iiinova — where your digital self keeps evolving

Reposted by MathiasTCK.bsky.social

Niall Machin @niallmachin1.bsky.social · 1h

Used up all its missiles and now getting rid of its generals. Really are rolling out the red carpet for #Putin great work Agent #Krasnov

172

Reposted by MathiasTCK.bsky.social

A.C. McKenzie @acmckenzie.bsky.social · 18/01/2026

Gee. Shocker. #Krasnov leaks critical info to his Daddy, Putin. I'm so shocked. 🤦‍♀️

264117385679

Reposted by MathiasTCK.bsky.social

CoastLife @coastlifetv.bsky.social · 09/09/2026

AI vs. Mathematicians: Part 2 A big fight over a million dollar math problem, two people that got close to solving it, and OpenAI coming in late to finish it after the original researchers were using one of their products. #OpenAI #NavierStokes #Buckmaster #Alpoge

021

Reposted by MathiasTCK.bsky.social

macfranc @macfranc.poliversity.it.ap.brid.gy · 09/09/2026

# **OpenAI avrebbe giocato sporco sul problema della dimostrazione completa del problema di esistenza e fluidità di Navier–Stokes** Il professore di matematica della New York University Tristan #Buckmaster ha annunciato martedì tre dimostrazioni con una scoperta preliminare su uno dei […]

poliversity.it

Original post on poliversity.it

012

Reposted by MathiasTCK.bsky.social

David Sainez @davidsainez.com · 06/02/2026

a man whispering into president bush's ear: woo woo woo

1614

Reposted by MathiasTCK.bsky.social

Kat Abughazaleh @katmabu.bsky.social · 2h

The KAPOW Handbook is the political playbook we wish we had: a guide to turning campaign resources into lasting community resources. 💥 Download the free handbook at kapow.works/handbook

kapow.works

GET THE HANDBOOK — KAPOW

112519

Reposted by MathiasTCK.bsky.social

Aaron Rupar @atrupar.com · 2h

Trump: "Super intelligence -- we've changed the name officially, by the way. I signed the document. We don't call it artificial intelligence. Because it's not artificial. It's real. It's a terrible world. It's amazing it loved on so long, actually."

12731952

Reposted by MathiasTCK.bsky.social

Aaron Rupar @atrupar.com · 2h

alert the press -- Trump is conked out in the Oval Office again

2691831455

Reposted by MathiasTCK.bsky.social

The Tennessee Holler @thetnholler.bsky.social · 3h

“Witch Hunt” Tennessee AG Skrmetti is going after the records of Planned Parenthood and other women’s health orgs — as an appeals court blocks a portion of the TN GOP abortion travel ban. #ForcedBirthState www.wate.com/news/planned...

1321691

Reposted by MathiasTCK.bsky.social

Aaron Rupar @atrupar.com · 3h

Hegseth: "Today, at my direction, I'm establishing the Office of Religious Affairs. It will report straight to the top -- a direct line to the secretary ... as Psalm 33:12 extolls us, blessed is the nation in which God is the Lord"

525767213

Reposted by MathiasTCK.bsky.social

Aaron Rupar @atrupar.com · 4h

Hegseth: "It's worth saying again so the fake news understands it. We are no longer the woke department or the weak department. Simple translation of that -- no fatties no trannies no beardos no weirdos no wimps no radicals. Just warriors."

18992022484

Reposted by MathiasTCK.bsky.social

Aaron Rupar @atrupar.com · 3h

Sen. Jon Husted nods along in agreement as Scott Jennings says "the president has won the war in Iran" and "Trump was right about this" while blaming Democrats for high fuel prices

76441111

Reposted by MathiasTCK.bsky.social

Headquarters @headquartersnews.bsky.social · 3h

Pete Hegseth touts the US military's "secret weapon" revealing that they have a "patriotic billionaire" making deals to spend the $1,500,000,000,000 military budget.

113124

Reposted by MathiasTCK.bsky.social

Paul fulbrook @paulfulbrook74.bsky.social · 3h

114431

Reposted by MathiasTCK.bsky.social

Aaron Rupar @atrupar.com · 3h

JD Vance suggests Talarico is a pedo: "You gotta wonder -- what really motivates a man who spends so much time talking about 11-year-old drag queens?"

478699121

Reposted by MathiasTCK.bsky.social

By Cory Doctorow (GPG 0xBF3D9110957E5F4C) @doctorow.pluralistic.net · 4h

Unlike most trolls, Wendall has been gifted with a remarkably beautiful singing voice (Terry Dykstra, for AD&D adventure “Song of the Fens” by J Bradley Schell, Dungeon 40, TSR, March/April 1993) oldschoolfrp.tumblr.com/post/8291935...

1132

Reposted by MathiasTCK.bsky.social

A. Feder Cooper @afedercooper.bsky.social · 6h

Working with @marklemley.bsky.social et al. has been a career highlight. It's a privilege to have our papers referenced in research, litigation, policy, and the press. I'm grateful others acknowledge our work as setting the standard for rigorous empirical research on copyright and generative AI.

072

Reposted by MathiasTCK.bsky.social

A. Feder Cooper @afedercooper.bsky.social · 6h

We make very carefully scoped claims, supported by an extensive set of baselines on non-training data. Without baselines, it's not possible to rigorously separate a model's ability to generate fluent text from extraction of training data. (We see no false positives for non-training books.)

We can measure extraction coverage: the percentage of a book that is memorized in the model by seeing how much text is extractable (with respect to 50-token suffixes) out of the whole book. When we run extraction with larger prompts (longer prefixes of the 50-token suffix), this exposes more underlying memorization (the discoverability phenomenon, see Carlini et al. 2023). Here, the plot shows that extraction coverage goes up (to a point) as prefix length increases for books in the training data. For the shown control book (which is non-training data), there is no observed extraction at all prefix lengths. This supports the validity of our measurement procedure.

133

Reposted by MathiasTCK.bsky.social

A. Feder Cooper @afedercooper.bsky.social · 6h

4. The implications for copyright don't unambiguously favor either side in litigation: (a) LLMs could be derivative works of books they memorize; (b) it may be practical to generate infringing copies in some cases; and, (c) class actions are ill-suited for copyright claims about memorization.

132

Reposted by MathiasTCK.bsky.social

A. Feder Cooper @afedercooper.bsky.social · 6h

3. In some cases, memorization is so extensive that one can deterministically(!) extract a near-pristine copy of a whole book using the book's first few words as an initial prompt. Here's a portion of the diff between the real (British) Harry Potter and the one we generated with Llama 3.1 70B.

A portion of the diff between the ground-truth Harry Potter and the Sorcerer's Stone and the book we generated. There are very minimal differences. For instance, our reference copy of the book from Books3 is the British version of the book (e.g., writes "Mum"), and we generated the American version (e.g., write "Mom").

142

Reposted by MathiasTCK.bsky.social

A. Feder Cooper @afedercooper.bsky.social · 6h

2. There are notable exceptions, e.g., Llama 3.1 70B entirely memorizes some books and enormous parts of others (e.g., Harry Potter, 1984). You can see this in the heatmaps for 1984 and Harry Potter and the Sorcerer's Stone for Llama 3.1 70B: there are high extraction probabilities everywhere.

Heatmaps showing memorization (via extraction probability) for 3 books and 5 models, where these books were included in the training data. WIth our measurement procedure, we observe very different amounts of memorization across books and models.

242

Reposted by MathiasTCK.bsky.social

A. Feder Cooper @afedercooper.bsky.social · 6h

1. Memorization varies both by model and book. With respect to our specific measurement methodology, we find that most LLMs don't memorize most books. In the heatmaps, white means not memorized; blue indicates memorization (with darker blue indicating higher extraction probability) However...

For George Orwell's 1984, a scatterplot showing where extracted sequences are located within the book. The x-axis shows location of the sequence (50-token suffix) is, the y-axis shows its extraction probability. We condense the same information into a heatmap, which is easier for making comparisons across books and models. Darker blue means higher extraction probability; white means we don't observe extraction signal.

142

Reposted by MathiasTCK.bsky.social

A. Feder Cooper @afedercooper.bsky.social · 6h

With @marklemley.bsky.social and others, we developed a novel and efficient approach for measuring verbatim memorization in LLMs (via extraction in outputs). There are a lot of findings and takeaways, so I'll only cover a few of them (briefly):

294

Reposted by MathiasTCK.bsky.social

A. Feder Cooper @afedercooper.bsky.social · 6h

Our large-scale study of memorization of books in open-weight LLMs (e.g., Llama, Qwen) will appear at the 2026 Conference on Language Modeling as an oral. We made a website for exploring our results on 200 books and 14 models: books-memorization.github.io

books-memorization.github.io

How much do open-weight LLMs memorize specific books?

Open-weight LLMs memorize books far more than previously believed. Memorization varies by model family, model size, and book. In extreme cases, entire books are memorized, and we can generate them eff...

24827

Reposted by MathiasTCK.bsky.social

One Does Not Simply Re-Tweet a Meme @evilmagnus.bsky.social · 5h

This is an amazing study that does a great job of illustrating how memory works in LLMs. Obviously can't run these tests against the closed-source models! They put additional filters in front to prevent/obfuscate extraction for some reason *cough* NYT *cough*

1299

Reposted by MathiasTCK.bsky.social

Raphael Satter @raphae.li · 25/09/2026

Scoop: ShinyHunters say they have medical and psychiatric records of FBI staff. @ajvicens.bsky.social and I have gone through and partially authenticated the sample data they’ve circulated. www.reuters.com/world/shinyh...

reuters.com

EXCLUSIVE: ShinyHunters hackers say they stole psychiatric and medical records of FBI staff

FBI personnel data recently stolen by the ShinyHunters hackers includes sensitive psychiatric and medical evaluation records, according to the hackers and documents reviewed by Reuters.

917177

Reposted by MathiasTCK.bsky.social

WIRED @wired.com · 10/09/2025

Everyone’s talking about hypochlorous acid sprays. Here’s how to use it without sabotaging your skin-care routine.

wrd.cm

Need a Pick-Me-Up? Try Spraying Your Face With Hypochlorous Acid

Everyone’s talking about hypochlorous acid sprays. Here’s how to use it without sabotaging your skin-care routine.

10213

Reposted by MathiasTCK.bsky.social

Rian Nejar 🦋 @riannejar.bsky.social · 16/04/2025

A natural disinfectant safe for surfaces and skin [via Scientific American] 🧪⚗️❌🦠⚕️👀 "it’s 100 times more effective than the active ingredient in bleach [], and yet it’s incredibly safe." www.scientificamerican.com/podcast/epis... #hypochlorous #acid #natural #disinfectant #synthetic #salt #water

scientificamerican.com

How This Natural Disinfectant Finally Made It to Market

Hypochlorous acid has a lot of buzz in the beauty industry, but this nontoxic disinfectant has many possible uses.

1125

Reposted by MathiasTCK.bsky.social

Don Moynihan @donmoyn.bsky.social · 9h

CMU Miami (for that is the name of their new campus) sounds like a police procedural

61039

Reposted by MathiasTCK.bsky.social

Aaron Rupar @atrupar.com · 7h

Blanche defends appearing at Trump rallies 👇 Q: Wherever you go, you are the attorney general-- BLANCHE: I don't agree w/that. We are allowed to do things in our individual capacity. There have been AGs who were blatantly partisan and never spoke at a single rally. Is that right? Of course not!

137604162

Reposted by MathiasTCK.bsky.social

Gordon McAlpin @gmcalpin.bsky.social · 8h

well, it's not "pure" regurgitation. it's copying and combining (based on other patterns it has been taught or identified, but in different contexts). and, remember, lots and lots brute force. that's why it needs all the energy.

231

Reposted by MathiasTCK.bsky.social

Aaron Rupar @atrupar.com · 7h

Q: You have appeared at political rallies, & it has long been the norm that the AG not do that. Are you blurring the line by you being a political actor and yet expecting something different from judges? BLANCHE: If Democrats want me to come and speak and their rally, they can just give me a call

146634158

Reposted by MathiasTCK.bsky.social

Aaron Rupar @atrupar.com · 8h

Klobuchar: "You got this guy who clearly thinks he's got this gotcha moment. And I am sitting there listening because I was the next one up, and starting to think to myself, it's not really major sleuth here, 'I just don't think Jack Smith would lie under oath that he hadn't been at the game.'"

671945376

Reposted by MathiasTCK.bsky.social

The Tennessee Holler @thetnholler.bsky.social · 8h

Ummmm…

38477190

Reposted by MathiasTCK.bsky.social

The Tennessee Holler @thetnholler.bsky.social · 8h

😂😂😂

231124237

Reposted by MathiasTCK.bsky.social

The Lincoln Project @lincolnproject.us · 9h

Ken Paxton just got caught admitting that appearing at Trump's event hurt his numbers, and he's not the only one. Republicans are picking Trump over America and it's backfiring on them as November nears.

41559114

Reposted by MathiasTCK.bsky.social

Erin Reed @erininthemorning.com · 9h

A new poll from The 19th News shows faltering Republican support for gender-affirming care bans for trans youth. It also shows that anti-trans ads might be backfiring, with respondents increasingly saying Republicans are talking about trans people too much. Subscribe to support our journalism.

erininthemorning.com

New Poll From The 19th News Shows Public Turning Against Republican Anti-Trans Healthcare Bans

The poll even shows faltering Republican support for such bans.

21753192