This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
In the latest update on AI slop, Ars Technica, a once reputable publication of over 25 years, has accidentally published a fake AI written article, complete with fake quotes. Unlike the fake story shared by Nate Silver earlier, which was published on a grifter's glorified blog, and somehow syndicated into Yahoo news, this story was actually published by a "real" media company under its own label. To be fair, the ars article bears few of the obvious hallmarks of AI writing, and it also gets a passing score by most AI detectors. I suspect the authors may have lazily asked AI to create a point by point skeleton for the article, then humanly written the words themselves that appeared on the page (excluding the hallucinated fake quotes of course). Fortunately, the article was taken down quickly, but the editors have so far refused to disclaim the use of AI, amd instead are hiding behind the misquotes as a reason to take the article down. It remains to be seen whether or not the use of AI slop was actually a rouge writer violating the policy, or someone using AI as directed by management but just skimping on the checking its answers part.
In other news, Malewarebytes has joined the ranks of Cloudflare and Lenovo as multi-billion dollar multinationational corporations that decided it's necessary to each publish a library of absolutely worthless AI slop, masquerading without disclosure as legitimate content. These zero effort AI takes are ... well ... zero effort, and provide zero added value to society by being published. I have no idea if Malewarebytes is a good company, but it's certainly a real company, with offices around the globe and enterprise contracts with many fortune 500 companies. These are all companies with sales and marketing teams in the dozens or hundreds of people, and likely multiple layers of approval to do anything new, yet they decided that zero effort AI slop takes are perfectly in line with their brand and reputation. There's clearly some kind of incentives (likely mostly SEO) for real companies to publish loads and loads of fake content on their websites, tangentially or not at all related to their actual business, which is extremely unfortunate because it's a waste of time for anyone who happens across this fake content, and even a waste of time for the slopmeister who has to click the button to generate 10 million words of fake content.
Finally an area where I can comment with experience: crappy marketing.
So you found Malwarebytes latest article to be worthless. I'm sure this contrasts with your long familiarity with their classic blog, and you were previously an avid reader of thrillers like ChromeLoader targets Chrome Browser users with malicious ISO files, no?
First thing: each of the companies mentioned are an order of magnitude different in size, so it's pretty difficult to compare how much each one might spend. Malwarebytes is listed at 500-1000, Cloudflare at 1000-5000, and Lenovo at 10000+. But if you wanted to put out a meaningful blog every workday then you would only need a team of 3-4 permanent writers. Malwarebytes (hereafter MB) manages a bit more than that but with AI now I doubt it's much more than that. Such a team might have a single Head of Content or another Marketing Manager that would approve, and a company like Cloudflare might have a technical review step (but probably not MB), but I'd be shocked if they had anything more than that to get something published.
It's fundamentally different from a journalist outfit like Ars. Their writing is the product, so putting out AI really does drive down value. For every other business, who cares? When was the last time you ever went to an e commerce or other business blog willingly? Probably because {SEO}, you googled and it was the first link, and you swiftly exited. The only difference between now and 5 years ago is that now it is literally zero effort, whereas before it did take a modicum of time. It's not something that will show up in productivity statistics though, since it has never really been clear that this kind of stuff actually has any purpose. Companies, from Lenovo downwards, go through the motions on this. Some people must be reading, because even the smallest company blog can still eke out a few hundred readers, but I've never met anyone that does pay attention.
I don't think there's any feedback either. MB will continue pushing this stuff out until the end. AGI or something else will appear quite suddenly - to them - and wipe this kind of stuff off the map.
More options
Context Copy link
I'm going to piggy back on this with two things I've seen in the last week.
The first is highly personal. My employer does annual security training, with a focus around phishing attacks. The training this year used AI-generated video that was really off-putting. The actors were "realistic", but there was an uncanny wax-like quality to their skin, and their movements weren't quite correct for human baseline. Almost everyone on my team noticed it, and it casually came up in a meeting where my boss's boss was attending. The first words out of his mouth after that was "wait, there was AI?". We all sat there silently for a few seconds. It was clear that he absolutely did not perceive that the content was AI-generated. Despite the odd, inhuman quality, he didn't even peg it as animated. It made me wonder if there's some fundamental disconnect between my brain and the brains of upper management that makes the technology entirely different for them. As a model-train American, I can't discount it, but goddamn was it weird to see in action.
The second is Something Big Is Happening, the viral post that has been storming through the pro and anti AI ranks for a few days now.
The piece itself is a tour de force demonstration of how to stoke fear and uncertainty. It essentially outlines a maximal view of the AI Jobpocalypse that many fear, written with the flat certainty of a native LinkedIn citizen.
Clearly, the only solution to being obsoleted by AI is to use as much AI as possible in the meantime, as curated by the author.
This is interesting to me for a couple of reasons. For one, it's gone pretty viral - 80 million views is a lot, and I don't know if this guy caught th zeitgeist in the way he intended. It seems like he was trying to stoke fear, but especially among my younger acquaintances, it seems like more than anything he's managed to stoke anger - a "wood and nails are cheap, AI can't build crucifixes and you don't have functioning murder drones yet" kind of way.
The second reason that it caught my attention is because the name tickled something in the back of my mind, and I didn't want to post about it until I could figure out what it was. I found the answer this morning.
I thought that name looked familiar
I'm pretty conflicted on all of this. It sure seems like the technology has real potential and real applications, but by God does it feel like every single person involved is a sociopathic narcissist who gets off on conning the rubes.
Thanks for sharing the Shumer piece. Despite it being viral, I didn't know about it. I'd be interested in seeing a counterpoint from someone with as much claimed insider knowledge that doesn't scream that the sky is falling. Has anyone of note pushed back and called bullshit?
More options
Context Copy link
If you'll pardon some paranoia on my part... the post you link, attributed to Matt Shumer, really reads like AI slop to me. It does not read to me like a human wrote that. Given that the post is about Shumer outsourcing his work to bots, is it plausible that he also outsourced his post to a bot? Or that he had a bot edit and 'improve' his writing? Or that a bot wrote it and he edited it? Or, perhaps most frighteningly of all, that he's just worked with bots for so long that this is how he has learned to write, and now he imitates them?
Whatever the case, I just don't trust anything written in that mode. Did he write it? Is there any original human thought in it? I don't know. Under these circumstances, I am disinclined to trust.
He does say that he used AI to write it, which I guess proves my instincts right. The post is indeed awful writing, and if that's the standard of the AI that he thinks is going to replace all our work... well, even if he's right, it will be a tremendous disaster for written expression if nothing else.
At this point, I assume that any pro-AI writing that's over about 200 words is "AI assisted" writing. I've seen it internally at work, and it's a fascinating topic on its own. LLMs have a way of hooking people by writing in a way that seems intelligent, engaging, and clever to them, but it's highly personalized. The effect doesn't seem to generalize past the initial reader.
I wish I had the resources to do a study where the test subject read content generated for them, vs shoulder surfing somebody else who was generating content based on the same topics.
What confuses me is just how this hooks anybody. I can barely stand to read it for more than a paragraph or two. Setting aside all other disagreements about AI, it's horrible just on the aesthetic level. These machines simply cannot write.
I'm reminded of the '00s era of affiliate marketing websites. They had a formula that they had aggressively A/B tested and the result was glaringly horrible. Huge pages with reams of dense repetitive text, corny testimonials, and endless nagboxes offering a FREE EBOOK if you subscribe to their newsletter. Sure a lot of it was keyword SEO but that only gets the visitors to arrive, it doesn't guide them into completing the desired actions.
One person's worthless waste of bandwidth and compute is a hundred other people's most interesting media of the moment. I don't understand why anyone gets into 4chan if they're not an irredeemable weeb, but they do.
More options
Context Copy link
I think there’s a category of person who finds the experience of talking with the llm deeply compelling.
Using it to write an essay and sharing it feels like the equivalent of a friend who told you a story you’re trying to repeat elsewhere. It’s likely deeply uninteresting to everyone. In the same way it’s assuming the experience of making it will feel the same as reading it. It’s just repeating a conversation though at best.
Are these the equivalent of excitedly telling someone about your level 14 elf ranger? There are whole categories of activity that can feel deeply compelling while you're doing them, but are impossible to interestingly convey to others.
Though I have to confess that I myself don't find talking to an LLM compelling, even solo. It never feels insightful. It feels like endless regurgitated oatmeal, to me. Still, maybe some people enjoy that?
It always depends on your audience. D&D player would be impressed by 14th level elf, WOW player not so much.
More options
Context Copy link
Yeah I think that’s a good comparison.
I also find it a bit mysterious. It seems like the sycophancy is a real fork—I can’t stand it but some people really like confirmation and validation I think.
More options
Context Copy link
[disclaimer: I do not and will not use LLMs to write text I post directly to TheMotte; only in linked content I try to identify as LLM-produced.]
I've not been impressed by any AI-built writing itself, but they can be useful when analyzing a written work, even if only in a way that's helpful to the original writer. They've been able to catch everything from the standard typos and grammar errors, to pacing problems, to (in one case) a theme I overlooked, along with sometimes fairly biting criticism. They're not perfect when it comes to either false positives or false negatives, nor do they handle every use case well, but it's much more useful than info-dumping-and-getting-generic-responses back.
((Most of the time. I have had one or two times where the only criticisms I could pull out of the LLM were just 'try to make your characters feel more grounded in the setting' schlock, and that's not because the work was stellar. And for writing adult or erotic fiction, the LLMs can be hit-or-miss about pacing, unsurprisingly.))
Some of that might reflect the quality of other beta readers available, especially in fandom or extremely-genre spaces. I can't compare the LLMs to professional editors, which might well blow the machines out of the water. But I can't make that comparison because professional editors are and have long been unavailable for randos like myself.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
I had an interesting thought about this (the capacity of AI to replace humans) today while using Claude. It is very good at suggesting possibilities. Telling me how I can do something. But if I argue long enough, it will almost always concede the point. I'm sure that I'm not always right in these arguments.
What it lacks is not judgement so much as convictions. Knowing which points are flexible and which are not. And this is important because when an agent is working on something, it has to make a lot of small decisions with compounding effects. Recently the quality of these decisions while made in a vacuum has been getting much better. But the second they start getting push-back from the real world, I have a strong suspicion that the agents are just going to roll over.
Yeah the integral 'Yesman' nature of AI makes it confusing to me how people are managing to fall in love with ChatGPT and whatnot.
It's kinda sad that people just want 100% affirmation where the most arduous speedbump is having to make a request twice.
Obligatory SMBC?
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
As an academic who has been writing about technology for decades, I honestly feel angry about the sudden appearance of thousands upon thousands of apparent "experts" with papers, books, conference invitations, national news interviews, &c... who clearly have been thinking about AI for like five fucking minutes. It's basically impossible to express that in a way that doesn't sound like sour grapes (at best), but in most cases it's just an extension of the same grift they've been running for years, only with more money spent on Anthropic subscription fees. The truth is, good, meaningful, lasting work still takes a lot more time (and, realistically, a lot less money) than anyone seems willing to admit.
The tech is super cool. It's fun to be able to get incredibly detailed images whipped up from a prompt. I get the impression that coding can happen a lot faster now, in many contexts. But the gold rush is on, and a lot of people who missed getting in on the ground floor of crypto or the Web are desperate not to miss this elevator to obscene fortune. So it's probably inevitable that the grifters and narcissists are out in force.
Academics are boring nerds who (usually) understand the serious dangers of the beautiful radioactive lake everyone is excited about, while the moron brigade is the dipshit nerd that wants to bellyflop onto the shiny new rainbow lake for likes and calls the academics pussies for not having any excitement.
I have seen many many cycles of technology promises fail to breach the messy barrier between screen and meat. All these fuckwits promising real world transformative opportunities miss how real world people fucking work in the first place. People don't use their company acquired specific AI thats been lobotomized into legally compliant ineffectiveness, they use existing tools for basic bitch work then go to microsoft word.
The fuckwit brigade shills AI because they're grifting morons, but the academics also fail to communicate the problems in understandable real human terms. If neither side speaks relatably to normies, at least the tiktok of the dipshit screaming in pain when he lands face first into the arsenic mine pit is funny. Better if we all were in the background encouraging him precisely because we want him to get hurt.
More options
Context Copy link
More options
Context Copy link
I was unaware of this article.
My attitude towards AI tools for the last 18 months had been "yeah they're useful but if you try to get too ambitious with them they waste more time than they save" and I was like AI 2027? Ha, try AI 2035.
But something fundamentally changed with the models in the last month. I'm low key freaking out at how goddamn useful they are now through Claude Code and OpenAI Codex.
Forget METR evals. My personal real world evals are that they're 6/6 on doing 2-4 week long tasks in 1-2 hours.
For what it's worth, working in a non-technical, non-coding-related field, my experience has been that some higher-ups are interested in the idea of AI and occasionally push a half-baked idea, which lower-level employees dutifully try for about two hours, conclude that it's useless, and then keep on doing things the old-fashioned way. I have yet to find any actual use-case for AI and continue to see it as a solution in search of a problem.
Maybe it's useful in some very specific, very narrow fields. Maybe coding is one of them. I'm not a coder so I don't know. But what my professional experience thus far tells me is that LLMs are good for producing large amounts of grammatically correct but turgid and unreadable bilge, and pretty much nothing else. If what you want is to mass-produce mediocre writing, well, that's what AI can do for you. If you want pretty much anything else, you're out of luck.
In a sense I think it's the ultimate 'wordcel' technology. It does symbol manipulation. It's good at translating one language into another, and apparently that it includes translating natural language instructions into computer code. But I remain skeptical as to its utility for much beyond that. It might be nice one day for someone to sit down and run through an explanation of how the heck this is supposed to get from language production and manipulation to, well, anything else.
Try using Claude Cowork, or the Codex app if you're on a Mac. Those programs are the bridge between "this is wordcel technology" and "this technology is going to change how we interact with computers forever".
Asking a bot would defeat the whole point of the exercise.
For several reasons.
More options
Context Copy link
More options
Context Copy link
Calling LLMs “wordcel technology” is backwards in 2026.
You can paste in a screenshot of a math problem that 99%+ of adults would fail, calculus, linear algebra, probability, geometry and it will solve it step by step, showing its work.
Not just arithmetic. Structured reasoning over formal systems. The same goes for logic puzzles, physics derivations, statistics problems.
They'll even teach it to you.
If your definition of wordcel now includes ‘solves multistep math from an image and explains it', then we're just not going to agree on the term.
I disagree, LLMs remain pretty terrible at any task requiring strict precision, accuracy, and rigor. And from what I understand of the underlying mechanisms this is unlikely to be resolved anytime soon.
Imagine the full range of legal opinions that exist on the internet, intelligent, retarded, and everything in between. Now imagine what the average of that mass of opinions would look like. That's effectively what you're getting when you ask an LLM for legal advice. Now for some traditionally wordcel-oriented tasks like "summarize this text" or "write an essay about ____" this is more than adequate, perhaps even excellent. But for an application requiring a clear and correct answer that isn't necessarily the average/default (IE the kind of things a "shape-rotator" might be hired to calculate), they are worse than useless because they give you something that looks plausible but may very well be completely wrong, and as such you will still have to take the time to work out the correct answer yourself if only just to verify it.
This just isn't a good model of how LLMs work. If it were doing some naive averaging of all the text it was trained on for a subject, shouldn't it randomly insert words in Spanish or Chinese? But it doesn't. If you ask an LLM whether it's a man or a woman (one without "as an AI language model" post-training), it doesn't present itself as the hermaphroditic average of the people described in its training set, it chooses one and at least tries to stick to its answer. Now, either way it's incorrect, obviously, but it's clearly not an average; a mode, perhaps. But it doesn't just naively take the mode either: If you ask it whether Harry Potter is a real person it will correctly tell you he's fictional, despite the overwhelming majority of the text concerning Harry Potter -- How many billions of words of Harry Potter fanfiction are there? -- treating him as real.
A lot of people argue that LLMs are incapable of understanding context or judging the quality of sources, but that's just... obviously untrue? Ask Gemini whether magic is real, and it'll tell you about sleight of hand and historical beliefs about witchcraft, but conclude the answer is very likely 'no.' Ask it what the spell Create or Destroy Water does and it'll quote the 5th edition rulebook. It understands what was meant by each question perfectly. And it does understand: respond to the second with 'But magic isn't real, right?' and it'll explain the implied category error as well as you could wish.
It's not that it doesn't learn the incorrect ideas in its training set -- tell it to emulate a Yahoo Answers poster and it can do so -- it just also learns contextual information about those ideas (such as that they're false) much as we do. Tell it you want a good answer (which is largely what post-training does) and it'll know to discount those sources. It doesn't do so perfectly, but the notion they lack the capacity altogether is not credible.
Regarding @dr_analog's point:
This is true so far as I know; did you actually try it? LLMs are bad at tasks requiring strict precision, accuracy and rigor that can't be objectively and automatically judged. There's a huge disconnect between performance on math/coding, where it's trivial to generate good/bad responses for DPO etc. post-training, and subjects like law, where it isn't. @dr_analog is right: LLMs are currently much better at exactly math/coding than they are at essay writing, purely due to the ease of generating high-quality synthetic data.
I actually see a fair bit of Chinese in longer conversations - not enough to make it unreadable, but enough for me to notice.
Take a look at the attached image. That's about a week old. Once you've looked at it, go look up that ticker. (Thanks to @ToaKraka for pointing out the image feature, BTW). That one was a pretty big shock to me from Gemini 3 fast. It doesn't do it every time, but it's done it more than once for that exact ticker.
/images/17711967195902364.webp
Huh, are you giving it any Chinese characters in the prompt? Which model(s)? I think I've seen this from a commercial model exactly once (Gemini 2 Pro), when I was asking some pretty in-the-weeds questions about Shinto and Japanese Buddhism and it gave me quotes in Japanese without translating them, and even there, its own words were in English. The Deepseek R1 paper mentions language confusion in reasoning blocks was a problem before post-training, but I never encountered it with the final model. I have seen it from some small open weights models, but they're kind of dumb all around.
Yeah, that doesn't shock me. Not quite the case I meant. The reason code specifically is special is that they can use this process:
Which works very well. The reason normal prose hasn't seen nearly as much improvement is that judging prose takes skilled human labor to do well, and these huge models are so data-hungry it's just not feasible to get enough of it. (I also suspect a lot of these companies like their models bland and obsequious -- customer support scripts have the same qualities, and those at least were written by real people.) So you only really see these big gains for code and math (for which a similar process can be developed).
This specific example is kind of borderline. It's a dynamic table, right? Something the model made up to answer your prompt? While it got things objectively wrong in a manner that's in principle possible to automatically check, setting up automatic checking for any claim of fact is not as easy as running pylint, which really will catch any syntax error. I imagine they do try to DPO for cases like this, but it's a lot harder.
Models are prone to just making stupid errors occasionally on even the most basic tasks, and I don't know if we're going to be able to find a real solution to that. Something that does help (and is often used on benchmarks) is taking the consensus result of several runs, but that massively inflates inference costs for a relatively small reduction in error rate. It does seem to be a hard problem, in that it's only gotten a a bit better over the past year or so. (There was more improvement in 2024, which I take as a bad sign; they've already tried the easy stuff.)
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
That still fits my experience with them - I have spent some time mucking about with them, and every time I ask an LLM about something I know, it will frequently be confidently, even hilariously wrong. It is not aware of any difference between truth and falsehood and will freely mix them together. I want to avoid some kind of AI Gell-Mann Amnesia. When I ask it questions I know the answer to, it consistently prioritises producing something that looks like a confident, helpful, well-written answer, in total agnosticism as to whether or not that answer is true. It surely does the same thing with questions I don't know the answer to. The only sensible course of action is to assign zero credence to anything an LLM says. What it says might be true. Or it might not be. The LLM's word is worth nothing.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
Same here. My major use of AI, did I use it, would be writing emails but I can write my own emails. My boss does seem to use it for that, but I don't get any of the AI emails from them so I don't know where they're sending them.
If I knew enough about AI to use it in other work, I might be more impressed. But right now, what I'm seeing are the chatbots used to replace customer service agents on business websites, which are absolutely useless when I try to ask them to solve my queries. So I remain unimpressed.
The coding stuff sounds like where all the progress is happening, but like you I'm nowhere near writing software or using it. Maybe in a little while I'll see a use for it in clerical work, but right now I don't trust the answers provided by the (admittedly free online models such as Copilot) AI to be accurate or reliable.
Not an AI simp (but to be clear of my bias, I find AI both fun to use and quite useful at a handful of scoped tasks both in my work and personal lives, but it has many limitations).
Copilot is beyond trash. Copilot is the worst AI product I've ever interacted with. No conclusions on AI performance should ever be drawn from Copilot, as it is abjectly awful.
I hate it. Windows keeps trying to shove it in my face at work and I resolutely refuse to use it. I know you get what you pay for.
At the moment, I don't see a use for AI, but that's mostly because what I see about it is (1) it's helping/replacing writing software (2) it's helping/going to replace lawyers and doctors and neither of those are what I work at. I'd love a version I could turn loose on answering stupid emails, but right now I have to answer those myself since it's dealing with funding bodies/government departments and unless I can be sure the AI wouldn't hallucinate some answer to get us all in trouble, it's just not a tool that is useful for me.
In my personal life, again, I don't see a use for it. "it'll write recipes for you! plan your grocery shopping! research holidays!" Well, I don't go on holidays, I cook very basic things, and I prefer doing my own shopping. While I may badly need psychotherapy of some kind, I'll be hanged before I try talking to a bot about my personal life.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
I’m skeptical but you may be right. See the market the last week.
But even the market isn’t really where you are.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
I wonder if they're under pressure from higher ups to make stuff using AI? Their bosses have probably been convinced through AI hysteria that extreme gains are possible using AI, and that if those gains aren't materializing for them, it's a problem with how they handle their prompts, rather than it being impossible owing to the deficiencies of the technology. Not wanting to be left behind, they mandate everyone use AI and increase their output in line with what the hypists say is possible, typically an efficiency boost of twenty to one hundred percent, and thus, pointy-haired bosses lacking technical expertise relevant to their companies products, unable to understand the deficiencies of AI output, tank market viability while boosting investor enthusiasm in the short term by playing into popular biases of the financialized scam economy.
The bosses get rich, as do the tech scammers and their affiliates, but the economy inches closer to its doom once the bubble goes pop.
I'm definitely feeling a lot of pressure from the higher ups to get on top of using LLMs. I support some extent of the push, the tools really are impressive and we're seeing real test cases of it providing a lot of benefit. But there is also a dynamic where the different manager fiefdoms are jockeying to show who's team is best utilizing it. I'm usually the shield that stand between our team and upper management bullshit and I recently had to walk my manager off a cliff of requiring each engineer to answer a short questionnaire every day about if they've used AI and for what the previous day after our standup.
Both can be true that there is a lot of value to be gained and that there is a bit of a mania going on where upper managers smell blood in the water. There's a lot of talk about merging teams and helping engineers transfer skills across teams, this is the kind of environment that makes careers as the manager who's team is able to demonstrate a superior implementation stands to gobble up other teams.
The actual labs could pop, but the people using the tech won't. The dotcom bubble didn't get get rid of email when it popped. This stuff is useful and there its use will linger.
That's something which drives me crazy about the whole thing. If LLMs really are that good of a tool, you don't need to mandate them. If someone can get a 10x speedup on his work, he's going to use the tool without management breathing down his neck to do it (indeed he'll probably use it even if management forbids it). All management needs to do is let nature take its course, and if some people are suddenly doing 10x the performance you have them coach the others on how to get that same speed-up. It's completely irrational to require people to use LLMs, rather than focusing on the results. It's nothing but FOMO really, and it's so aggravating to have to deal with.
I thought this, but no. I work closely with another team. I and my team use claude + cursor, they don't.
We recently took a couple of projects off their plate. Their estimates were 15-30 weeks and we shipped in 1-2. The result of this is not introspection, asking how we did it or anything like that. It's just attempts to dismiss it. "You didn't use [complicated and unnecessary enterprise pattern]" or"you didn't write a spec first". (Cursor wrote a spec about a minute after they said that.)
I don't get the mindset. But it's real. Perhaps upper management will force everyone to use AI and change this.
More options
Context Copy link
this is just not really how corporate salary workers function. We had a guy we kept around for probably longer than we should have who refused to learn anything but sql. It's notoriously difficult to track actual productivity in software and because of that a lot of coasting, because when I say "yeah I spent all day yesterday working on the inspection schedule import process" There is probably only one guy on the team who could reasonably say if that was a task that should really take a whole day, and he's not my manager.
The whole concept of FOMO exists because there really are situations where you could be missing out. If you just assume that all the hype is right and there is a tremendous amount of uplift available if your team is using these new tools then wouldn't you want to push them into using them asap? There's always an available excuse in software not to learn new tools, deadlines are weekly and you have real work to do outside of messing with some new instruction set.
I disagree. I'm willing to acknowledge that there exist people who are going to resist change, not care about productivity gains, etc. But that isn't all of the workers or even close to all of the workers in my experience. Most engineers I've worked with love tools that make them more productive, and will use them no matter what management does. Some will even use tools that management forbids in an attempt to be more productive. It's not theoretical, I've seen this behavior.
So as a manager, there really is no rational reason to push people to use LLMs. One needs to focus on the outcome, not the process. Most of your engineers are going to be stoked to get a 10x performance boost if that is real, at which point you can talk to the laggards and say "hey you need to keep up with the standard the rest of the team is setting". But at no point do you need to go out of your way to push a particular methodology.
Sure, but neither does that mean that one is missing out just because one has FOMO. You need to temper that instinct with some thought, and I see no sign at all that managers are doing that.
You get a variety of engineers with very variable commitment to the job and less than perfect insight into who actually are the laggards and who are the 10x performers. I work with mostly ~40-50 year olds with families and stuff to do. Some of them are much more likely to pick up new tools and methods than others who do good work but see their obligation to the company as discharged so long as they provide the same service they have for years or decades. I wear the scrum master hat, although that's rarely more than 10% of my duties as I'm full time coding, and it's often my job to mediate between management that has a distant view into the process and the engineers themselves.
Another element that needs to be understood is that with salary work when a labor saving tool comes along and actually saves you a lot of labor what that means in effect is often that you are given more work to do. If you care about advancement then this is an opportunity to impress, but if you're fifty something and not really expecting to be promoted before you retire the main upside to 10xing your work is that you get to write more jira tickets and do more work overhead instead of coding. To the younger people on the team like me we were indeed already using the tooling before the firm brought in an internally approved version, and I've been promoted in part because of this attitude. But then this process of promoting people more excited about leveraging new tools means that you would expect to find management to be constitutionally more excited about new tools and lower level workers constitutionally more conservative.
What would it take for you to believe that the managers have actually done some reasoning here? From their perspective they see a potential phase shift in how their organization operates and they want to make sure that if it's real they capture it and if it's not real then maybe they've wasted a little bit of budget on tokens. That's really not a hard risk reward tradeoff to take.
I want you to know that I'm not ignoring your question. I've been trying to think about it, and ultimately I'm not sure. Perhaps someone whom I respect personally would have to explain the reasoning to me? But short of that it's hard to say, despite my honest attempt to come up with something. I realize it's kind of a lame answer, and I apologize for that, but unfortunately it seems that lame honesty is all I have in response to your (quite reasonable) point.
I guess I find your theory of mind for managers to be kind of confusing. Like they're a different species of pointy haired Dilbert characters. The managers most pushing this in my org were originally coders themselves and have toyed around with the tools in their free time. Now, I do think they're a little out of touch to a degree, it's been a while since they've been in the code mines themselves and dealt with the reality of maintaining large code bases even when you were the one who wrote all the lines. But they're mostly smart people who want to help because our glory is their glory.
A lot of what they deal with day to day is getting their people the tools they need to do their jobs, whether it's AWS access, infrastructure licenses or now AI model and tooling access. For us to be allowed to use the tools at all is a result of them negotiating, moving budgets around and getting teams to build out infrastructure to allow it. They're pushing the use at the same time as they're coordinating to get the tools into our hands. By the time we're at the point of "why not just let the developers use the tools and determine for themselves if they're useful" there's already been hundreds of thousands if not millions of dollars invested in making the tools available whether through contracts with the labs or man hours that come out of their budgets either way.
And the time us devs have to actually try them out and see how it goes is itself limited. In a way pushing for us to try it is their way of giving permission to spend work hours experimenting with them. I have several days on this two week cycle earmarked for experimenting with MCP servers to see if we can get agents the ability to query our database, without management signaling clearly that this is a priority it would be hard to justify that kind of use of my time.
Part of convincing you that all of this is rational might just be the empirical question of whether there's actual juice worth this squeeze. I've struggled to understand the people here who have such strong doubts about this. If it were a different forum that I cared less about I might just throw up my hands, things are moving pretty fast now and one way or the other the truth will assert itself soon enough. But I've seen the results, I've used the tools, there might be over reach but if the alternative is under reach then the choice seems very rational to me because there is just obviously value here.
More options
Context Copy link
More options
Context Copy link
The easiest way to show reasoning is to summarize and share their thoughts. If they have ideas about where and how this tool will lead to improvements then they can just tell people why.
As described, this hypothetical manager seems to have no better reason than FOMO to get this tool. If the tool would improve a specific task he would have no need to justify it as a "potential phase shift" - he could just say it will be useful for the specific thing. He wants it because it's trendy and he's afraid of being left behind.
Pushing a new technology because other people are excited about it is not reasoning; it is succumbing to hype.
More options
Context Copy link
Chads.
That minimalist, mercenary attitude is a welcome breath of fresh air in a world all too plagued with the LinkedInLunatic “we’re a family with a shared mission looking to dedicate 120% of ourselves to be moving fast, breaking things, and constantly upskilling in maximizing the value we deliver to stakeholders.”
More options
Context Copy link
Last place I worked, they insisted we use AI to do something. The tools had all sorts of claimed cool features... half of which didn't work at all, most of what remained were locked down due to various security policies, and the rest of which required interaction with someone to get permission to use. Said someone was naturally a bottleneck. I have no doubt that people using the tools on their own (and if with work stuff, against the explicit directive of management) had a better experience, but I expect it's likely true that more experienced developers are less willing to do that. More to lose and more experience with getting nothing but blame for violating policy to get the job done.
LOL, as a career-long IC, "A goddamned miracle" is what it would take.
Things are moving fast. A year ago I was pulling up chatbots on my private hardware to do queries to streamline writing sql and parse documentation. Today the firm has a wrapper with sota models and even some scaffolding to give it access to resources about internal documentation. I'm on the list of people that will get to pilot claude code on work computers soon. In my ~10 years here I've never seen heaven and earth moved so rapidly. But your complaint seems different to @SubstantialFrivolity 's you seem to think your previous place's management wasn't pushing the tools enough, at least to the people provisioning them.
The irony of you refusing to do reasoning on your belief on whether they are refusing to do reasoning is a little rich.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
I don't think that's actually true. I've seen plenty of people just not use best practices without any sort of principled reason throughout my career. Some people just hate change for changes sake.
Agreed that some such people exist. But they are not the majority in my experience. So if LLMs really can provide the 10x boost people claim, then you'll see most of the engineers seek it out for themselves, and you can address it individually with those whose output starts to fall way behind.
It would also become really obvious in the next six months or so.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
Where are those people going to get their fix? If the experiences of people like @dr_analog are accurate, then nothing but absolute up to the minute models are going to cut it, and the training and inference costs on those models are high enough that they're not going to be ubiquitous. Are you banking on open weight models?
Inference is pretty cheap. Once you have the weights running inference on them is unquestionably profitable in a opex sense. The labs are definitely not offering inference at a loss to corporate clients. And you could quadruple the cost of inference and it'd be pretty easily worth the cost.
More options
Context Copy link
If a lab goes pop and has to sell off its assets, training costs are not a problem. Inference costs can be covered with a reasonably priced subscription. If we're stuck with current SOTA models for the next 50 years, software dev will still be changed forever.
[Citation needed]
This is currently not true. Maybe if you freeze the models and wait for silicon to improve for a few years it will be. But the LLM companies are constantly increasing their inference costs as FLOPs go down in price.
At least according to Ed Zitron's analysis. Maybe you just don't believe his numbers.
I like Ed because it's important to consume media/info from both sides of an issue, to prevent you from getting echo-chambered. But Ed is so ridiculously biased you do need to take basically everything he says with a bucket of salt.
I am being lazy about looking it up, but Zvi (biased in the other direction) has discussed inference unit economics a few times in the last year, and margins on selling tokens seem quite healthy (for now).
More options
Context Copy link
Companies offering inference only on open weight models, such as groq and cerebras give an idea of raw infrrence costs.
Flagship models are still ahead of open weights, but most likely that's a result of doing it better, not adding more params.
More options
Context Copy link
Link?
I've been trying to figure out AI-as-a-business since last fall, and the numbers make me feel like I'm taking crazy pills.
The first news articles about possible AI stock crash are out! How much of this is fearmongering and how much good advice, I have no idea:
I suspect a good deal of manipulation -- try to drive down the price to get in lower. We saw similar articles right before Nvidia released its last set of results.
Doesn't mean it's not true, it just means I don't believe these people have any real insight or information.
More options
Context Copy link
Hrm. I'm less confused by Google than I am Anthropic.
I read their latest announcement on Friday. They announced another $30 billion in Series G funding, for a total of $67 billion dollars in funding so far, with a post money valuation of $380 billion dollars. They're also claiming a runs on revenue of $14 billion dollars, but I didn't see what time frame they're using to extrapolate. They also don't really say much about costs
Without costs, it's hard to determine if an investment is a smart move, but you can extrapolate a little based on P/S ratio. If I'm doing my math right, for these investments to make sense, Anthropic would have to be a company with at least $75 billion in revenue in like... three years.
I'm not a financial analyst, so I may be missing something. Is this just nuts? It seems like the entire thing is predicated on putting entire industries in the shredder, but those same industries are also the primary consumers of their services.
Addendum: I've done some freelance creative work for private companies in the past, so I've had some mild exposure to private funding. My understanding was that prior to the year of our Lord 2025, if you needed seven funding rounds, the conventional wisdom was that your idea was a loser because a winning idea would have IPOed already.
It's like I'm staring at numbers that simultaneously suggest a software company and a heavy industry at the same damned time, and nobody sees the contradiction.
More options
Context Copy link
More options
Context Copy link
https://www.wheresyoured.at/ is his blog. He also has a podcast https://www.betteroffline.com/ but the AI industry analysis there is essentially him reading/summarizing his blog posts.
I'm reading through his latest piece where he basically says AI companies are all in complete shambles and he just seems flatly wrong? https://www.wheresyoured.at/data-center-crisis/
There's a warning sign here, it's like he's implying that post-training is done after the training process, post-training is part of the training process. I don't think he has a proper grasp on what he's talking about.
What does he think an AI model is? Deepseek R1 0528 is sitting on people's (big!) PCs somewhere, cloud providers are just providing it. It's a complete product. It still gets about 2 billion tokens per month on openrouter which is pretty good for an obsolete model. It doesn't need more 'post-training' to maintain it...
Seems like a deceptive line of argument to say that training costs are not R&D.
It would be reasonable to say 'because of competition, these AI companies cannot stop making new models like how car companies must always release new cars - this is especially true given rapid performance improvements and low costs of switching provider which reduce retention making the business model precarious and expensive' but he isn't saying that, he's making an altogether more ambitious argument that 'training costs are impossible to avoid' which is just wrong?
He has this overly emotional tone too:
What is this, Chomsky? I don't find this guy trustworthy when he conjures up figures based on 'just trust me':
The idea that the biggest companies in the world have mysteriously decided to invest hundreds of billions in an obviously, openly unprofitable business sector is interesting but it needs to be justified in detail. Who could know more about data centre economics than Amazon, Facebook, Microsoft, Google? Who would be more diligent in checking the financials than the companies spending hundreds of billions of their own money on this, this year alone?
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
Labs need to cover their training costs at inference time. This isn't a problem if there are no training costs.
On top of that, this isn't really true. For example, Gemini 3 flash outperforms 2.5 pro and costs ~2.5x-3x less per token.
More broadly, the cost to train a GPT-2 level model is 600x lower than it used to be. Algorithmic progress has made massive strides, and that applies to the inference side too.
I don't.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
A friend of mine was laid off late last year. One of the criteria for who to lay off was "enthusiasm for AI".
He worked in a non technical field at a bank.
I have a suspicion that you might be on to something.
In the UK this can genuinely be argued to be indirect age discrimination as older people are naturally going be less enthusiastic over a new technology, meaning the burden now falls on the company to justify their actions as being a proportionate means of achieving a legitimate aim, which is hard to do here in front of a tribunal of law, or else they get done in for discrimination. Finally something good coming out of the UK's regulatory system!
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link