplored side of MAGA, I think, is that it's a backlash to how cozy nerd culture, queer culture, and neurodiverse culture have gotten with each other, especially among Gen Z and younger). But no one is shaming anyone for looking up how Excel works! Even before AI there were a million videos and forum posts to go to: people helping people. In a way that was barely even monetizable! This is the equivalent era to Amazon selling everything at a loss. Once AI has killed the peer-to-peer help approach, getting answers from it will get a lot more expensive. Learning from people is good! "Learning" from AI is bad! Not that hard to figure out.

By: joannemerriam

joannemerriam — Wed, 17 Sep 2025 05:54:21 -0800

DreamerFi: "It helps to realize these CEO's see their workers as tools too, not as human beings." Ironically, this is one of the things I am writing about in my current WIP! I'm still working out the plot so it'll be awhile before it sees the light of day but I want to explore the importance of seeing people as people. Anyway, I am not as worried about AI making it impossible for me to make a living as a writer as, perhaps, I should be, for two reasons: 1. I think people will always want human-written novels. Artisanal novels, if you will. But mainly 2. it's already impossible for me to make a living as a writer. I'm only managing a semblance of that right now because I am living off of my savings (and hoping I have explosively amazing sales with my first book) and actually one of my tasks this week is to start applying for work because my money is drying up. I've been in the industry long enough to know that a decent middle-class living as a writer is basically like winning the lottery. That's not to say I don't see AI as a threat—I do—but I see it as a threat to society, as a tool that makes us stupider and more complacent, rather than a threat to me personally. I hope I'm not wrong.

By: TheophileEscargot

TheophileEscargot — Wed, 17 Sep 2025 06:29:56 -0800

I'm reminded of the horror some people in the painting world had at the introduction of photography.

As the photographic industry was the refuge of all failed painters, too ill-equipped or too lazy to complete their studies, this universal infatuation bore not only the character of blindness and imbecility, but also the color of vengeance. That such a brainless conspiracy, in which one finds, as in all the others, the wicked and the dupes, can achieve absolute success, I do not believe it, or at least I do not want to believe it; but I am convinced that the ill-applied advancements of photography have greatly contributed, like all purely material progress, to the impoverishment of French artistic genius, which is already so rare. Modern Fatuity may well roar, belch out all the rumblings of its rotund stomach, spew out all the indigestible sophisms with which a recent philosophy has stuffed it. Nevertheless, it is obvious that this industry, by invading the territories of art, has become art's most mortal enemy, and that the confusion of functions prevents any from being well fulfilled. Poetry and progress are two ambitious people who hate each other instinctively, and when they meet on the same path, one of them must serve the other. If photography is allowed to supplement art in some of its functions, it will soon have supplanted or corrupted it altogether, thanks to the natural alliance it will find in the stupidity of the multitude.

Photography didn't supplant or corrupt painting altogether. But it did seriously affect the profession of painting. Miniature portraits in particular went from a successful specialist trade to near extinction. Painters had to figure out what they could do that photographs could not.

By: mrgoat

mrgoat — Wed, 17 Sep 2025 07:01:18 -0800

In a world where there is real value in human cognition and the ability to formulate meaning via language Here's my take, for the moment. (My opinions on LLMs are still evolving) I want to live in a world where this is true, and I personally believe we live in a world where it is still true in many places, but these techbro prompt fondlers are trying very hard to erase that, and "value" is subjective. In terms of monetary value specifically, I see a lot of corporate work where there really isn't value in human cognition, where people really do want a lot of corpo-speak flavored bullshit slung around for the purposes of looking like there's useful work being done when there... isn't. LLMs are awesome for that! We can now effectively pair the "nobody ever actually reads this" kind of work with an equally valueless machine that doesn't actually write anything. It's efficient! This isn't the fault of AI - humans have already managed to devalue critical thought and generate whole industries of paid, but useless make-work - but AI sure is helping. But people get paid for it, so in the "pay my rent" sense, it has plenty of value that human cognition doesn't. For all the places where human cognition still counts, I do also believe that using the giant bullshit machine is eating the seed corn. Especially when our current gen stuff has no concept of, or ability to evaluate correctness. It's bad for helping people learn. It can't write a passable book, not least of which because it can't fundamentally do anything more than remix the training corpus. It can't really do anything new. But man, could I ever bulk up those TPS reports that Lumbergh wants. I haven't yet, but every time I'm asked to write something up that not only won't be read, it will be not read by someone who doesn't have the domain knowledge to understand it even if they did read it, it's tempting. So it's probably here to stay.

By: mittens

mittens — Wed, 17 Sep 2025 07:11:18 -0800

Of course if Nvidia falls, maybe all this becomes moot? "On Wednesday morning, the Financial Times reported that China's Cyberspace Administration had told tech companies to stop using Nvidia chips which had specifically been manufactured for the Chinese market."

By: the sobsister

the sobsister — Wed, 17 Sep 2025 08:59:57 -0800

I Against I?

By: Artful Codger

Artful Codger — Wed, 17 Sep 2025 09:08:49 -0800

Earlier this year, the Trump administration told Nvidia that it would require a license to sell its China-designed H20 processors in the country. [CNBC; underline mine] Note also the success of the Chinese-designed DeepSeek, which was developed for a small fraction of the investment poured into the US efforts. It's free version does some useful stuff that is only available with the paid versions of ChatGPT. Another facet of China's AI efforts. China's moving forward with or without US AI tech. Also, the job threat we most fear is AGI, right? - a digital drop-in replacement for the average knowledge-worker. I don't think that the fantasy of AGI will be achieved with bigger LLMs eating bigger dumpster-fulls of unvalidated data. AGI may prove elusive; it could be like fusion reactors - perpetually 10 years away. It's my guess that while AGI stumbles, more basic, limited AI will quietly be adopted as force-multipliers and natural language front-ends for expert data in specific applications like medicine, law, etc. In creative endeavours, slop has been with us since well before AI, and I expect (hope) that people will continue to treat it with disdain.

By: mittens

mittens — Wed, 17 Sep 2025 09:24:51 -0800

Also, the job threat we most fear is AGI, right? That's such an interesting question. I don't fear it because I don't think AGI is going to happen, but I am extremely fearful of the job threat posed by current-state LLMs? Not the capacity of those LLMs to replace workers, but the capacity to replace workers regardless of whether the technology works. I suppose my mid-range nightmare is: Job losses due to current LLM uptake, with some small hiring around the margins to fix the problems of LLMs; job losses due to the inevitable market crash when LLMs do not prove very good at replacing workers; and job losses due to a deflationary spiral when the current administration proves unable to help the economy recover after the crash.

By: Artful Codger

Artful Codger — Wed, 17 Sep 2025 09:51:42 -0800

Companies Are Pouring Billions Into A.I. It Has Yet to Pay Off. [NYT]. There are other reports of underwhelming gains. The use cases, or killer apps, that fully justify the intense investment are yet to emerge. We seem to agree that AGI isn't imminent. Your mid-range nightmare is possible... but there are so many other external factors right now that are also threatening to wreck the economy/civilization/planet that AI will have to take a number and sit over there.

By: sickos haha yes dot jpg

sickos haha yes dot jpg — Wed, 17 Sep 2025 10:00:07 -0800

We also have expert consensus when it comes to climate change. I tend to take this as a clue that the experts might be on to something. There are at least two groups of experts with consensuses on this one, and those consensuses are meaningfully different, so it is not actually the same. Thinking AI's a problem isn't like being a climate change denier, but I guess if that's how you've been framing it, it certainly explains a lot.

By: mrgoat

mrgoat — Wed, 17 Sep 2025 10:00:40 -0800

the capacity to replace workers regardless of whether the technology works. I also see this as a huge potential issue. Not specific to LLMs or AI, but of a culture that's happy to chase profits all the way to the bottom. New technology has always displaced workers in various industries, but what we're looking at is replacing workers with something that looks like it works, but doesn't. Horse-drawn carriages went out with the rise of cars, but at least cars actually carry and move things. I'm also not worried about AGI right now, not least of which because we haven't even got to the point where we have a solid working theory of what the "I" part means.

By: sickos haha yes dot jpg

sickos haha yes dot jpg — Wed, 17 Sep 2025 10:01:16 -0800

Not the capacity of those LLMs to replace workers, but the capacity to replace workers regardless of whether the technology works. THIS IS IT, THIS IS THE ONE

By: Zumbador

Zumbador — Wed, 17 Sep 2025 10:07:24 -0800

Ha ha I was about to quote post that exact same phrase " Not the capacity of those LLMs to replace workers, but the capacity to replace workers regardless of whether the technology works." Yes! This is exactly the problem.

By: TheWhiteSkull

TheWhiteSkull — Wed, 17 Sep 2025 10:08:05 -0800

I feel like the question "will the humanities survive AI" is mainly being posed by people operating from the predetermined conclusion of "they won't, which is why we should just go ahead and destroy them now."

By: TheWhiteSkull

TheWhiteSkull — Wed, 17 Sep 2025 10:18:28 -0800

"Them" being "the humanities" in this instance.

By: Ryvar

Ryvar — Wed, 17 Sep 2025 11:50:25 -0800

Also, the job threat we most fear is AGI, right? Whose definition of AGI? If you take the classical definition of machine sapience (eg Data from Star Trek TNG), then, no: median estimate by actual researchers in the field had been hovering around three decades from now, although that began moving up around 12-18 months ago. My gut feeling is that both parameterizing new predictive "world models" from the semantic map of LLM-like structures at runtime AND detecting when it is a good time to spin up said predictive models are going to be riddled with a host of recursion issues. I don't think we appreciate the lengths the human mind goes to in order to minimize radical alterations to knowledge representations in the brain (read: changing your mind too much, too rapidly, threatens overall network integrity). These issues may well require partial reinventing-the-wheel of ANN fundamentals to address the overarching structural complications. So I'm betting on least twice that: over six decades until anybody is shaking Data's hand (this is consistent with my estimate of ~75 years back in 2019) . Like, we just discovered two fundamentally new types of human neurons last year: triangular and multi-synaptic ...when I'd previously mentioned major arteries/branches/trunks/highways in the context of neurotopology I always meant those strictly as emergent/derivative topological features, but it turns out there are actually literal, physical entities. That has to throw several wrenches into any nasceant plans for RNN-based continuous training models, right? If you mean OpenAI's definition of AGI, which is the one I suspect most US CEOs are interested in: "you can fire half your paper-pushers," then yes we are probably on course for that to be feasible in 2027-2028. What will probably happen is that many CEOs will fire 80 or 90% of their knowledge workers before discovering that while a lot of what was being done was make-work, there was more genuine need than they understood for a sapient intelligence with an abstract mental model of how their work integrated with the various systems of their broader field/industry. This won't reduce the resulting short, mid, or long-term damage from the initial cuts - damage to both the workers and the broader economy. As for writers/writing: I think LLMs will finish killing off the Listicle/Buzzfeed bilge that was already the proverbial "slop" (just human-powered). As someone who spent a few years engaged to a freelance writer in the early 2010s this was always a disturbing amount of the daily grind / money earned for people in that job market. I don't think humanity would be impoverished by nuking it altogether, and automating it is value-neutral. I think that investigatory journalism was already nearly dead in the US and in serious jeopardy elsewhere, and this doesn't change it other than increasing the already fierce competition for the very few seats. In terms of the impact on students and very young adults: much like Internet search and Internet discourse platform enshittification I think this is just nailing shut a coffin largely built from the Covid era and slow-boiling immiseration of parent-laborers via wage theft over the last several years. Ask any teacher about their students for the last three years if you want an absolutely harrowing trauma-dump on what the reduction in available parental attention/energy has produced. And yet: you still have to reason to use these systems effectively. You still have to read/write to use them at all. Damage? Certainly. But hardly the end. Finally: yes the US investment bubble on AI will implode soon. Sooner and NVidia will be fine due to China's ongoing teething issues with domestic inference hardware, and the broader economic fallout will be somewhat limited. Later and the fallout will be far worse. After it implodes, China's deep-learning-based companies will sail on unbothered: DeepSeek is already (just barely) profitable as a service because of its amazing efficiency (despite both model and source for key innovations being freely available to download under an MIT do-what-you-want-with-this-except-suing-us license). Any jobs legitimately lost (the 50%, not the 80%) are going to stay lost, but prices for using LLMs are never going to suddenly inflate when open source models consistently track just nine months behind state of the art and you can run them at home. A corollary to "today is the worst the leading AI models will ever be at this task" is: "today is the worst the user-friendliness of local LLMs will ever be."

By: straw

straw — Wed, 17 Sep 2025 15:07:44 -0800

We all laugh about "memes" as images which shortcut a bunch of the conversation to set up a joke, but dropping back to earlier conversations about them, they're a set of communication mechanisms/shorthand which propagate ideas through the culture. And I think we can all see that modern social media, journalism, indeed all of publishing has waves of the same idea hitting all of the outlets all at once. And that quite often that happens because a publicist wanted it to happen. At the simplest level, every tech website got a press release that was targeted to their particular writing style at the same time, all some journalist had to do was change a few words to make it "theirs", submit it to their editor, hey we're on top of things because everyone else is making the same observation at the same time. And we know solidly that this is happening with AI companies, where even critical articles use framings that increase the anthropomorphic perception of LLMs, in terms that short-circuit our usual critical thought. Further, we know that we are often not aware of the ways in which we're being manipulated. We can all say "advertising doesn't affect me", and yet the huge amounts of money spent, and the ways that markets react, indicate that, yeah, we are all hella suscpeptible to being swayed. And LLMs are becoming part of the tool to help sway us (they can, for instance, rewrite that press release for a particular style type, so that the reporter is more likely to copy&paste larger bits of it into their article). I don't think it's too large a leap to say that it is the job of the AI companies to plant that "this is scary and inevitable" in the society, and that they're demonstrably doing so, and we should look at the all of these articles happening at once in the way we do fawning reviews on the newly released but very stupid product.

By: storybored

storybored — Wed, 17 Sep 2025 21:17:51 -0800

As a practising Luddite I am of course familiar with having all arguments in support of the Luddite position casually dismissed as "screeds" so as to avoid any temptation to engage with the actual points raised. What form would this engagement take? One way would be to run this essay through AI and have it come up with a rebuttal. Would that be useful? But we are probably all familiar with that rebuttal too. I call it a screed because it talks in generalities, indulges in hyperbole. Comforting if you are already in that particular tribe. But does it have any new insights? Look at this: "Not only is the ratio of AI's resource rapacity to its productive utility indefensibly and irremediably skewed, AI-made material is itself a waste product: flimsy, shoddy, disposable, a single-use plastic of the mind." This sentence is meaningless without some sense of quantification. What percentage of AI product is flimsy, shoddy and disposable? (He actually means "LLMs" and not "AI" but of course we are familiar with that mistake as well). Tylenol kills thousands every year. Is that a flimsy, shoddy product? No? Well, what's the standard for LLMs then? That would be an interesting discussion to have. But we're not going to get it because this is a screed.