Paris court blocks auction of earliest-known calculator
Ukrainian teen saboteurs recruited on Telegram to attack their own country
well, now we have a study on this attack mechanism...
"Adversarial Poetry as a Universal Single-Turn Jailbreak Mechanism in Large Language Models"
(many authors)
In 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.
By way of Zarf (Andrew Plotkin), who earlier noted (2023):
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!
ClaireBell
( Read more... )
For Leonard, Darko, and Burton Watson
by Ursula K. Le Guin
A black and white cat
on May grass waves his tail, suns his belly
among wallflowers.
I am reading a Chinese poet
called The Old Man Who Does As He Pleases.
The cat is aware of the writing
of swallows
on the white sky.
We are both old and doing what pleases us
in the garden. Now I am writing
and the cat
is sleeping.
Whose poem is this?
More than 250 arrested in Charlotte as US immigration crackdown escalates
What do we know about the Epstein files?
Scale of one trillion dollars
If Elon Musk achieves certain benchmarks for Tesla over the next decade, he gets a $1 trillion bonus. While unlikely Tesla gets there, a trillion is kind of a lot, especially for one person. But our human brains aren’t great at imagining numbers at that scale. So, for the Washington Post, Alyssa Fowers and Leslie Shapiro scaled a trillion by total U.S. workers in a given job.
I like to think in units of number of Jack in the Box tacos I can buy, but I guess that’s more useful for smaller values. Although less so recently. Thanks, inflation.
It’s crazy that just a few years ago we were looking at how comical Jeff Bezos’ net worth of $172 billion was at the time. Pocket change now.
Tags: Elon Musk, jobs, money, scale, Washington Post

Thursday ✎ Therapy [DW]
Today's theme is therapy. There are dozens of forms of therapy in the world. Everyone has their personal favorites and go-to ways to relax, not to mention experience different levels of effectiveness depending on the therapy. So let's think about therapeutic activities, sensations, and anything that can help a character let go of their troubles for even just a minute. What works? What doesn't? Which part of someone's health and well-being needs some extra care?
Feel free to add specifics to your prompts, like whether you'd prefer a gen fill over something shippy, or if you have a squick or trigger you hope to avoid. Original fiction, fanfiction, and fanfic crossovers are always welcome. ~_^
Just a few rules:
No more than five prompts in a row.
No more than three prompts in the same fandom.
Use the character's full names and the fandom's full name
No spoilers in prompts for a month after airing, or use the spoiler cut option found here. Unfortunately, DW doesn’t have a cut tag, so use your best judgment when it comes to spoilers.
If your fill contains spoilers, warn and leave plenty of space, or use the above-mentioned spoiler cut.
Prompts should be formatted as follows: [Use the character's full names and fandom's full name]
Fandom, Character +/ Character, Prompt
Some examples to get things started...
+ any Middle-earth (Tolkien) fandom, any (+/ any), watching nature return and thrive as they themselves heal
+ Resident Evil (game/CGI 'verse), any playable character +/ any, checking over their weapons and gear/inventory (finding/mixing herbs?) is a calming routine, whether they're near danger or not
+ author's choice, any +/ any (+ any pets!), getting warm and cozy after a while outside in the cold
We are on AO3! If you fill a prompt and post it to AO3, please add it to the Bite Sized Bits of Fic from 2025 collection. See further notes on this option here.
Not feeling any of today’s prompts? You can use LJ’s advanced search options to limit keyword results to only comments in this community. Fret not, DW members; we are working on a way to search through old entries for prompts for you! As of right now, the best way to search for a lonely prompt on DW is to search the community’s archive, which can be found [[HERE]].
While the use of LJ's advanced search and DW’s archive are options, bookmarking the links of prompts you like might work better for searching in the future.
As a friendly reminder about our schedule, Lonely Prompts and sharing completed fills are encouraged on Sundays, while new themes and prompts are posted on Tuesdays and Thursdays. Saturdays are a Free for All day. We'll share our posts on DW and LJ for everyone's convenience. Keep an eye out for notifications!
If you have a Dreamwidth account and would feel more comfortable participating there, please feel free to do so… and spread the word!
tag=therapy