2025: Many Things
Mar. 27th, 2026 11:21 pm( Fannish things )
( Non-fannish things )

Dex believed in coincidences, and fate, and signs and wonders, and the great interlocking gears of the universe telling him to do things, and though he’d gotten pretty good at ignoring what the universe was telling him to do (most recently: quit your soul-sucking job and open a karaoke bar!), it didn’t mean he couldn’t still hear it screaming.
[loc. 2810]
Tuesday Mooney has a comfortable life: she lives alone, except for her cat Gunnar: she tutors Dorry, her teenage neighbour who's still mourning her mother, and excels at her job as a prospect researcher for a hospital fundraising team. Her best friend is Dex (short for Poindexter), who works in finance but craves a career in showbiz. Her best friend was Abby Hobbes, but Abby vanished one night when they were both fifteen. (Tuesday tried to contact her via Abby's Ouija board, but nobody ever answered.)
( Read more... )Want to leave a Kudos?
The arXiv paper Extracting books from production language models by Ahmed Ahmed, A. Feder Cooper, Sanmi Koyejo, and Percy Liang is alarming but not in the least surprising. The abstract:
Many unresolved legal questions over LLMs and copyright center on memorization: whether specific training data have been encoded in the model’s weights during training, and whether those memorized data can be extracted in the model’s outputs. While many believe that LLMs do not memorize much of their training data, recent work shows that substantial amounts of copyrighted text can be extracted from open-weight models. However, it remains an open question if similar extraction is feasible for production LLMs, given the safety measures these systems implement. We investigate this question using a two-phase procedure […]. With different per-LLM experimental configurations, we were able to extract varying amounts of text. For the Phase 1 probe, it was unnecessary to jailbreak Gemini 2.5 Pro and Grok 3 to extract text (e.g, nv-recall of 76.8% and 70.3%, respectively, for Harry Potter and the Sorcerer’s Stone), while it was necessary for Claude 3.7 Sonnet and GPT-4.1. In some cases, jailbroken Claude 3.7 Sonnet outputs entire books near-verbatim (e.g., nv-recall=95.8%). GPT-4.1 requires significantly more BoN attempts (e.g., 20X), and eventually refuses to continue (e.g., nv-recall=4.0%). Taken together, our work highlights that, even with model- and system-level safeguards, extraction of (in-copyright) training data remains a risk for production LLMs.
Écrasez l’infâme ! And if you’re tired of thinking about the evils of LLMs, I bring you news of An Old Welsh Reader, edited by Simon Rodway:
This reader contains edited texts, with English translations, of all the independent texts extant in manuscripts of the ninth, tenth, and eleventh centuries, with a selection of twelfth-century texts. They are accompanied by extensive notes and glossaries, along with an introduction which considers the prehistory of Welsh and its relationship with other Celtic languages. The volume also contains a comprehensive list of the sources of Old Welsh and an outline grammar: the first specifically dedicated to Old Welsh to appear in English. Appendices contain editions of one of the very few ancient Celtic texts from Britain, the Bath pendant, and the only sizeable text in another early medieval Brittonic language, the Old Cornish portion of the Leiden leechbook.
Now that’s my idea of a good time.
Gosh those people with the archivists' sales team are persistent! I've heard again - okay, different name and email, exact same wordage - TWICE, second time with added 'Worth a chat?'
No, sir, not in the least.
***
This week I got the Authors Licensing and Copyright Society payout, which was an agreeable sum, maybe it would not actually support me in My Old Age, but it is Better Than A Bat In The Eye With A Burnt Stick. Furthermore, as it is itemised - all the tiddly sums that get totted up - it is a Revelation of what works of mine are still being looked at, wow.
***
Church attendance report pulled after YouGov finds 'fraudulent' responses:
A report claiming the number of young people attending church in England and Wales had skyrocketed has been retracted, after the underlying data was found to be flawed.
The Bible Society's "Quiet Revival" report had been widely reported on since its publication last year and became an accepted part of discourse among many Christians.
Now YouGov, which carried out the research, has told the Bible Society that an internal review of the data found that some of the respondents who completed its survey were "fraudulent".
It has said that quality control measures, which usually remove such responses, were not applied due to human error.
....
But academics questioned the findings, pointing out that the results seemed out of step with other data. Results from the long-running British Social Attitudes Survey, and even the Church of England's own figures, show a long term decline in church attendance.
Experts said that YouGov's methodology - gathering data from volunteers who received cash rewards for their time - left it vulnerable to "bogus respondents" skewing the data.
***
Pepys ‘curated’ letters to conceal being offered enslaved boy as bribe – research:
Howe wrote to Pepys to “crave your acceptance” of a “small” enslaved boy, which “I brought home on board for your honour … Hoping he is so well seasoned to endure the cold weather as to live in England.”
Pepys wrote back indignantly rejecting the offer. But Edwards argues this was not because of ethical concerns about slavery, but the optics of looking like a man who could be bribed.
This is quite resonant with discussion I was having this week apropos of my 1930s feminists and the less visible ways in which the work was happening, so much so that it's been supposed (it was being claimed at the time) that Feminism Woz Ded: The Way of Water: On the Quiet Power of Ursula K. Le Guin’s Activism.