This is unfortunate, I wish we lived in a world where Google didn't have this much control over the Web. It's a shame that we got what we did, when the dream of "the Web" was so ambitious (and in my opinion, inspiring).
Politely, I'm seeing more and more articles with comments claiming "llm slop" and no justification. Have we forgotten that LLM looks like article text because it's trained on it?
People may not like the writing style - which I won't quibble with either way - but bad writing does not necessarily mean an AI wrote it. There was plenty of crap prose before AI.
Human bad prose is different from LLM slop. Should we go and list all the tells every time we find slop? The information density, how it goes from an idea to the next and back and repeats itself, the constant "it's not X it's Y", the stupid headings.
You start reading, then at some point you notice it just keeps going on and on and on, and it has already made all the points it was ever going to make, but somehow still reiterates. New section. More of the same. Then you stop and notice it: oh, the computer has wasted my time again. I am tired of this.
Why are you rushing to defend this article? What makes you think it was not written by an LLM?
Why are you rushing to defame this article? What makes you think it was not written by a person?
Yeah, I think you should list all the tells. It's lazy to do a post like you did, and it's getting tiresome. It's entirely possible he's just not good at writing.
This reminds me of Brandolini's Law. If I can paraphrase it just slightly,
The amount of energy needed to refute bullshit slop is an order of magnitude bigger than that needed to produce it.
FWIW, the author isn't doing themselves any favors with the obviously slopped up hero image either (unless that's the new Discord logo now). If you make me scroll past this before reading your prose, why would I expect what's below to be any better?
I only read the article just now, but the most jarring thing for me was the article didn't have "memory". The author introduces the idea that Chrome is the default twice. First it talks about Firefox and Safari's quirks for a few paragraphs, then it talks about how Chrome's at fault:
The quirks files aren’t just fixing broken sites; they’re often compensating for Chrome’s control over what “working” means in the first place.
The pattern goes like this: Chrome ships a feature, developers use it because Chrome dominates the market, and other browsers scramble to either implement the feature or add site-specific quirks to paper over the difference. By the time Safari or Firefox catches up, the quirk has already shipped to millions of users.
And then the next section jumps back to the previous topic of Firefox and Safari hacks without reference to Chrome, and then the section after that goes back to Chrome:
You might have noticed something. I’ve shown you Firefox’s about:compat and WebKit’s Quirks.cpp, but where’s Chrome’s equivalent?
Chrome doesn’t really need one, and not necessarily because Chrome is better engineered. The web is already built for Chrome. [...]
Chrome doesn’t add quirks; it sets the agenda. When Chrome changes how something works, sites update to match, and other browsers follow or break.
It's a slightly different point, but the text already told the reader about Chrome's control. There should be some sort of connection between the two sections. As it stands, it's like the author just forgot about it.
I could believe a human made that mistake, but I could believe an AI made that mistake a whole lot more easily. Esp when the human has already written three books and is presumably not a writing beginner.
LLMs, at this point in time, generally tend to write in a way where the sentence structures follow a very typical repetitive phrasing and overal very formulaic approach of story telling. It is easy to read, for sure. It also means the human "author" did minimal prompting to change that phrasing. Which raises questions about the content of the article, the correctness of things claimed, etc. To give that more context, generally speaking I find that there are lazy and non lazy ways to use LLMs. The more an article shows signs of heavily being LLM generated and not just used as an editor the more I have to ask myself how critical the author has been about the content of the article.
It also often means that the article could have been a lot shorter and to the point. As I said, there is a lot of repetition in the writing. The top comment alludes to that by saying there is an interesting story hiding behind the llm slop text.
It's also possible there could be ... a lot of repetition in the writing. Given how negatively people regard AI generated posts, and with good reason because I myself put weeks of work into things I write, I'd like to think there's room for reasonable doubt. Instead we're jumping to slap the scarlet A on and tar the post, regardless of its merits.
I mean, you've talked about repetition several times in your post. What would you conclude about you, on that standard?
It's also possible there could be ... a lot of repetition in the writing
Of course, that is why I said it is a very typical and formulaic approach of story telling, not just repetition. I also am not talking about repeating something to drive through a point. When I say repetitive writing I mean the way the sentences are structured and the patterns a text follows.
I'd like to think there's room for reasonable doubt.
Sure, in this instance I think that lands somewhere in the middle. I can see the author put in work themselves, but I also recognize a lot of the typical LLM generated tells. And again to be clear, I don't discount the use of LLMs entirely. If we are going to have an honest conversation about this I do feel like it should include the entirety of what I said as context matters.
I mean, you've talked about repetition several times in your post. What would you conclude about you, on that standard?
No I am ascribing bad faith based on 1) you ignoring the majority of my input responding out of context to things I said explicitly within a context and 2) doing that again even when I point that out and ask that you not do that.
Ah, so people think this is AI generated, but I didn't. I thought the RMS + bzr article was AI generated, go figure. I don't know if I'm worse than average at detecting LLM articles or about average. Thank you for taking the time to explain.
Many people (you included, apparently) seem perfectly content reading LLM output, which contributes to its popularity. If they didn't, nobody would care to post it.
I find it to be extremely verbose and shallow, saying the same things over and over, and it gets old really quickly. I am sick of it really.
Yeah the contrast is almost non-existent. Even in light mode it's not great (but slightly better than dark mode). I suspect the creator of the site has not actually viewed it themselves and just blindly coded it.
These sites sniff for Chrome and serve degraded experiences to everyone else, so rather than let Safari users suffer, WebKit lies about what browser it is.
Happens to everything in this industry, it seems. Computer manufacturers not rarely ship ACPI firmware that hides things from unsupported operating systems, forcing them to lie to the firmware by pretending to be Windows instead.
This creates a feedback loop. Developers build for Chrome because Chrome dominates. Their sites work best in Chrome. Users who hit bugs elsewhere blame the browser, not the site, so they switch to Chrome, reinforcing its dominance.
There are two feedback loops. The one mentioned here and the one where site owners say they don't see Firefox in their stats while their site is broken with it (it can also happen if the quirk is to change the user agent).
The implementation is the specification is a widespread issue in the field. In a previous job we used a workflow language that had conformance tests with the idea that there would be many implementations and the workflow would be portable.
The key issue was that there is low economic incentive to make things fully portable. You want your implementation to do extra things so people stay on your product.
You want to push ahead and build that new feature because no one got time to do software by committee.
The implementation is the specification because we be human society.
This is not new: non-majority browsers have had site specific hacks for the IE of era, as the alternative is the site being broken.
A couple of decades ago web developers made sites that only worked in IE and just said "use IE to have this site work", and now it's that same but with Chrome. It does not matter if Chrome is right or wrong: the sites only work with Chrome, and if the other browsers don't ensure that these sites get chrome's behavior people just say "this browser is broken" and switch to chrome.
I'm genuinely curious: do people think Gecko and WebKit leave these sites broken on principle? or that everyone should just use chrome and not any of the alternative browsers? Because those are the alternatives to site specific hacks.
In the first screenshot one of the compat sites is ace7.acecombat.jp
“Big sites” indeed….
Maybe they just do this on a per case basis and it just happens that “big sites” by their very nature are doing stuff that causes them to encounter such things a lot.
Probably many more users file bug reports about big sites not working. Also, for smaller sites it’s more likely people blame the site rather than the browser.
It's a product of bug reports. Large/high profile sites get noticed during standard testing/QA so get the workarounds immediately, slightly smaller ones are more likely to get bug reports (notice: the browser is getting bug reports, not the site), and then smaller and niche sites get work arounds if someone files a bug and the complexity of the workaround is balanced by the number of users (e.g lots of workarounds are basically "apply this quirk" not anything that requires really customized work)
This is the core issue though. You can either take the google position: the browser needs to be able to replace the entire OS, in which case it inherently needs to approximate the size of a large amount of a modern operating system, or you should adopt a position that it shouldn't in which case they can be smaller.
At the same time remember that electron apps depend on the browser being an entire OS. It's why even the most basic electron app is so huge (both binary size and in memory footprint), they clobber battery life so badly, and have such broken usability and OS interaction: each one is basically a virtual machine running an entire separate OS, which shares no common libraries with any other electron app, and duplicates huge amounts of the actual OS functionality.
I (as you might have guessed from the above) would have no problem if all electron apps disappeared overnight: they're all awful to use, despite huge amounts of work by google they all have terrible battery life impact, and they fail at the most basic of OS integration.
ambee | 23 hours ago
This is unfortunate, I wish we lived in a world where Google didn't have this much control over the Web. It's a shame that we got what we did, when the dream of "the Web" was so ambitious (and in my opinion, inspiring).
cosarara | 22 hours ago
There is an interesting story behind all the llm slop text.
orib | 21 hours ago
Unfortunately, the writing is so bad that I couldn't find it. The article may as well have stopped after the headline.
abnercoimbre | 20 hours ago
It's astounding they apparently wrote three (!) technical books. How tragic.
Rovanion | 20 hours ago
All who upvote this comment should probably report the story as spam.
eyesinthefire | 19 hours ago
The jump back and forth between sensationalizing it and giving cool browser compat override info was really jarring
classichasclass | 19 hours ago
Politely, I'm seeing more and more articles with comments claiming "llm slop" and no justification. Have we forgotten that LLM looks like article text because it's trained on it?
People may not like the writing style - which I won't quibble with either way - but bad writing does not necessarily mean an AI wrote it. There was plenty of crap prose before AI.
cosarara | 18 hours ago
Human bad prose is different from LLM slop. Should we go and list all the tells every time we find slop? The information density, how it goes from an idea to the next and back and repeats itself, the constant "it's not X it's Y", the stupid headings.
You start reading, then at some point you notice it just keeps going on and on and on, and it has already made all the points it was ever going to make, but somehow still reiterates. New section. More of the same. Then you stop and notice it: oh, the computer has wasted my time again. I am tired of this.
Why are you rushing to defend this article? What makes you think it was not written by an LLM?
classichasclass | 18 hours ago
Why are you rushing to defame this article? What makes you think it was not written by a person?
Yeah, I think you should list all the tells. It's lazy to do a post like you did, and it's getting tiresome. It's entirely possible he's just not good at writing.
dzwdz | 12 hours ago
This reminds me of Brandolini's Law. If I can paraphrase it just slightly,
FWIW, the author isn't doing themselves any favors with the obviously slopped up hero image either (unless that's the new Discord logo now). If you make me scroll past this before reading your prose, why would I expect what's below to be any better?
hwayne | 6 hours ago
I only read the article just now, but the most jarring thing for me was the article didn't have "memory". The author introduces the idea that Chrome is the default twice. First it talks about Firefox and Safari's quirks for a few paragraphs, then it talks about how Chrome's at fault:
And then the next section jumps back to the previous topic of Firefox and Safari hacks without reference to Chrome, and then the section after that goes back to Chrome:
It's a slightly different point, but the text already told the reader about Chrome's control. There should be some sort of connection between the two sections. As it stands, it's like the author just forgot about it.
I could believe a human made that mistake, but I could believe an AI made that mistake a whole lot more easily. Esp when the human has already written three books and is presumably not a writing beginner.
kghose | 20 hours ago
I need someone to be explicit about this comment and the attached thread. I found the article easy to read and informative. What is the problem?
creesch | 18 hours ago
LLMs, at this point in time, generally tend to write in a way where the sentence structures follow a very typical repetitive phrasing and overal very formulaic approach of story telling. It is easy to read, for sure. It also means the human "author" did minimal prompting to change that phrasing. Which raises questions about the content of the article, the correctness of things claimed, etc. To give that more context, generally speaking I find that there are lazy and non lazy ways to use LLMs. The more an article shows signs of heavily being LLM generated and not just used as an editor the more I have to ask myself how critical the author has been about the content of the article.
It also often means that the article could have been a lot shorter and to the point. As I said, there is a lot of repetition in the writing. The top comment alludes to that by saying there is an interesting story hiding behind the llm slop text.
I hope that clears things up for you?
classichasclass | 17 hours ago
It's also possible there could be ... a lot of repetition in the writing. Given how negatively people regard AI generated posts, and with good reason because I myself put weeks of work into things I write, I'd like to think there's room for reasonable doubt. Instead we're jumping to slap the scarlet A on and tar the post, regardless of its merits.
I mean, you've talked about repetition several times in your post. What would you conclude about you, on that standard?
creesch | 17 hours ago
Of course, that is why I said it is a very typical and formulaic approach of story telling, not just repetition. I also am not talking about repeating something to drive through a point. When I say repetitive writing I mean the way the sentences are structured and the patterns a text follows.
Sure, in this instance I think that lands somewhere in the middle. I can see the author put in work themselves, but I also recognize a lot of the typical LLM generated tells. And again to be clear, I don't discount the use of LLMs entirely. If we are going to have an honest conversation about this I do feel like it should include the entirety of what I said as context matters.
Not much, as I hopefully have made clear now.
classichasclass | 17 hours ago
I hope we can all grant authors that level of grace too.
creesch | 17 hours ago
You are again cherry picking. Which does not leave a sense of good faith with me as I just specifically asked you not to do so.
classichasclass | 16 hours ago
Since you're willing to ascribe bad faith to me on the basis of a genuine discussion, I wish you luck.
creesch | 16 hours ago
No I am ascribing bad faith based on 1) you ignoring the majority of my input responding out of context to things I said explicitly within a context and 2) doing that again even when I point that out and ask that you not do that.
kghose | 16 hours ago
Ah, so people think this is AI generated, but I didn't. I thought the RMS + bzr article was AI generated, go figure. I don't know if I'm worse than average at detecting LLM articles or about average. Thank you for taking the time to explain.
jfloren | 9 hours ago
Oh the bzr article was definitely generated too
cosarara | 18 hours ago
Many people (you included, apparently) seem perfectly content reading LLM output, which contributes to its popularity. If they didn't, nobody would care to post it. I find it to be extremely verbose and shallow, saying the same things over and over, and it gets old really quickly. I am sick of it really.
kghose | 16 hours ago
I guess I'm no good at detecting LLM output. It wasn't the world's shortest article, but I've seen worse posted on this site.
dubiouslittlecreature | 21 hours ago
I was waiting to see if anyone else had said it
cmcaine | 23 hours ago
The blockquotes are unreadable for me. (Maybe a dark mode thing).
Thanks for sharing some of the details of the workarounds.
Sharparam | 21 hours ago
Yeah the contrast is almost non-existent. Even in light mode it's not great (but slightly better than dark mode). I suspect the creator of the site has not actually viewed it themselves and just blindly coded it.
matheusmoreira | 22 hours ago
Happens to everything in this industry, it seems. Computer manufacturers not rarely ship ACPI firmware that hides things from unsupported operating systems, forcing them to lie to the firmware by pretending to be Windows instead.
dubiouslittlecreature | 21 hours ago
I don’t like that this reads like AI prose
vbernat | 21 hours ago
There are two feedback loops. The one mentioned here and the one where site owners say they don't see Firefox in their stats while their site is broken with it (it can also happen if the quirk is to change the user agent).
thisalex | 22 hours ago
As far as I remember, classical Opera (Presto) was the first who started to implement such compatibility layer.
kghose | 22 hours ago
The implementation is the specification is a widespread issue in the field. In a previous job we used a workflow language that had conformance tests with the idea that there would be many implementations and the workflow would be portable.
The key issue was that there is low economic incentive to make things fully portable. You want your implementation to do extra things so people stay on your product.
You want to push ahead and build that new feature because no one got time to do software by committee.
The implementation is the specification because we be human society.
hc | 12 hours ago
i think it's funny that firefox and safari pretend to be chrome via useragent
but chrome's useragent string has the fossilized remnants of pretending to be both a mozilla and an apple browser
this line of code has geological layers:
olliej | 15 hours ago
This is not new: non-majority browsers have had site specific hacks for the IE of era, as the alternative is the site being broken.
A couple of decades ago web developers made sites that only worked in IE and just said "use IE to have this site work", and now it's that same but with Chrome. It does not matter if Chrome is right or wrong: the sites only work with Chrome, and if the other browsers don't ensure that these sites get chrome's behavior people just say "this browser is broken" and switch to chrome.
I'm genuinely curious: do people think Gecko and WebKit leave these sites broken on principle? or that everyone should just use chrome and not any of the alternative browsers? Because those are the alternatives to site specific hacks.
3digitdev | 22 hours ago
Don’t read just the headline.
In the first screenshot one of the compat sites is
ace7.acecombat.jp“Big sites” indeed….
Maybe they just do this on a per case basis and it just happens that “big sites” by their very nature are doing stuff that causes them to encounter such things a lot.
uncenter | 19 hours ago
On the first page of https://searchfox.org/firefox-main/source/browser/extensions/webcompat/data/interventions I can see Bank of America, steamcommunity.com (Steam), DirecTV, Xiaomi, Santander, USPS, and Yahoo. Those sound like big sites to me!
sjamaan | 20 hours ago
Probably many more users file bug reports about big sites not working. Also, for smaller sites it’s more likely people blame the site rather than the browser.
olliej | 15 hours ago
It's a product of bug reports. Large/high profile sites get noticed during standard testing/QA so get the workarounds immediately, slightly smaller ones are more likely to get bug reports (notice: the browser is getting bug reports, not the site), and then smaller and niche sites get work arounds if someone files a bug and the complexity of the workaround is balanced by the number of users (e.g lots of workarounds are basically "apply this quirk" not anything that requires really customized work)
gcupc | 22 hours ago
Just another example of the "reckless, infinite scope of web browsers".
olliej | 15 hours ago
I'm not sure what you mean by this? Are you saying that everyone should just use chrome?
dbremner | 14 hours ago
They're quoting an old Drew DeVault rant which complains about the size of web browsers; here's the lobste.rs discussion.
olliej | 13 hours ago
Ah ok.
This is the core issue though. You can either take the google position: the browser needs to be able to replace the entire OS, in which case it inherently needs to approximate the size of a large amount of a modern operating system, or you should adopt a position that it shouldn't in which case they can be smaller.
At the same time remember that electron apps depend on the browser being an entire OS. It's why even the most basic electron app is so huge (both binary size and in memory footprint), they clobber battery life so badly, and have such broken usability and OS interaction: each one is basically a virtual machine running an entire separate OS, which shares no common libraries with any other electron app, and duplicates huge amounts of the actual OS functionality.
I (as you might have guessed from the above) would have no problem if all electron apps disappeared overnight: they're all awful to use, despite huge amounts of work by google they all have terrible battery life impact, and they fail at the most basic of OS integration.