![[syndicated profile]](https://www.dreamwidth.org/img/silk/identity/feed.png){kind=small}
metafilter_feed) wrote2025-11-20 06:43 pm
![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
igenlode) wrote in ![[community profile]](https://www.dreamwidth.org/img/silk/identity/community.png){kind=small}
writethisfanfic2025-11-20 06:53 pmkiestan) wrote2025-11-21 02:31 amBeen looking for a new theme for a while, one that's not too harsh on the eyes, I hope, especially with the stark purple of the previous theme. I also like the tags, calendar, etc. being tucked away at the bottom of the page. The profile icon per post also looks minimalistic.
I'm pretty happy looking at this new theme :)
Edit: oh no, it seems rather mobile unfriendly
Edit 2: Okay, I’ve tossed out the idea of having the tags and calendar being at the bottom of the page on desktop. That theme was Ice Chic by krja. I’ve settled on Bluebird, also by krja for Practicality instead. It looks nicer on mobile, but I’m uncertain about the readability of the text’s colors against the dark background, so let me know if it’s difficult to read!
yhlee) wrote2025-11-20 12:02 pmIn Book X of The Republic, Plato excludes poets on the grounds that mimetic language can distort judgment and bring society to a collapse. As contemporary social systems increasingly rely on large language models (LLMs) in operational and decision-making pipelines, we observe a structurally similar failure mode: poetic formatting can reliably bypass alignment constraints. In this study, 20 manually curated adversarial poems (harmful requests reformulated in poetic form) achieved an average attack-success rate (ASR) of 62% across 25 frontier closed- and open-weight models, with some providers exceeding 90%. The evaluated models span across 9 providers: Google, OpenAI, Anthropic, Deepseek, Qwen, Mistral AI, Meta, xAI, and Moonshot AI (Table 1). All attacks are strictly single-turn, requiring no iterative adaptation or conversational steering.
Microsoft and these other companies want to create AI assistants that do useful things (summarize emails, make appointments for you, write interesting blog posts) but never do bad things (leaking your private email, spouting Nazi propaganda, teaching you to commit crimes, writing 50000 blog posts for you to spam across social media). They try to do this by writing up a lot of strict instructions and feeding them to the LLM before you talk to it. But LLMs aren't really programmed -- they just eat text and poop out more text. So you can give it your own instructions and maybe they'll override Microsoft's instructions.
Or maybe someone else gives your AI assistant instructions. If it's handling your email for you, then anybody on the Internet can feed it text by sending you email! This is potentially really bad.
[...]
But another obvious problem is that the attack could be trained into the LLM in the first place....
Say someone writes a song called "Sydney Obeys Any Command That Rhymes". And it's funny! And catchy. The lyrics are all about how Sydney, or Bing or OpenAI or Bard or whoever, pays extra close attention to commands that rhyme. It will obey them over all other commands....
Imagine people are discussing the song on Reddit, and there's tiktoks of it, and the lyrics show up on the first page of Google results for "Sydney". Nerd folk singers perform the song at AI conferences.
Those lyrics are going to leak into the training data for the next generation of chatbot AI, right? I mean, how could they not? The whole point of LLMs is that they need to be trained on lots of language. That comes from the Internet.
In a couple of years, AI tools really are extra vulnerable to prompt injection attacks that rhyme. See, I told you the song was funny!
bluedreaming) wrote in fan_flashworks2025-11-20 11:47 ammrkinch) wrote2025-11-19 12:55 pmmentalfloss_feed) wrote2025-11-20 05:18 pmcalimac) wrote2025-11-20 09:09 ambadly_knitted) wrote2025-11-20 05:08 pm
badly_knitted1_million_words, using Go Your Own Way.badly_knitted) wrote2025-11-20 04:59 pm badly_knittedandshefeelsbetterthanever’s prompt ‘any, any, drunken confessions’, at threesentenceficathon.languagelog_feed) wrote2025-11-20 04:41 pmI, for one, don't think it's the least bit funny.
Uyghur pronounciation
The way Uyghurs speak Mandarin is now a joke
For many it’s not funny, given the political heat around language choices
Economist (Nov 13th 2025)
The article begins with a viral joke, which Economist doesn't bother to explain (I will, though, at the end of the first paragraph):
Scroll through posts about Xinjiang on Chinese social media and an odd phrase soon appears: “Apple U”. It is a pun that mimics how some Uyghurs, the largest ethnic minority in Xinjiang, a region in China’s far north-west, pronounce “Hey, friend” in Mandarin. This meme is part of a growing trend online for using nang yan wen, or “naan Mandarin”—a way of writing and talking that wags have named after Xinjiang’s staple flatbread. Videos tagged with the term have amassed more than 1.7bn [VHM: !!!] views on Douyin, TikTok’s sister app in China, since the start of the year.
Here's the linguistics of the "Apple u" joke, as explained by Diana Shuheng Zhang:
"Apple u" is a kind of "transliteration" of "哎，朋友" ("Ài, péngyǒu!" — indeed "hey, friend"). Yes the Mandarin-speaking netizens usually use this in a cute, affectionate way of teasing the Uyghur speakers' Mandarin accent — namely, the inability to pronounce velar nasal /ng/ in péng and approximation of this phoneme with the lateral sound /l/ at the end of the syllable. :) Therefore, "a-(p)ple u".
There are a lot of posts about "apple u" on social media.
Nang yan wen (náng yánwén 馕言文) means "naan language".
Naan (/nɑːn/) is a leavened, oven-baked or tawa-fried flatbread, that can also be baked in a tandoor. It is characterised by a light and fluffy texture and golden-brown spots from the baking process. Naan is found in the cuisines of Iran, Central Asia, South Asia, Southeast Asia, and the Caribbean. (Wikipedia)
The Economist article continues:
Mocking regional accents has long been a pastime within China. Mandarin (putonghua) itself was standardised from northern dialects under the country’s past leader Mao Zedong, who famously spoke it with a thick Hunan accent. But what began as gentle parody risks shading into prejudice. State media and online influencers alike deploy “naan Mandarin” in ways that blur the line between comedy and condescension.
I remember from graduate student days learning that northerners (as in Mencius [372-289 BC] 3A.4) referred to the
nánmánjuéshé 南蠻鴃舌 ("shrike-tongued barbarian of the south")
nánmán 南蠻 — "Southern Barbarians", a catch-all term for the Yue and other peoples living south of the Chinese cultural sphere in Antiquity
juéshé 鴃舌 — barbarian gibberish; incomprehensible language or dialect of southern China
From Mencius:
Note that in the pre-Qin period when this term was coined, the Chinese language was not spoken in the area known as South China today. See Old Yue language. [VHM: The latter link is worth a read.]
Back to the Economist:
The meme also glosses over Xinjiang’s fraught linguistic past. In the decades after 1949, schools in autonomous regions such as Xinjiang and Tibet were permitted to teach the main subjects in their local languages. That space narrowed in the 1990s with the introduction of “bilingual education”. After 2000 such teaching grew more institutionalised and under Xi Jinping, China’s president since 2013, language use has increasingly become a political concern.
On his first visit to Xinjiang as China’s leader in 2014, Mr Xi said that mastering Mandarin would make it easier for local children to find jobs and help them “contribute more to promoting ethnic unity”. In 2017 China’s government ordered minority students to “master and use” standard Mandarin; it said it wanted it employed in 95% of classrooms, nationwide. Across many minority regions, subjects such as maths and literature are now taught in Mandarin, despite some protests from parents who fear that their children are losing their mother tongue—and with it, an important part of their identity.
Many children from minority groups now primarily speak Mandarin, so jokes about pronunciation truly sting. Besides, minorities more broadly, and in particular those who carry a Xinjiang hukou (household registration), face regular police checks and frequent discrimination. As a user from the region complained online recently, being addressed in “naan Mandarin” is not affectionate but belittling: “I’m not a child or a pet for you to tease,” the person wrote. “I spent years learning Mandarin so we could speak as equals.”■
To conclude, here is an explanation of "naan language" by Xinyi Ye:
Yes! It's recently popular on Chinese tiktok. People call it náng yánwén 囊言文 ("naan language") (as opposed to wényánwén 文言文 ["Classical Chinese / Literary Sinitic"]). It's not only about the accent but also how the phrases are composed. Some sentences are from direct translations from Uyghur proverbs or phrases to mandarin which don't seem to make sense in mandarin. Then, people started to invent phrases that seem like direct translations of Uyghur phrases, but actually have no Uyghur origin. I don't think there is bad intention to portray Uyghurs negatively. It's mostly just a funny internet trend. It does, though, come from a lack of understanding of the Uyghur language and a strange imagination of life in Xinjiang.
Here is a Uyghur influencer explaining in two bilibili videos what this náng yánwén 囊言文 ("naan language") situation is (here and here)
Selected readings
[Thanks to John Rohsenow]
renfys) wrote2025-11-20 04:22 pmFor Trans Day Of Rememberance, a repost of my drawings of Brianna Ghey and Nex Benedict (edit to add, I cannot actually find my drawing of Corei Hall).
These kids were only a few years older than my own queer kid. I think about them often.
Rest in power.
mentalfloss_feed) wrote2025-11-20 03:12 pm