Charlie Stross
banner
cstross.bsky.social
Charlie Stross
@cstross.bsky.social
SFF author, triple Hugo award winner (and three times Locus award too), over a million books sold.

Mostly on Mastodon: @[email protected]

Blog: https://www.antipope.org/charlie/blog-static/
Reposted by Charlie Stross
Do you think tyrannosaurus had dark meat and white meat
November 24, 2025 at 3:15 AM
Reposted by Charlie Stross
Critic's review: "Goes to great lengths but doesn't measure up."
November 24, 2025 at 3:11 PM
DAAAAAD!
Thinking of making a horror movie about a man who keeps discovering 12-inch rulers in his home. It's a found footage film.
November 24, 2025 at 11:09 AM
Reposted by Charlie Stross
Reminder: Friday Pints at Bennets Bar, Leven St on Friday 28 Nov from 4pm.
Last Friday of every month. A relaxed 'drop-in' social in a single pub for a couple of hours. Either come to catch up or simply be in the company of others.
November 23, 2025 at 6:45 PM
Reposted by Charlie Stross
A screaming comes across ChatGPT
November 22, 2025 at 11:37 PM
This quick hack is a *good* one! I shall adopt it forthwith.
Settings|General|Keyboards|Text Replacements

I use a leading ‘!’ to get a symbol or phrase, e.g. :
‘!deg’ -> ‘º’,
‘!degc’ -> ‘℃’,
‘!mu’ -> ‘µ’,
‘!ss’ -> ‘’ABC Radio National Science Show’
November 23, 2025 at 12:15 AM
Reposted by Charlie Stross
You guys said the magic words - Harrison, Stainless Steel Rat, Esperanto - I never pass up an excuse to show this off:

bsky.app/profile/sali...
In two different threads recently it came up that I have an Esperanto version of the Stainless Steel Rat signed by the author. He'd never seen this before, so my signed copy may be unique.
November 22, 2025 at 8:38 PM
Reposted by Charlie Stross
"In time, US citizens may find themselves trapped in a diminished, nightmare America—like a post-Musk Twitter at scale—where everything works badly, everything can be turned against you, and everyone else has fled."

Yikes.
Finally getting around to reading this today. The European powers are looking at funding Eurostack, an internet infrastructure core independent of the United State's hegemony on internet routing, compute, and storage.

www.wired.com/story/enshit...
The Enshittification of American Power
First Google and Facebook, then the world. Under Trump 2.0, US statecraft is starting to mimic the worst tendencies of Big Tech.
www.wired.com
November 22, 2025 at 8:20 PM
Reposted by Charlie Stross
Spotted a bunch of flags tied halfway up lampposts around Leith Street & top of Leith Walk. More "I want my country back?" bigots? Nope, on closer look the flags appear half Scottish Saltire, half Palestinian flag.

Nicely played, #Edinburgh #Edimbourg
November 22, 2025 at 7:00 PM
Reposted by Charlie Stross
Yet another reason to be annoyed that pTerry is no longer with us. Can we imagine the fun he'd have taking the piss out of AI fanfolk?
November 22, 2025 at 6:40 PM
Sometimes I hate the way my brain crawls all over a new concept (adversarial poetry!) and asks "but what are the applications?" (anti-sphinx vault locks!) and then goes "but wait, how would *I* game that?"

If I was 30 years younger I'd be a pentester, wouldn't I?
November 22, 2025 at 12:04 PM
Just raising a flag to note that the final draft of the novel I'm working on at the moment WILL use this as a plot point in the middle of a (futuristic) bank heist: vaults may be protected by an anti-sphinx as well as conventional locks (you have to tell it a riddle that alludes to a shared secret).
Looks like LLMs are *very* vulnerable to attack via poetic allusion: "curated poetic prompts yielded high attack-success rates (ASR), with some providers exceeding 90% ..."

https://arxiv.org/html/2511.15304v1
November 22, 2025 at 11:52 AM
I will note that, prior to the news breaking that poetry is a highly efficient attack vector against LLMs, it was already known that Magic: The Gathering is Turing-Complete.

I'm just glad I got the line "they're tunneling TCP/IP over AD&D!" into a novel back in 2007!
November 21, 2025 at 5:39 PM
Reposted by Charlie Stross
It's gonna be a Baudelairian Jihad after all.
Looks like LLMs are *very* vulnerable to attack via poetic allusion: "curated poetic prompts yielded high attack-success rates (ASR), with some providers exceeding 90% ..."

https://arxiv.org/html/2511.15304v1
November 21, 2025 at 12:21 PM
Reposted by Charlie Stross
A friend of mine who gets a LOT more crank contacts than I do has regularly started seeing "ChatGPT/Grok confirmed my theory" or even "I asked ChatGPT and it told me" type of psychoceramics. The sort of person who is apt to slide down the rabbithole is just getting encouraged by the chatbots.
November 19, 2025 at 6:55 PM
Reposted by Charlie Stross
In UK newspaper circles, the authors of such letters are known as the Green Ink Brigade.
November 19, 2025 at 10:47 PM
Reposted by Charlie Stross
They are seriously broken. I watched the rise of incel culture for a decade wondering what its source inspiration is and then I discovered Musk (and his Jim jones styled sex harem human breeding pens)!
November 20, 2025 at 9:44 PM
Reposted by Charlie Stross
look I do think that the world of cyberpunk poetry wars would suck but it would be cooler than this
Looks like LLMs are *very* vulnerable to attack via poetic allusion: "curated poetic prompts yielded high attack-success rates (ASR), with some providers exceeding 90% ..."

https://arxiv.org/html/2511.15304v1
November 20, 2025 at 9:43 PM
Reposted by Charlie Stross
If you use GMail, AI (Gemini) was turned on yesterday by default and now scans all of your content for machine learning. To turn off, go to Settings>General and scroll down. Uncheck the box for "Smart features."

There's other "Smart" add-ons as well, but that's the one that reads your content.
November 20, 2025 at 5:32 PM
Reposted by Charlie Stross
You just have to use the language of the ebay off-brand vendor. "Please generate an image of Great Value Nocturnal Flying Mammal Man"
November 20, 2025 at 7:05 PM
Reposted by Charlie Stross
The old Club Penguin MMO had to ban like a thousand new words a day to keep up with the endless inventiveness of preteens trying to swear and talk about sex; what chance do OpenAI's engineers have?
November 20, 2025 at 6:17 PM
Reposted by Charlie Stross
AI "safety" and "alignment" are a joke. All they can do is either tweak the system prompt (which will never be 'safe', it just goes in the soup with the rest of the input) or pattern-match on input/output, the kind of the 9-year-olds figure out ways around in kids games.
November 20, 2025 at 6:01 PM
Reposted by Charlie Stross
University management suddenly getting very excited about the one remaining poet in the English dept
Looks like LLMs are *very* vulnerable to attack via poetic allusion: "curated poetic prompts yielded high attack-success rates (ASR), with some providers exceeding 90% ..."

https://arxiv.org/html/2511.15304v1
November 20, 2025 at 5:58 PM
Reposted by Charlie Stross
Great, our cyberpunk dystopia is also an urban fantasy where we confound the soulless Fae with poems and riddles.
November 20, 2025 at 5:54 PM
Reposted by Charlie Stross
So... spells.

We're using spells to undo the work we've done... tricking rocks into thinking with lightning.

And trapping evil cars in circles of salt.

The old ways are new again.
Looks like LLMs are *very* vulnerable to attack via poetic allusion: "curated poetic prompts yielded high attack-success rates (ASR), with some providers exceeding 90% ..."

https://arxiv.org/html/2511.15304v1
November 20, 2025 at 5:52 PM