I'm not sure what the primary signatories want, exactly. "No mathematics" safeguards added to frontier models, like the "no cybersecurity or bio" ones that currently ship? That seems like a bleak future. Protectionism like tariffs.
I think implication being organizations with 40,000+ employees and even more consultants and contractors plus a lot of budget are also using LLMs to draft public facing content instead of paying for content writers or even just proof readers .
It points to friction rather than cost economics. Same reason we are always surprised why multi billion dollar product companies with millions of install base prefer electron instead of a native app.
This does not imply that the organization is not paying for content writers or proof readers. It does suggest that they are not getting the value of paying for content writers or proof readers.
According to https://news.ycombinator.com/item?id=49417480, Anthropic hires writers who do not use LLMs, and reading their updates I also feel that they don't use LLMs for communication.
Pangram has an extremely low false positive rate. Even on adversarial examples.
One trade-off is even some obviously LLM text won't get detected by them, but they work really hard to ensure false positives are rare since a false accusation is much worse for society than someone getting away with LLM meatpuppetry.
I think you can't trust Pangram in a high stakes situation, but it is absolutely better than random noise at detecting AI-generated text. Which isn't surprising. If the distribution of probabilities can yield blatant Claudisms, it's not surprising it would also have more subtle deviations.
(Addendum: As I recall, LLM-generated outputs roughly follow Zipf's law, but the distribution still tends to have some subtle distinctions vs human text; pretty interesting, but I don't know where I heard this, so nothing to cite. Sorry.)
To be honest with you, I don't think I would be able to identify with high certainty that the bottom text is AI generated, so it definitely goes a long way to obscure the AI-generated nature of it, but I also think it still feels unnatural somehow. I realize my framing naturally calls into question whether I'm being honest, but I am being honest. Given my experience with similar "skills" (it's just chunks of prompt, nothing magical after all) I expected even less.
But still, this is all very strange because it wasn't that many generations of AI models ago that AI writing was a lot better - I'm talking GPT 4.1, Claude 4.5, that sort of era.
Anthropic newsroom posts on the other hand are carefully constructed and well-written in a way that I have not seen demonstrated by LLMs yet, past or present. I expect that they have well-paid staff who are careful with every detail of their public communications. When you put it that way, it almost feels unfathomable that they wouldn't, doesn't it?
I don't feel like either one of you really has a strong claim. "Doesn't appear to" is subjective, and of course it's impossible to prove one way or another.
You're simplifying the exchange a little too much. I said:
> Anthropic doesn't appear to use Claude for blog posts
My claim is literally the lack of evidence, which, yes, can't prove anything. This claim can be contested easily by showing evidence that they in fact, do appear to be using Claude to write prose in blog posts.
They said:
> Anthropic _absolutely_ does
Sounds pretty certain Anthropic is in fact, using Claude to write blog posts. Enough to emphasize "absolutely". That doesn't read like "I'm going off of vibes", that reads like "I can prove it". So, fine. Prove it. I don't believe it, and I want to hear the proof.
I'm skeptical, but it wouldn't be my first time being wrong. But flatly, if you make claims with this kind of certainty, yes I want to hear your proof.
My point in saying "Even Anthropic doesn't appear to be using Claude for blog posts" was not meant to be some striking revelation, I literally was considering it a prior to make another point. This on the other hand sure does sound like a striking revelation to me, that a lot of people across the Internet would be curious to hear. Like I'm sure these people would be interested:
I will admit that I am unnecessarily aggressive sometimes, but I wouldn't have changed my response much in any case. If you're going to make a strong claim like this, I want your evidence, not your vibes. Otherwise, the claim should be a lot weaker.
I also realize that this sort of brashness upsets HN a bit, but it is what it is. I pandered comments for votes in my 20s a bit, time to grow up, sometimes people won't like you. Sometimes I feel something deserves a brash response.
I don't really have any opinion on your tone; I just still don't agree with your framing. A lack of evidence would be neutral like "there's no evidence to indicate either possibility is more likely", but your phrasing conveyed that one possibility was more likely than the other. I pushed back against your follow-up because it seemed like you were arguing for a higher threshold of evidence than you provided.
Well, to be fair, you're correct. I am asking for a higher threshold of evidence. It's a stronger claim. I feel a stronger claim deserves stronger evidence.
I guess that's where we disagree. I feel like either claim is equally hard to falsify from the outside (partially because I've never had much confidence in my ability to spot whether text is from an LLM outside of the most glaringly obvious cases, and likewise don't have any clue whether people who have high confidence are accurate or deluding themselves).
McDonald's food is not even that unhealthy. I just tried a Burger King burger the other day and it's terrible. I think it's like 2000 calories in a single burger or something.
2k cal would be around 250ml of oil. or 350grams of peanuts. So doing with bread, meat, and other stuff alike requires over 600g of food, an excellent value to energy.
I get the impression that Nvidia employees don't care too much - I started seeing fully AI-written "documentation" on some of their smaller projects more than a year ago (i.e., before it was even slightly a good idea).
Yeah lol it’s basically the only reliable way to know how things work. Pre-AI, I read documentation for libraries that I used almost every day.
And now with AI I’m using it to fact check Claude. And still reading it for myself to understand why other peoples code is written a certain way. It’s basically the most important thing to reference when coding.
Sure today Claude can just read the library code and tell you what a function does or how to do something. But it still won’t tell you why something is a certain way or won’t figure out specifically-designed usage patterns as reliably as the author telling you “this is an example of doing x”
I really appreciated a friend reaching out to me with some PHP questions today. It was, to me, fairly basic but he was having a hard time grokking the documentation vs reading what his coworker wrote (some code using output buffering).
I brushed up on the docs since I haven't touched it in a couple years, explained my understanding of the ob_* functions, and gave him a very brief demo on a PHP playground.
He could have asked any LLM to tell him what that chunk of code did, and to explain the three functions, and instead he reached out to me. That felt _good_. Talking shop has always been a good way for me to form connections, because the pressure to socialize becomes task-oriented and you start to learn about how people think and feel, and that opens up easier paths for actual connection. It was nice.
Just like the Old Internet still exists - niche websites, mailing lists, probably a BBS or two (likely more right?), the pre-LLM world will trudge on, for a time. I hope LLMs actually lead to good things for people in the long run, and for now I personally will remain sparse in my usage of them.
The good news is your attention to actually reading and understanding documentation will differentiate you more and more as others (short-sighted, IMO) outsource understanding to an LLM.
when you start to internalize that these kinds of statements are an indication of how the average developer of the last 10-15 years operated the adoption rate of AI makes a lot more sense
Do you have the stomach to walk into a high school in the USA these days? Teachers use AI to generate assignments. Students feed the assignments to AI and submit the responses. Teachers feed the student submissions to an AI for grading.
As a high schooler going to a school with stricter rules on AI than most in my area, I can say that it's been going downhill ever since GPT 4. Teachers constantly use AI to create assignments(my French Teacher regularly handed us work with GPT 5.1 prose and emojis). Students are also rampantly using AI and bypassing school restrictions(We have a google account, making it easy to use Gemini if we just sign out), causing an inflation in GPAs and test scores. There's no easy solution to the problem, banning AI-tools only help somewhat as even typing into Google has AI web results, and students are quickly overcoming ways to restrict them. I have a friend that vibe coded an application that allowed his Mac Mini's desktop to be mirrored on his school chromebook, bypassing every restriction with sub 1-second latency. Of course, that opens the can of worms to whether schools should allow students to use AI...
> Of course, that opens the can of worms to whether schools should allow students to use AI...
We are starting to see results indicating cognitive decline due to AI in education, so no, we should do everything possible to ban it except for very limited fields.
LLMs aren't calculators or even computers, their generated output is too flexible, generic and basically starts replacing thinking.
Most likely they should only be allowed during late highschool years or just at university level, when people at least have a chance to learn how to research on their own.
I know many teachers who actually have respect for the profession, themselves, and the students. Thankfully that means they don't do this.
Whether this is a widespread macro trend is another issue, and would be terryfying.
If true, however, it would reflect on the values of the organization: we have spent decades underpaying teachers, and doing a poor job of pretecting schools from frivoluos lawsuits. Add into that, districts have thrown money into new buildings, have been suckered by Big Tech to adopt their policies (common core was pushed by Big Tech and has been a distaster as well as computers in classrooms). As a nation (the USA) we can't get our act together for a rigorous national exam, etc etc.
About half of the states in the USA require the ACT or SAT for high school graduation.
Alabama is one state that requires the ACT. The mean score in Alabama is below 18/36. Wisconsin is another. Its students score on average about 1 point higher than the national average of 19.4/36.
If you prefer states that require the SAT, the mean SAT score of students from Delaware is less than 980/1600, about 50 points below the national average.
I'll leave it to others to argue about whether these exams are rigorous.
I have two kids in engineering programs at a state University. They are allowed to use AI for homework assignments, but the homework is no longer worth any credit. They have a lot more papers, quizzes, and tests in class that count for their entire grade.
FYI, the world is a lot more decentralized than we think and even during the Dark Ages, guess what, that was happening in Europe and many places in the world were booming scientifically, technologically, etc.
Meh. It's not like we forgot how to make copper wires for landlines. We'll be fine. We'll live more or less like in 1880 or 1920, it's not a horrible life. I do hope we get to keep antibiotics, though.
It's true of most work in many and soon most white collar jobs, too. Claude writes some dense useless thing, everyone else uses Claude to summarize and write a reply to the thing. The Claude-submitted PRs get automatically reviewed and commented on by a GitHub Claude review bot. The programmer asks Claude to check out Claude's review comments to Claude. Claude pushes a commit to the branch and writes a comment. The Claude review bot reviews the commit and leaves a comment. The human [...].
My hot take is that it's not really that terrible in the long run for work since I think LLMs will probably be nearly or actually AGI and better white collar workers than most humans within 5 years of today. But it is very funny and surreal in the meantime.
It is definitely bad for school, though. Kids IMO should actually be encouraged to use LLMs but not in or for class work outside of an AI best practices class. Probably stop giving them homework (90% will always try to find a way to make AI do it) and have them solve problems in class hours with no electronics so that they're forced to not defer learning. This will become even more important once we have AGI.
Dude I am in slop fucking hell right now. There is still room for a human touch, without which the agents will lever us harder and faster into a world of incomprehensible garbage.
I totally concur. I'm almost lost for words at this stage. I need me some land to grow vegetables on and that's about it. Maybe some chickens. Every single day brings more despair (and not the prosperity we were promised).
It is the number 1 thing I cannot stand with Claude slop. It's a sort of anthropomorphization of language. Every "thing" does, produces, feels, wants, asks, answers, etc....
- "Launch is checked"
- "Question is asked"
- "The implementation answers"
- "The model wants"
- "The results name"
- "The connection surfaces"
- "The prompt wires"
- "The feature rides the mechanism"
Every single fucking thing is alive, wants things, and does things.
It's terrible. Infuriating. I want to rip my eyeballs out reading this filth. All. The. Time. "The anger is real".
Create any page with a file uploader. They all look the same now. It's like the Twitter Bootstrap days of responsive design. You'll get an icon which looks like ones on (on the drop space) those sites which are like "you must wait 60 seconds for this file to download".
It's so horrible. The human element has been completely removed and replaced by..... mediocre.
The human element has been completely removed and replaced by..... mediocre.
No it hasn't. The human element is still there, prompting the LLM. The change is that the human is happily accepting the first thing they get rather than critically looking at it and seeing a problem.
I don't think it's that because I see a lot of this in businesses where the human isn't paying the bill, or is even aware of what the bill is.
Humans are seeing either a shortcut to go faster (accepting low quality to move on immediately; reasonable if they're short on time) or a shortcut to lowering effort (accepting low quality because they don't care; not so reasonable but probably has a deeper root cause).
I hadn't read the article and read this comment as though NVIDIA themselves were implying that this library was checked but not trusted by them since it was fully LLM generated.
It gets fun when someone uses an uncensored model to bypass a refusal, but they accidentally pick one that was trained for erotic writing and brings its particular talent to the documentation task.
Yeah. I think if the text is written for other machines, then by all means have an LLM generate it, but if it is intended for a human audience, have a human being write it.
We are still much better at writing in a way that doesn't waste other people's time.
Was almost going to praise this article for not overtly sounding like Claude, but then I read this:
> and the incentive structures we have built are the experimental apparatus.
> Astra improved a term in a bound on these gaps that had remained unchanged for more than 80 years. We’re sharing the proofs and abridged chain of thought and verification materials for both results.
Looks like they listened to Terry Tao’s request for CoT in his talk on LLM use in mathematics?
reply