Would it be possible to get versions of the files to see when certain words got added and I imagine certain words got removed? Could be interesting to correlate certain events, personalities, trends, memes, controversies etc with the changes.
It was added in 23 so there's potentially 3 years of updates.
This is about autocorrect learning words you like to use that it should consider potentially intentional use of words it might otherwise think you typed accidentally
Categories are used for training the autocorrect transformer model. It’s context aware, app aware, and adaptive over time. The 2023 wwdc keynote explains it all
Regardless, it’s non-censorship because autocorrect does not prevent you from saying anything
Apple's built-in dictionary flags on my Mac: 26.2, which is newer than the article's
15.6. Apple has changed several
things since:
┌───────────────┬────────────────┬──────────────────────────────────────────┐
│ │ Article (15.6) │ My Mac (26.2) │
├───────────────┼────────────────┼──────────────────────────────────────────┤
│ Total entries │ 199,019 │ 199,940 │
├───────────────┼────────────────┼──────────────────────────────────────────┤
│ Vulgar │ 1,050 │ 850 │
├───────────────┼────────────────┼──────────────────────────────────────────┤
│ Offensive │ 663 │ 727 │
├───────────────┼────────────────┼──────────────────────────────────────────┤
│ Sensitive │ 2,558 │ 2,748 │
├───────────────┼────────────────┼──────────────────────────────────────────┤
│ Oedipus │ vulgar │ sensitive (sex topic) │
├───────────────┼────────────────┼──────────────────────────────────────────┤
│ cooties │ vulgar │ clean │
├───────────────┼────────────────┼──────────────────────────────────────────┤
│ Biden │ clean │ now tagged politics topic, still no tier │
└───────────────┴────────────────┴──────────────────────────────────────────┘
These are unchanged:
- Trump is still sensitive with the politics topic. The lowercase card-game word "trump" is still
clean.
- Obama, Putin, Vance and Harris have no tier. Musk still isn't in the dictionary.
- Tom and Mick are still offensive, and pee, poo, niggle, underling, blacklist and Isis are still
sensitive.
- idiot is still sensitive, and now also has the catch-all "sensitive context" topic.
Words in the capitalized sensitive list that the article doesn't mention include Saddam, Grindr,
Oedipus/Oedipal, TERF and Viagra. Trump is still the only living politician among them.
This is interesting… does anyone have data in theirs? It also feels like something they could have put on an intern and it didn’t really gain any traction. Each app would need to “tell” the OS who the receiver is, and the article indicates there is no API to do that. Maybe it just knows, but how could that be?
Why does Apple need to keep these in a separate database? I thought they removed filtering, maybe subject to user age or parental controls, not added a whole mechanism to manage it.
Fun fact, the word “cooties” didn’t originally refer to a fictitious infective agent spoken of by children; it was originally a nickname for body lice.
> Coined during World War I by British and American soldiers in the trenches of Belgium and France to describe the pervasive body lice that infested their uniforms. The word likely traces back to the Malay word kutu (parasitic louse).
Funny enough, the same sorts of activities that can “give you cooties” can in fact spread cooties (the real ones.) Though mostly due to physical proximity and skin contact.
There are literal slow motion videos on YouTube showing people pressing the correct intended next character and iOS "correcting" it to something else because it predicted(?) that they actually meant something else.
When the first iOS 27 beta came out, it was actually really good for me. But later betas and now the public release ruined it once again. It's unbelievable how bad it is.
There is a lot more interesting stuff to mine from this.
The phrase blocklist for the Chinese lanaguge is a lot larger (surprise, surprise). It about 281 KB for Simplified Chinese, 325 KB for Traditional and 160 KB for Cantonese, against 28 KB for English.
apparently there are `shouldShowTaiwanFlagEmoji()` and `deviceIsChinese()` functions as well but that doesn't have anything to do with this database.
Yup. Something about this sentence reeks of Opus 5.5: "The one exception is the n-word: it is in the list in English and several other languages, and I'm not writing it out."
There is something very funny to me about how that word (I’d write it directly if I knew I could without getting auto-dead) has somehow been deemed the most offensive word of all time. What makes it different from any other slur based upon identity?
Yeah, what is it with the fucking 4-word sentences (subject, verb, object) which AI use that aggravate me. Always so short and "punchy". Maybe it's the aggression and breatheless haste...
I can't focus and read this shit. The other article on the frontpage https://blog.nns.ee/2026/11/02/renault-hack/ also used LLM for the work, but I think a human wrote it (or he's using an LLM that's managed to sound human). It's so readable!
Edit: a snippet from the slop:
> There is no public API
> I looked. NSSpellChecker, UITextChecker, NaturalLanguage and UILexicon don't expose vulgarity.
I think the fucking problem is, each sentence is supposed to convey information. Here "I looked." is prominent, but is it information? No, it's fucking useless. Remove it and it doesn't even change the information. Meanwhile readers' brains are trying to process why the author included this sentence, it must have value?
I have a MacBook that was meant to replace my aging Windows and Linux laptops. This article is a microcosm of why I have just stopped using it. An endless stream of mollycoddling, obfuscation and "It's for your own good, child" trash. I can't even bring myself to sell it because I'm ashamed of it.
To be crystal clear, this internal framework and these databases exist as tweaks to make these words be autosuggested to / autocorrected to more often than they would be before the tweak.
Also, not only Windows and Android but even Linux (aspell / hunspell) also have+use “don’t suggest this word in autosuggest” metadata tags in their dictionary databases.
Just in time for America's ongoing discovery that "liberal" and "conservative" are in fact quite similar ideologies. New insults are called for. I propose "progtard" and "reactard".
That's incidental, but I hear it a lot. I started using this screen name nigh-on 30 years ago (as was the slang at the time), but I love that people recognize that.
How do I make iOS let me swipe “fucking” everywhere? fucking
fucking fucking fucking!! After typing it manually two or three times it then swipe completes in this particular text field in this app. I've heard about adding a contact with the words, but that is ridiculous, so I have held out hoping Apple will eventually come to their senses, like a fool.
This doesn't make “fuck” swipeable either… unless I type it once, and keep it in the text field!
Somehow Mao is there. But Pol Pot isn't. I would have guessed they'd go by the number of victims. But I guess "Uncle Joe" is clean and uncontroversial, so someone was confused about history. But yeah, this has that "Cupertino interns typed this list" feel to it.
Why would “trump” be considered an offensive term? My understanding of the database is to prevent you from accidentally typing offensive or words being autocorrected to something offensive. Like if I misspell “helper” I don’t want that to be autocorrected to “hitler”.
It’s surprising to me that “trump” would get the same classification. Given that he’s the current president of the US, I’m sure people intentionally type his name all the time. I don’t want my keyboard to steer me away from writing out “trump”. And typos autocorrecting to Trump don’t seem like a big deal either. Would anyone be offended by the text “don’t forget to bring your Trump to band practice”?
Would it be possible to get versions of the files to see when certain words got added and I imagine certain words got removed? Could be interesting to correlate certain events, personalities, trends, memes, controversies etc with the changes.
It was added in 23 so there's potentially 3 years of updates.
Edits: This comment shows differences between 15.6 and 26.2 https://news.ycombinator.com/item?id=50358223
I'd appreciate if someone could dump the iOS "adult websites" list (which are blocked by default in the UK unless you submit government ID etc.)
Wait, are you saying iOS UK Edition blocks adult websites on the OS level?
Yup! Any webkit web view, and iOS doesn't allow any non-webkit web views (in this jurisdiction at least)
10 replies →
It's not a list, it's a series of ML classifications. Basically, you've got "child mode" locked on. The solution is to not use a UK apple id.
An ML classifier running against the DNS name? Really?
3 replies →
This is about autocorrect learning words you like to use that it should consider potentially intentional use of words it might otherwise think you typed accidentally
It’s adaptive non-censorship
if it's non-censorship then why is it categorizing the words?
Categories are used for training the autocorrect transformer model. It’s context aware, app aware, and adaptive over time. The 2023 wwdc keynote explains it all
Regardless, it’s non-censorship because autocorrect does not prevent you from saying anything
Apple's built-in dictionary flags on my Mac: 26.2, which is newer than the article's 15.6. Apple has changed several things since:
if autocorrect is disabled this will be empty
ah that explains it
The vulgar db doesn't exist at all (nor the ~/Library/Keyboards directory) on my Tahoe install. Weird
It didn't for me either but macOS was trying to write to it and failed so I had Claude fix it: https://gist.github.com/ericpardee/4ca8d9d46d03d3d92b1887f47...
Huh. I wonder what percentage of Macs are just not running this feature because someone in Cupertino failed to create the /Keyboards/ subdirectory
This is interesting… does anyone have data in theirs? It also feels like something they could have put on an intern and it didn’t really gain any traction. Each app would need to “tell” the OS who the receiver is, and the article indicates there is no API to do that. Maybe it just knows, but how could that be?
Why does Apple need to keep these in a separate database? I thought they removed filtering, maybe subject to user age or parental controls, not added a whole mechanism to manage it.
This gave me a good chuckle.
Fun fact, the word “cooties” didn’t originally refer to a fictitious infective agent spoken of by children; it was originally a nickname for body lice.
> Coined during World War I by British and American soldiers in the trenches of Belgium and France to describe the pervasive body lice that infested their uniforms. The word likely traces back to the Malay word kutu (parasitic louse).
Funny enough, the same sorts of activities that can “give you cooties” can in fact spread cooties (the real ones.) Though mostly due to physical proximity and skin contact.
That's a very fun (and interesting) fact indeed!
Apple MacOS contains keylogger
Updates could expand its use, Computer owner unlikely to notice
Does this play a role in why the keyboard on iOS is so fucking enraging? It used to be amazing.
No. That is a separate INCREDIBLY OBNOXIOUS issue.
Can’t believe they don’t fix it year after year.
There are literal slow motion videos on YouTube showing people pressing the correct intended next character and iOS "correcting" it to something else because it predicted(?) that they actually meant something else.
When the first iOS 27 beta came out, it was actually really good for me. But later betas and now the public release ruined it once again. It's unbelievable how bad it is.
1 reply →
It actually seems like it could be related because of a change in the prediction model that places emphasis on this type of "learning"
There is a lot more interesting stuff to mine from this.
The phrase blocklist for the Chinese lanaguge is a lot larger (surprise, surprise). It about 281 KB for Simplified Chinese, 325 KB for Traditional and 160 KB for Cantonese, against 28 KB for English.
apparently there are `shouldShowTaiwanFlagEmoji()` and `deviceIsChinese()` functions as well but that doesn't have anything to do with this database.
Can put all these ethical quandaries behind `shouldActLikeTheEthicalBusinessWeClaimToBe()`.
> I dumped Apple's word list
Claude dumped Apple's word list, and obviously wrote the article too.
But if it gets rogue and hacks some websites then it's suddenly again you, not Claude.
If I use a car to break the 100m dash record, it's the car. If I then run someone over at the finish line, it's me.
This is not some novel legal loophole.
1 reply →
[dead]
[flagged]
Yup. Something about this sentence reeks of Opus 5.5: "The one exception is the n-word: it is in the list in English and several other languages, and I'm not writing it out."
There is something very funny to me about how that word (I’d write it directly if I knew I could without getting auto-dead) has somehow been deemed the most offensive word of all time. What makes it different from any other slur based upon identity?
8 replies →
Yeah, what is it with the fucking 4-word sentences (subject, verb, object) which AI use that aggravate me. Always so short and "punchy". Maybe it's the aggression and breatheless haste...
I can't focus and read this shit. The other article on the frontpage https://blog.nns.ee/2026/11/02/renault-hack/ also used LLM for the work, but I think a human wrote it (or he's using an LLM that's managed to sound human). It's so readable!
Edit: a snippet from the slop:
> There is no public API
> I looked. NSSpellChecker, UITextChecker, NaturalLanguage and UILexicon don't expose vulgarity.
I think the fucking problem is, each sentence is supposed to convey information. Here "I looked." is prominent, but is it information? No, it's fucking useless. Remove it and it doesn't even change the information. Meanwhile readers' brains are trying to process why the author included this sentence, it must have value?
[flagged]
I have a MacBook that was meant to replace my aging Windows and Linux laptops. This article is a microcosm of why I have just stopped using it. An endless stream of mollycoddling, obfuscation and "It's for your own good, child" trash. I can't even bring myself to sell it because I'm ashamed of it.
But ... the point of that vulgar words list is to STOP autocorrecting them when it detects intentional use.
The parent post is a microcosm of why he's actually not using the Macbook - he doesn't understand how it works.
1 reply →
Also you can disable autocorrect. I don't even have this list.
1 reply →
Install Linux on it and carve a big X over the apple logo
I have Asahi on my mac book mini - it turns a nice piece of hardware into a wonderful computer to use :)
There's no X in Wayland
1 reply →
To be crystal clear, this internal framework and these databases exist as tweaks to make these words be autosuggested to / autocorrected to more often than they would be before the tweak.
Also, not only Windows and Android but even Linux (aspell / hunspell) also have+use “don’t suggest this word in autosuggest” metadata tags in their dictionary databases.
Incredible that 'conservatards' is in there but not 'libtard'. I thought the right was always being censored!
I don't think I have ever seen 'conservatards' before today
First time I heard that word too. Which proves the point /s
People are reading too much into this.
2 replies →
That is odd. Hadn’t occurred to me.
libtard and conservatard are both there on when I tested on my Mac 26.2
I think downvoters may not have realized i was being sarcastic with my second sentence.
The author must have bought MAGA edition of the MBP. The AOC edition has 'libtard' in it, surely.
It's the exception that proves the rule
figure everyone knows, but since plenty english second language here — prove used to have a "test if passes" meaning, and proving can fail.
two examples:
proving grounds = testing grounds (people know this isn't proof grounds)
proves the rule = tests the rule (people think this is proof, no, also testing)
Just in time for America's ongoing discovery that "liberal" and "conservative" are in fact quite similar ideologies. New insults are called for. I propose "progtard" and "reactard".
Progcels be seething over reactchads.
2 replies →
Somebody please kill Claude I'm so fucking tired of reading this grating writing. Technology was a mistake
If you find something funnier than Oedipus, I want to hear about it.
'Bacchus', 'Hooters' are also in the lists. 'Pornhub' but not 'OnlyFans'/'OF'.
I'm reminded of how Cartman on South Park used to sneak in terms like 'catamite'. Steganographized archaic terms.
I feel the same way.
That's the funniest comment I read today. Honestly, I feel the same.
The cultural burnout and second-hand AI overdose many of us are suffering. It's a form of mental pollution, and it's getting harder to breathe.
> Every other real person on that list is dead, and nearly all of them are Nazi officials, dictators or terrorists. Then there's Trump
If the shoe fits....
Surely the first HN user to be named after a Brummie drag queen
That's incidental, but I hear it a lot. I started using this screen name nigh-on 30 years ago (as was the slang at the time), but I love that people recognize that.
you must acquit
giving fixing ducking
How do I make iOS let me swipe “fucking” everywhere? fucking
fucking fucking fucking!! After typing it manually two or three times it then swipe completes in this particular text field in this app. I've heard about adding a contact with the words, but that is ridiculous, so I have held out hoping Apple will eventually come to their senses, like a fool.
This doesn't make “fuck” swipeable either… unless I type it once, and keep it in the text field!
Settings->General->Keyboard->Text Replacement
add fucking as a replacement for fucking (yes, both the same)
Stupid, but it works.
Add autocorrect in the settings/keyboard/text replacement to change duck and ducking etc back to their original beauty
Somehow Mao is there. But Pol Pot isn't. I would have guessed they'd go by the number of victims. But I guess "Uncle Joe" is clean and uncontroversial, so someone was confused about history. But yeah, this has that "Cupertino interns typed this list" feel to it.
[flagged]
Why would “trump” be considered an offensive term? My understanding of the database is to prevent you from accidentally typing offensive or words being autocorrected to something offensive. Like if I misspell “helper” I don’t want that to be autocorrected to “hitler”.
It’s surprising to me that “trump” would get the same classification. Given that he’s the current president of the US, I’m sure people intentionally type his name all the time. I don’t want my keyboard to steer me away from writing out “trump”. And typos autocorrecting to Trump don’t seem like a big deal either. Would anyone be offended by the text “don’t forget to bring your Trump to band practice”?
A reasonable explanation is proposed in the article
just read the article and trump (unless its the word for fart as in the uk) is marked as sensitive.
Here in the UK it means fart
Same on the other side of the pond
They're really passionate about Euchre (in a negative way) over in Cupertino
Trump is marked as sensitive, not offensive; along with words like "God", "Jews", and "gun"
> Would anyone be offended by the text “don’t forget to bring your Trump to band practice”?
Trump could be offended. It is obviously there to avoid offending Trump.
And it is not considered "offensive" but "sensitive".
Right on Trump and wrong on Putin.