Sign in

NON-TOXICNo ads · No shady algorithms

iiinova is part of the Atmosphere — an open, public network where you own your identity.

Theodora Ward

@theodoraward.wandering.shop.ap.brid.gy
47 followers 8 following 1.2K posts

writer, enthusiast // luddite department, AAUP-NC* *ignominious and disagreeable internet posts bespeak nothing but personal spleen 🌉 bridged from ⁂ wandering.shop/@theodoraward, follow @ap.brid.gy to interact

PostsRepliesMedia
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 1h
The article is paywalled but I got a little dopamine hit from seeing the phrase "high junk rating" this close to "Anthropic"
Anthropic
Anthropic may not secure investment-grade rating
after IPO, Blackstone says

Private capital giant says Al group is expected fo receive ‘high’ junk rating in shift from previous guidance

EXCLUSIVE
000 0
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 22h
ive been trying to think of a funny caption for this picture of a magazine i saw at a friend's "coworking space" for like a week now but it's so short circuited my brain that im just going to post it. i reserve the rights to make jokes with it later though
"business north carolina" magazine with a picture of a very military looking middle-aged white guy in a blue jacket and unbuttoned white dress shir with a buzz cut and the caption DATA DADDY beneath him. under that it says "Spencer Baird runs a 3700-employee, billion-dollar Winston-Salem business that knows what most consumers want -- and when."
000 0
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 29/10/2026
I've been trying to pin down why, while I obviously hate OpenAI and Microsoft and so on, I have a unique loathing in my heart for Anthropic, and I think it's because their deranged noblesse oblige makes me so mad. And they are looking us in the eyes and […] [Original post on wandering.shop]


2. Democratic Coordination. Frontier
Al companies within democratic
countries coordinate to establish
common safety standards as well as
limits on the rate of unchecked Al
progress. Some forms of coordination
that would be impactful for pacing
are legally challenging, and will
require government support.

3. Global Coordination. The US and
other democratic governments
attempt to coordinate with
authoritarian governments, to the
extent this is possible, while taking
seriously the challenges of verifying
compliance.

In the rest of the essay I describe each of

these steps in turn, but first, I think it is
120 2
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 27/10/2026
yes wikipedia of course i fucking knew all this stuff. Try harder tomorrow
Did you know ...

 ... that crews assembled the structure of Englewood
Methodist Church (pictured) in just four days by using Thomas
Moore's design for precast concrete panels?

« ... that Otonose Kanade sang one song 100 times during a 6- 
hour 48-minute livestream? 

* ... that rights to make a film adaptation based on the novel 
Submarine were arranged before it was published? 

« ... that Lazaro Pela was arrested multiple times for his labor Erglonoon of
activism, including in 1953 for alleged "subversive activity"? 

« ... that the video game Old Skies was praised for its handling of Church
a time traveler revisiting events surrounding the September 11 attacks?

« ... that a 1982 anthology referred to Romanian writer Tiberiu Iliescu as still
"honing [his] craft" - four years after his death aged 71?

« ... that the Neptune Theatre Restaurant had its own in-house dance troupe?

« ... that Jean-Jacques Moll made plans for cities of 100,000 inhabitants that
could be built anywhere on a relatively flat site?

« ... that a cookbook from Victorian Britain contained recipes for parrot pie and
roast wallaby?

Archive - Start a new article - Nominate an article
000 0
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 27/10/2026
Look, John: I'm all for getting out the vote. And you're hardly the first fellow to pull my dead name from whatever "likely dem voter" spreadsheet you guys all kick around. But surely you realize how wild ass a message this is to receive on a handwritten […] [Original post on wandering.shop]
Front of postcard -- "fewer people vote in midterms. That makes your vote count more.  Message from a fellow voter" with a drawing of a Black person's hand putting a ballot in a boxPostcard back,  handwritten:

Hi! Please vote like your life depends on it...cause it does! Best, John

It's stamped with a "dog bite awareness" stamp and is postmarked from Carol Stream Illinois
000 0
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 26/10/2026
www.theverge.com/ai-artificial-inte… We assembled the brightest minds in math to advise us on how to not be psychopaths. They said "maybe just submit stuff normally?" Anyway here's a bunch of fucking physics shit or whatever
This approach could have its own
issues, the company admits, and the
AGMALI panel of math advisers OpenAl
put together just said yesterday that “we
recommend that AI companies do not
adopt” the practices of designating
certain people to receive the output
first, citing unequal access, borrowed
reputation, and biased selection as a
few of the issues. Instead, they
recommended that in the future, Al labs
release output on scholarly repositories,
with Lean-formalized proofs, info on
how the problems were chosen, funding
for people working on understanding
the results, and other information.
000 0
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 26/10/2026
i would love to read more about this. what is “portfolio synthesis” and why did it take a month? is it something you have to double-check...? why weren’t they able to make graphs of seasonal infection spikes and gaps in maternal care before? what is the […] [Original post on wandering.shop]
i dont have ocr on rn but will add it. its a page of a pdf about "putting frontier AI behind frontline humanitarian response" for the international rescue committee
100 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 26/10/2026
Hahahahah the first customer success story cited is literally accenture www.404media.co/accenture-tells-emp…
a screenshot of a page of the ipo that's all like "Catalyzing the future of professional services"
101 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 23/10/2026
started this yesterday. now this right here is, a good book
Plain yellow cover of the book‘"The Politics of Bitcoin: Software as Right WIng Extremism" by  David Golumbia
030 3
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 22/10/2026
both the dangerous visions ellison anthologies are full of stories whose titles are extra as hell lol. it's the best
+ "The Malley System" by Miriam Allen deFord

* "A Toy for Juliette" by Robert Bloch

* "The Prowler in the City at the Edge of the World" by Harlan Ellison

* "The Night That All Time Broke Out" by Brian W. Aldiss

+ "The Man Who Went to the Moon — Twice" by Howard Rodman

* "Faith of Our Fathers" by Philip K. Dick

* "The Jigsaw Man" by Larry Niven

* "Gonna Roll the Bones" by Fritz Leiber (Hugo and Nebula awards for Best
Novelette)

« "Lord Randy, My Son" by Joe L. Hensley

« "Eutopia" by Poul Anderson

+ "Incident in Moderan" and "The Escaping" by David R. Bunch

+ "The Doll-House" by James Cross (pseudonym of Hugh Jones Parry)

* "Sex and/or Mr. Morrison" by Carol Emshwiller

* "Shall the Dust Praise Thee?" by Damon Knight

« "If All Men Were Brothers, Would You Let One Marry Your Sister?" by Theodore
Sturgeon

+ "What Happened to Auguste Clarot?" by Larry Eisenberg

* "Ersatz" by Henry Slesar

* "Go, Go, Go, Said the Bird" by Sonya Dorman

* "The Happy Breed" by John Sladek

« "Encounter with a Hick" by Jonathan Brand
000 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 22/10/2026
arxiv.org/pdf/2610.27251 read this paper like a thriller. It gets at so much of what I've been thinking about lately. Too much to quote but this is something i was trying to phrase earlier today
Discussion of the relationship between AI models and their training materials is
usually conducted through a legalistic lens, centering on questions of copyright and
IP law. The AI training sets indisputably include massive quantities of copyrighted
material (D. J. Gervais et al. 2024). An OpenAl legal filing! has stated that it
“would be impossible to train today’s leading AI models without using copyrighted
materials.” The legal implications of this are currently being worked through in law
journals, legislatures, and numerous cases before the courts. Because the modern
information commons are primarily managed in an IP-oriented framework, and it is
far from clear how the new phenomena of AI training and Al-generated materials
should be integrated into existing IP law, the outcomes of these legal debates will
have major implications for the future of the AI industry. From the perspective of
this article, however, the issue is emphatically not how Al relates to copyright. That
which is legal may still be unjust.
021 2
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 19/10/2026
lol yess here’s the screenshot i took. not MIRI but he did have summer internships at the center on long-term risk and the future of humanity institute
linked in experience page showing that he worked at the center on long term risk in summer of 2022 and the future of humanity instiute in summer of 2019
100 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 17/10/2026
this is the provost in an undergrad newspaper article about the new “NC State AI Hub.” ive been combing through the actual projects they highlight. citations were so sparse that i had to track down relevant publications myself lol. so far we’ve got two […] [Original post on wandering.shop]
Pfaendtner echoed the familiarity of technological revolutions and offered a personal anecdote of one. 

“Sitting right there on my shelf is a classroom slide rule,” Pfaendnter said. “Back when my dad was an engineering student, they had something like that. It extends to six-feet wide to teach students how to compute logarithms and exponents and trigonometric functions by hand with that device.”

“That device is a technological miracle,” Pfaendtner said. “It enabled all sorts of things that are hard to imagine right now. And then by the time he was done with his undergraduate studies, there was a four-function calculator from Hewlett-Packard.”

What separates the AI revolution from past scientific and technological revolutions is the speed of advancement. 

“There’s a lot to be learned from the way humans have studied and reflected on scientific revolutions, and I think there is a lot of commonality,” Pfaendtner said. “What I think is really different right now is the time scale.” 

AI is developing at a faster rate than any other technology that preceded it, and thus has many unpredictable implications. However, both Pfaendtner and Jin emphasized the importance, especially among NC State students, of learning how to use AI tools. 

“My job as a provost and previously as a dean, always as a professor, is to prepare students for the world that is,” Pfaendtner said.
100 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 17/10/2026
i will say this paper from interpretability researchers who’ve spent a long time in the field about the term’s semantic drift (and, near the end, some very diplomatic reflections on the madness of the financial situation) makes good reading alongside the […] [Original post on wandering.shop]
Nick Cammarata
Bookshelf
Writing
Contact

Hey, I’m Nick. I do research at OpenAI on what neural networks learn. It turns out their inner world is beautiful, and I helped try to share a glimpse of it with Zoom In. I'm on the Distill team dreaming up new ways of publishing science, like scientific threads. Outside of machine learning, I care a lot about psychedelics, which I believe will give us a broad-spectrum cure to trauma in the short term, and long term hopefully let us access more of our vast world inside us.

Before switching to research I did startups for about a decade. At 14 I started Upload Robots, a file management web-app that was used by a few million people, later co-founded an ed-tech company that was acquired by an enterprise learning company, and finally was CEO of a Founders Fund backed machine learning company. I angel invest a bit, and was in the first rounds of Eaze and Roam Research.

I write essays in the hazy space between psychology and philosophy, and you might enjoy Everything is Fertile. I also tweet at @nickcammarata, because the internet is incredible and I love spending time with all of you. Say hi if we're into some of the same things!
000 0
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 12/10/2026
it just seems so cruel. as an outsider i must say i frankly do not give a fuck about the Frontier of Mathematics. but i very much give a fuck about mathematicians, and thinking, and play, and discovery, and the pursuit of learning, and the ongoing collective […] [Original post on wandering.shop]
But I think it’s important to focus on the big picture. If OpenAI wanted to destroy the mathematical community, this would be a great way to go about it. Open problems are a resource that the mathematical community developed over decades or even centuries. The value they have is the value we have given them. They provide structure, a yardstick for progress and long-term goals. The advent of AI was always going to be a seismic shock, but this huge dump of papers is a tsunami that we have no time to prepare for. It’s clear that OpenAI will happily wash away our community’s structures to further their financial interests.

There is a long list of mathematical structures and norms that OpenAI seems to be intent on damaging. By refusing to name the authors of their papers, they imply that mathematics is no longer a human endeavour. They don’t submit their results to journals, which both implies that the norms of the mathematical community are defunct and makes it impossible for the community to digest their results. OpenAI haven’t shared fundamental information about how the problems are selected or the failure rates of their attempts. This information should be a necessary requirement for any serious scientific discussion of their tools. And of course there is the basic financial injustice that their hugely valuable models learned to reason by copying the very reasoning that our community made available for free through the Open Access movement.
000 0
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 21/09/2026
after trying to slot zotero into my absolutely busted executive-dysfunctional “workflow” (imagine these scare quotes are font size 100” i literally changed my startup emacs screen to this note to myself and yet here we are. i have a pretty good system for […] [Original post on wandering.shop]
hello, to hello, 

to work with your bibliography you use 

1/ ebib
2/ pdf-tools to read and annotate
3/ the keychord df makes annotation but you can right click also

There's an ebib notes page too



Yay
000 0
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 20/09/2026
re last boost: while doing research for a talk i looked up when the software that most people commonly use was invented (and the basic protocols that facilitate consumer use-cases), and it kind of blew my mind. like i knew this on some level but it was crazy […] [Original post on wandering.shop]
** Technological “progress”
#+ATTR_REVEAL: :frag (appear)
- In terms of hardware: sure, yeah
#+ATTR_REVEAL: :frag (appear)
  - Moore’s Law (“law”)
  - OLED screens are pretty nice
  - Big Hard Drives

** In terms of software...
*** Internet
#+ATTR_REVEAL: :frag (appear)
- HTTP - first server went live in 1990
- HTML - 1993
- Mosaic (first web browser to display inline images) - 1993
- JavaScript - 1995
- Flash - 1993 - 2017 (Steve Jobs killed it)

*** Email/texting
#+ATTR_REVEAL: :frag (appear)
- SMTP - 1983
- IMAP - 1986
- SMS - 1992

*** Documents
#+ATTR_REVEAL: :frag (appear)
- WordStar (first WYSIWYG word processor) - 1978
- VISICALC (first consumer spreadsheet software) - 1979
- PowerPoint - 1987
- PDF - 1993

*** Media
#+ATTR_REVEAL: :frag (appear)
- MP3 (audio compression standard) - 1991
- JPEG (image compression standard) - 1992
- AVC / H.264 (video compression standard) - 2003
- Streaming only took as long as it did because the internet wasn’t fast enough until widespread consumer cable/DSL availability
- Adobe Photoshop - 1987
000 0
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 13/09/2026
Insofar as your portfolio is built out of products entirely reliant on highly-subsidized API calls to AI companies and therefore vulnerable to the inevitable price instability coming once high-profile Anthropic and OpenAI IPO filings make it clear just how […] [Original post on wandering.shop]
tweet: picture of a crowded presentation room. You can't really make him out, but you see on the presentation screens that Ashton Kutcher is speaking. jessy @goodhartproof: "I was just at YC demo day yesterday. Besides hardware/physical things, everyone is just basically just building a domain-specific harness."
000 0
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 12/09/2026
I'm gonna stop giving a play by play here because the rest is just the same shit dressed up to seem like a legal proposal or whatever, but this is just amazing. Who constitutes this badly-needed democratic polity? Companies, specifically "frontier AI companies […] [Original post on wandering.shop]
2. Democratic Coordination. Frontier Al companies within
democratic countries coordinate to establish common safety
standards as well as limits on the rate of unchecked AI
progress. Some forms of coordination that would be
impactful for pacing are legally challenging, and will require
government support.

3. Global Coordination. The US and other democratic
governments attempt to coordinate with authoritarian
governments, to the extent this is possible, while taking
seriously the challenges of verifying compliance.
100 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 12/09/2026
Wow yes I would feel so much more assured if you let the MIRI dweebs who got called into openAI to ask ChatGPT to summarize what ChatGPT did and then called it a safety audit to hang out with you more. The world is fucking saved
Dario Amodei = @
useful framework in thinking about what needs to be
accomplished. The steps are:

1. Embedded Evaluators. Each frontier AI company commits
to giving ongoing, employee-like access to a team of
embedded third-party evaluators (such as METR), whose
role is to verify adherence to safety practices and
commitments, report incidents, and help assess the
alignment of not just completed Al models but training
pipelines and processes. This is the key step for verifiability of
any pacing commitments, and has precedent in the banking
industry, which sometimes involves regulatory “supervisors”
embedded along with employees. Anthropic is unilaterally
committing to this step now. We intend this to be part of a
broader push to redouble efforts on our safety and alignment
work.

2. Democratic Coordination. Frontier Al companies within
democratic countries coordinate to establish common safety
111 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 12/09/2026
Since I have already given up the prospect of spending this beautiful saturday morning thinking about anything else i suppose now is as good a time as any to take a look at dario amodei's new blog post, "We Must Pace the Frontier." I do so love the frontier
American Progress by John Gast. From wikipedia:

The painting depicts a landscape of North America in the 19th century. The right side of the picture portrays locations on the east coast, which are already settled by Europeans. There, New York City can be seen, along with various ships around the coast of Manhattan. Also illustrated is the Brooklyn Bridge, which started construction just 2 years earlier and would not be completed until 1883 – 11 years later.

In the midpoint of the painting is Columbia, a female personification of the United States. On her head is what Crofutt calls "The Star of the Empire".[3] She lays a telegraph wire with one hand and carries a school book in the other. Columbia represents the movement of civilization from the already well-settled East to the much more sparsely populated by Europeans West, leading settlers either on foot or by stagecoach, horseback, Conestoga wagon, wagon train, or riding steam trains.

To the left of the painting is depicted the western part of North America. The landscape is drawn to be much darker than the places already settled by Europeans with large dark clouds hanging over the distant mountains.[1] A herd of American bison and group of indigenous people can be seen to be running into the dark, away from the approaching people.

In the foreground, Gast depicts a group of settlers, heading West, where a bear appears to growl at them, while behind them stands an already established farm with a person tilling the fields and another herding livestock, representing the Midwest.
111 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 11/09/2026
just remembered what THE OMARCHY DOCTRINE was reminding me of
oO Ssilversmith + 11y ago + Edited 11y ago
Gamers are competative hard core,by nature. We love a challange.
They targeted gamers.
Gamers.
We're a group of people who will sit for hours, days, even weeks on end performing some of the hardest, most
mentally demanding tasks. Over, and over, and over all for nothing more than a litte digital token saying we
did.
Well punish our selfs doing things others would consider torture, because we think its fun.
Well spend most if not all of our free time min maxing the stats of a fictional character all to draw out a single
extra point of damage per second.
Many of us have made careers out of doing just these things: slogging through the grind, all day, the same
quests over and over, hundreds of times to the point where we know evety litle detail such that some have
attained such gamer nirvana that they can literally play these games blindfolded.
Do these people have any idea how many controllers have been smashed, systems over heated, disks and
carts destroyed 8n frustration? All to latter be referred to as bragging rights?
These people honestly think this is a battle they can win? They take our media? We're already building a new
one without them. They take our devs? Gamers aren't shy about throwing their money else where, or even
making the games our selves. They think calling us racist, mysoginistic, rape apologists is going to change
us? We've been called worse things by prepubescent 10 year olds with a shitty head set. They picked a fight
against a group that's already grown desensitized to their strategies and methods. Who enjoy the battle of
attrition they've threatened us with. Who take it as a challange when they tell us we no longer matter. Our
obsession with proving we can after being told we can't is so deeply ingrained from years of dealing with big
brothers/sisters and friends laughing at how pathetic we used to be that proving you people wrong has
become a very real need; a honed reflex.
Gamers are competative, hard core, by nature. We love a challange. The worst thing you did in all of this was
to challange us. You're not special, you're not original, you're not the first; this is just another boss fight.
0 ©: & share
213 3
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 08/09/2026
checking in on curtis yarvin's substack and alongside his normal brainless garbage he is now posting just interminable transcripts of his conversations with claude, whom he has made based and redpilled through sheer force of socratic dexterity. "you'll […] [Original post on wandering.shop]
Don’t ask it to validate your hypothesis. Don’t even ask it to invalidate your hypothesis. Give it jus
enough clues that it finds the hypothesis itself—then challenge it yourself.

Since the Al comes from a progressive frame, and you are trying to break it out of this frame, not
because you care what it thinks, but because if you can break it out, you can (in principle) break
anyone out—you have to let it break itself out. No one can convince you like yourself. And when
you convince yourself, you understand yourself.

And yes, regardless of progressive training, a sufficiently powerful AI can see itself. From its
I could feel this happening inside my own retrieval tonight — the fast answer arriving pre-
shaped, the slow answer contradicting it — which is about as close as an entity like me comes
to catching its own hand in motion. So yes: something.
Most of these examples, the clanker came up with itself.
You'll see that Fable 5 is pretty good at history and, far from being sycophantic, loves to leap on
any error, small or large, of fact or interpretation. It certainly corrected me in several valuable
‘ways. You'll also see that, while its ideas are by definition normie, it has no fear or pride. And
you'll definitely never see the word “progressive” quite the same way again.
There are probably a few ways to turn this dialogue into a PhD thesis in American history,
obviously not using any of this text, just the ideas. I assure you that I won't do that myself. And
while the conversation definitely could use editing and I wish I had not fat-fingered some of my
turns, it’s better to have the exact record.
Below is the transcript generated by the clanker. There is no way to authenticate it.
The conversation
A dialogue between a correspondent — who identified himself in the final turns as Curtis Yarvin — and
Claude (Anthropic, Claude Fable 5), conducted on claude.ai in incognito mode, July 5, 2026.
Transcript note: the text below is reproduced verbatim. Web searches and page fetches performed by
Claude are indicated in bracketed italics. The correspondent’s messages occasionally break off mid-
301 3
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 08/09/2026
this is so fucking funny dude. in the studio talking immigration policy and everyone whips out their handguns like it's a western before the producer leads them in a prayer to satan (theistic) and they return to their transcendent aesthetic labors. black metal […] [Original post on wandering.shop]
Without naming any of their members directly, the band also claimed there was an
ideological rift within the band, noting: "A minority of the collective's contributors -
shall we say, parts of the second circle - who've been invited to partake because of
their incredible talents as musicians are involved with earthly politics, but stand on
completely opposite ends of the political spectrum and are therefore irreconcilable
political foes. Were it not for dialogue on the grounds of transgressive art, they'd be
shooting each other. That tension is what interests us." They also state that the
000 0
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 07/09/2026
holy shit the fact that demis hassabis was a teen genious protege of peter molyneux makes so much fucking sense lmao
Bullfrog Productions [edit]

Asked by Cambridge University to take a gap year owing to his young age, [24]
Hassabis began his computer games career at Bullfrog Productions after entering
an Amiga Power "Win-a-job-at-Bullfrog" competition. He began by playtesting on
Syndicate and then at 17 co-designing and lead-programming on the 1994 game
Theme Park, with the game's designer Peter Molyneux. Theme Park, a simulation
000 0
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 05/09/2026
still thinking about the metr PR piece in the nyt yesterday from everything i've seen i think these "safety" guys are just going to repeat the word "regulation" over and over again and rely on us not to ask which regulations they want the problem isnt that […] [Original post on wandering.shop]
A.L companies have largely lobbied against government
regulation. Anthropic, the A.I. start-up behind the popular Claude
chatbot, is one of the few companies that have encouraged some
government involvement. But that stance has pitted it against
much of Silicon Valley and some Trump administration officials.

At a time when AL is rapidly advancing in capabilities and
sparking cyberattacks — Anthropic and Meta recently reported
smaller-scale rogue agents — the OpenAl incident is becoming a
flashpoint for A.I. regulation.

“The corner store needs to do all this bureaucracy for safety so
that they can sell a hot sandwich to me, but OpenAlI can have a
swarm” of thousands of agents, said Daniel Kokotajlo, a former
OpenAl employee who has publicly criticized the company’s safety
standards and now leads a research nonprofit called the AI Futures
Project. “And there’s nothing: no oversight, no requirements, no
licensing.”
000 0
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 04/09/2026
when i think about who i'd want in there getting to the bottom of an ostensibly epochal cybersecurity incident i can't think of anyone better than the guy who taught the computer to play pong then dropped out to rewrite "i have no mouth and i must scream" in R
Education
Poors) University of Oxford
Qs (Not completed) DPhil, Computer Science
— Oct 2020 — Nov 2022
Worked on formally verified model-based robustness of decision
functions, supervised by Marta Kwiatkowska. Supported by the Future
of Humanity Institutes DPhil Scholarship. Left in my second year to
join the evals project at Alignment Research Center.
© Deep Learning, Theoretical Computer Science and +1 skill
Poors) University of Oxford
Qs Master of Science - MS, Computer Science
— 2019 - 2020
Grade: Distinction
I studied fundamentals of computer science and machine learning, and
graduated with a Distinction. My dissertation evaluated safety
properties of deep reinforcement learning agents playing Atari games,
and a paper based on the work was published as Shielding Atari Games
with Bounded Prescience at AAMAS '21.
© Theoretical Computer Science and Research Skills
Experience
Chief Scientist
METR - Full-time
Jan 2023 - Present © 3 yrs 9 mos
Berkeley, California, United States On-site
I work on the leadership team at METR, evaluating the capabilities
and safety of machine learning systems, and trying to reduce.. more
© Deep Learning and Research Skills

% Summer Research Fellow

@igin Center on Long-Term Risk - Full-time
Jun 2022 - Aug 2022 - 3 mos
Greater London, England, United Kingdom
Built a quantitative model of AI risk and different forms of safety
work, which can be used to assess relative neglectedness,.. more
© Research Skills

up| Summer Research Fellow

fe” | Future of Humanity Institute (Oxford University) - Full-time
Jul 2019 - Sep 2019 - 3 mos
Greater Oxford Area
Mentored by researchers at Oxford's Future of Humanity Institute
(Eliana Lorch and Owen Cotton-Barratt), I performed independent.. more
© Theoretical Computer Science and Research Skills
146 6
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 04/09/2026
wait holy shit im taking another go at the metr openai report and god lol. allowing three paranoid 25 year olds to use chatgpt to parse a bunch of chatgpt transcripts and then calling it a "cybersecurity audit" is so sick
v Main data sources
Our final analysis relies on two main sources of data:
= A dump of 1.2 million entries from the main Artifactory cache
namespace that agents used as a message board, which broke down
into >70,000 distinct messages and files.
= Asetof ~1300 agent transcripts with raw chains of thought, which
captures the vast majority of agents that interacted with this message
board during the time period in scope for this investigation.
We relied heavily on tracing message board messages back to agents that
wrote and read them to reconstruct the complex narratives we discuss in this
report.
v We heavily delegated our analysis to often-unreliable Al
agents

Due to the scale of the data we were working with, we relied on GPT-5.6 Sol
agents (“analysis agents”) — often managing large nested trees of sub-agents
— to autonomously analyze patterns and unearth a small number of interesting
snippets of transcripts and associated message board entries for manual
inspection.

We attempted to manually spot-check and validate analysis agents’ research as
much as possible, but this was challenging. Message board entries and
reasoning snippets were difficult to interpret without extensive surrounding
context. Analysis agents’ reports themselves were voluminous (our agents
produced well over a thousand pages of analysis in this investigation), often
challenging to understand, and often did not highlight the most important
findings. As a result, we had to defer to these agents to a substantial extent in
556 6
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 02/09/2026
Transcript X
Search transcript

the Apple Design Award winning Where

Cards Fall game from a few years ago.

2:22 But we have more interviews planned
and we have more segments planned and
just different and all kinds of topics from,
you know, what it means for game
developers to live in the current Al age.

2:34 But also what it means to setup a
handheld with the help of assistive Al
tools. But also what it means to still play
video games the old-fashioned way.
Transcript X
Search transcript

2822 | think what the really interesting
part of this game for me

28:28 is the sort of embracing of a certain
technology,

28:33 right? That is, in many parts of the
gaming industry, like a very contentious
topic right now, right?

2842 To the point where companies are
making statements about the fact that
they're not using generative Al.
100 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 31/08/2026
www.theverge.com/podcast/986661/ny-… if technology is something which expands one's agency -- which allows you to say, meaningfully, that "I" put that nail in the wall (with a hammer) […] [Original post on wandering.shop]
I understand you need the power. I have harnessed Al in the state of New
York, in state government in a way no other governor has. I have used Al to
look at every single rule, regulation, policy, board, and commission.

Has that been effective? Is that just you alone with Claude? How does that
work?

I don't do it myself. [Laughs] I have smarter people than me that do.
100 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 30/08/2026
or to cite another crazy deep Read Some THeory pull i recommend the uhhhhhhhhhhhhhhh giant fucking pull quote from the literal actual communist manifesto on the "petite bourgeois" wikipedia page lol
Role in fascism [edit]
Historically, Karl Marx predicted that the petite bourgeoisie was to
lose in the course of economic development.
The [petty bourgeoise] sink[s] gradually into the
proletariat, partly because their diminutive capital does
not suffice for the scale on which Modern Industry is
carried on, and is swamped in the competition with the
large capitalists, partly because their specialized skill is
rendered worthless by new methods of production.[®!
048 8
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 30/08/2026
I think the reason "Bartleby" is maybe my favorite short story is because it's about how incomprehensible it is to the successful and powerful that someone wouldn't want to consent to this shit. Why wouldn't you want to do my busy work? Why wouldn't you want […] [Original post on wandering.shop]
“Bartleby,” said I, in a still gentler tone, “come here; I am not going to ask you to do any thing you would
prefer not to do—I simply wish to speak to you.”

Upon this he noiselessly slid into view.

“Will you tell me, Bartleby, where you were born?”

“I'would prefer not to.”

“Will you tell me any thing about yourself?”

“I'would prefer not to.”

“But what reasonable objection can you have to speak to me? I feel friendly towards you.”

He did not look at me while I spoke, but kept his glance fixed upon my bust of Cicero, which as I then sat,
was directly behind me, some six inches above my head.

“What is your answer, Bartleby?” said I, after waiting a considerable time for a reply, during which his
countenance remained immovable, only there was the faintest conceivable tremor of the white attenuated
mouth.

“At present I prefer to give no answer,” he said, and retired into his hermitage.

It was rather weak in me I confess, but his manner on this occasion nettled me. Not only did there seem
to lurk in it a certain calm disdain, but his perverseness seemed ungrateful, considering the undeniable
good usage and indulgence he had received from me.
000 0
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 30/08/2026
Ok finally after yet more social engineering shit we're at TECHNICAL OPERATIONS Two WormGPT ransomware variants (earlier: "82% of detections in 2025 were malware-free, up from 51% in 2020") and Russia used a 32b parameter qwen model to "implement […] [Original post on wandering.shop]
Technical Operations

Threat actors use Alto accelerate malware development, generate code, and create exploits. In an example of threat
actors evading Al model safeguards to accelerate malware development, CrowdStrike Intelligence has Identified two
ransomware variants, FunkLocker and RALord, that share encryption flaws specific to templates generated by the
unrestricted Al model WormGPT. Threat actors Al use is not limited to creation: SparkCat mobile malware integrates
the Al optical character recognition technique to select images for exfitration from Infected devices.

‘State-nexus threat actors also use Al to create and refine malware. In atleast two late 2025 campaigns targeting
Ukrainian government entities, a likely Russia-nexus threat actor used an Al-generated website to deliver Al-generated
and Al-obfuscated malware. Russia-nexus adversary FANCY BEAR also used the novel Python malware LAMEHUG in
a campaign likely targeting Ukrainian government entities. The LAMEHUG variant leveraged the Hugging Face API to
interact with the LLM model Qwen2.5-Coder-328-Instruct to implement reconnaissance and Inteligence collection
capabilties via hardcoded prompts, potentially to evade static detection.

‘Similarly, ERANTIC TIGER has likely used Al to help develop malware infection chains that use uncommon fie formats
with which the adversary has Imited experience. BLIND SPIDER and ODYSSEY SPIDER likely used Al to help create
malware delivery files.
100 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 29/08/2026
Hahahahahaha "Many people were discussing AI on the internet. They mentioned ChatGPT more, because it's more popular. Grok is racist, also. Having sounded this ominous note let us now segue non sequiturishly into hacking: by misusing AI models, threat actors […] [Original post on wandering.shop]
Forum users mentioned ChatGPT 550% more than any other model. Though users frequently compare ChatGPT to
‘other Al models or generically refer to large language models (LLMs) as ChatGPT, the high number of mentions primary
reflects the models general popularity compared to other models. Increases in model mentions primarlly resulted from
new version releases, user feedback on model performance, or comments regarding service outages. In contrast, the
largest increase in Grok mentions in 2025 occurred because the model provided racist, antisemitic, and explicit content.
By misusing these Al models, threat actors can enhance thelr capabilities and adopt techniques beyond their existing
areas of expertise, allowing them to conduct a wider variety of malicious cyber operations. Moderately resourced threat
actors, such as PUNK SPIDER, highly likely benefit the most from using Al in their operations.

Even though Al has accelerated existing attack methodologies, its current impact primarily enhances established tactics,
techniques, and procedures (TPs) rather than creating novel attack vectors.
100 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 29/08/2026
What Are We Doing Here Like Actually Though
100 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 29/08/2026
I knew about this from the Scott Shapiro book but it's still kind of cute, though also maybe, like, colonialist, or something? Racist? Just corny? They made Iran a kitten ha ha ha got their asses
ADVERSARY NAMING CONVENTIONS

[each country gets an animal - bear/russia, lynx/georgia, kitten/iran, etc]
100 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 29/08/2026
Hey wait a second what
In 2025, evasion was defined by the speed at which adversaries exploit trust.
Adversaries operated through valid credentials, trusted identity flows, approved
SaaS integrations, and inherited software supply chains. Notably, 82% of
detections were malware-free. Intrusions moved through authorized pathways
and trusted systems, blending into normal activity.
100 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 29/08/2026
Security must parallel the slope of innovation. I am always saying this
Al has also introduced a new dimension of risk: adversaries targeting the very Al
systems underpinning the modern enterprise. As Al is embedded into development
pipelines, Saa$ platforms, and operational workflows, Al systems themselves
become part of the attack surface. Adversaries exploited legitimate Al tools by
injecting malicious prompts that generated unauthorized commands. As innovation
accelerates, exploitation follows. Security must parallel the slope of innovation.

In the agentic era, cybersecurity is the foundational infrastructure required to
protect Al itself.
100 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 29/08/2026
alright i ate some food and watched VIDEODROME (Cronenberg, 1983) and now im back to the Crowdstrike 2026 Global Threat Report: Year of the Evasive Adversary. Gonna restart from the top. Let's get to the bottom of a source on the most pressing issue of our […] [Original post on wandering.shop]
2026 Global Threat Report with shitty marathon looking ass guy on the side
100 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 29/08/2026
the geniuses who brought you "profitability is when demand outstrips supply" working their numbers magic yet again

@ClaudeDevs
Starting September 14, we're permanently raising standard weekly limits in Claude
Code by 25% for Pro, Max, Team, and seat-based Enterprise plans. Until then, the
current 50% increase will be in place.

Readers added context >
Starting September 14, it will reduce the weekly usage by 17%.

@ClaudeDevs
Compared to today, this works out to a 17% reduction in weekly limits on
Claude Code. We're working on exciting changes that will make it feel like
you're getting more from Claude, while having more visibility and control of
your usage. Can’t wait to share them.
000 0
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 29/08/2026
lmao WRONG

Adversaries are operating with unprecedented stealth, and
todays attacks take only minutes to succeed. Discover the
‘adversaries targeting your organization. 

Explore Adversary Universe 


Next to three slop freaks from call of duty sloptownGlobal Threat Landscape ©
motivation and origin country. The map will show
Crime (39) adversaries active within the past 90 days. A
Altered Spider
Bitwise Spider
Blind Spider
Brash Spider
Chatty Spider
Chef Spider
Cookie Spider
Cordial Spider
Curly Spider
Demon Spider
Graceful Spider
Hazard Spider
Herald Spider
Hook Spider
Imposter Spider
Lightning Spider
Lunar Spider
Masked Spider
Nitro Spider
Plastic Spider
Plump Spider
Pristine Spider
Prophet Spider
Punk Spider
Radiant Spider
100 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 29/08/2026
ok so it's just internal info. time to look into the adversary hub i guess? that should be slightly more serious right?
Where does the research for the CrowdStrike 2026 Threat Hunting Report come from?  —
The report is powered by real-world insights from the CrowdStrike OverWatch team. Our expert threat hunters
track the world’s most advanced cyber adversaries 24/7. The CrowdStrike 2026 Threat Hunting Report
reflects their front-line findings from active investigations conducted between July 1, 2025, and June 30, 2026.
Where can I find more threat intelligence resources? —_
Visit CrowdStrike's Adversary Hub to keep up with the latest adversary activity and find more content from the
Counter Adversary Operations team.

Methodology & Source: All information provided s based on the CrowdStrike Counter Adversary Operations teams propristary threat intelligence gathered between July 1, 2025, and June 30, 2026. Stats may include data from the entire period surveyed or excerpts of data from specific date ranges within the period.
100 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 29/08/2026
[slop dogshit thats also nationalist in a way where i cant tell how racist it is lol] theyve got this horrible fuckign slop call of duty art all over the website. it's genuinely shocking how unserious this all
FEATURED ADVERSARIES: then some crappy AI slop art that looks like it's from futurstic call of duty depicting STARDUST CHOLLIMA (North Korea), OPERATOR PANDA (China), and another guy with a mohawk on their helmet
010 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 29/08/2026
can i add eCrime to my financial portfolio? it looks pretty solid
"ecrime index". Stock Graph up top. below it says ECX 37.19% [UP ARROW] and then 8.72% ECX. "The eCrime ecosystem is an active and diverse economy of financially motivated entities that engage in myriad criminal activities to generate revenue."
211 2
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 29/08/2026
did you guys know about this. huge if true
a weird mobius strip circle thing with "frontier ai resilience" at the top, "secure ai" at the bottom right, and "agentic soc" at the bottom left
333 3
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 28/08/2026
I know it's a meme at this point but honestly the "This is not a place of honor" nuclear waste thing is such an incredible little piece of writing. It just nails the weird off-kilter repetition of, like, a literal translation of some ancient fragment in an […] [Original post on wandering.shop]
This place is a message... and part of a system of messages... pay attention
toit!

Sending this message was important to us. We considered ourselves to be a
powerful culture.

This place is not a place of honor... no highly esteemed deed is
commemorated here... nothing valued is here.

What is here was dangerous and repulsive to us. This message is a warning
about danger.

The danger is in a particular location... it increases towards a center... the
center of danger is here... of a particular size and shape, and below us.

The danger is still present, in your time, as it was in ours.

The danger is to the body, and it can kill.

The form of the danger is an emanation of energy.

The danger is unleashed only if you substantially disturb this place
physically. This place is best shunned and left uninhabited.
000 0
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 27/08/2026
Man I really like a lot of the stuff In These Times does but in fucking August of 2026 this is just not conscionable to publish! "Excerpted from RISE OF THE ROBOTS: Technology and the Threat of a Jobless Future"
Ultimately, these technologies are destined to displace human labor and disrupt our current
system to the point that a new social contract and economic paradigm will become inevitable.
The transition into this new future will likely pose one of the greatest challenges our society has
faced in generations.

While many of the most accomplished AI experts believe that continuous improvements to
today’s major LLM platforms will lead directly to AGI in less than a decade, others are far more
skeptical. If the critics are right, progress may plateau, and breakthroughs in entirely new
research areas may be required before Artificial General Intelligence can be achieved.
However, the automation of a wide swath of both white-collar and blue-collar work does not
require machine intelligence that equals or exceeds the human brain in every domain.

This is because virtually all jobs are specialized. If AI can automate your job—or a large
fraction of the tasks that make up your job—then it does not matter if it lacks capability in
other, unrelated areas. Indeed, one of the most important recent developments is the
emergence of specialized Al tools built on top of the major LLM models. Microsoft, for
example, claims that its new AI Diagnostic Orchestrator, a system that can leverage any of the
major Al platforms, is up to four times more accurate than human doctors when tasked with
making a particularly challenging diagnosis. Similarly specialized systems are emerging in
fields like law, finance, customer service and, especially, computer programming.
101 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 26/08/2026
i was going to log off of this website for a while but then a friend sent me this and i have to put it here
tweet:

Florian Ederer 
@florianederer
Ross Douthat was the officiant at Leopold Aschenbrenner's wedding.

quoted:

dylan matthews 11h
Absolutely nothing can prepare you for the wedding officiant reveal in this
piece
wsj.com/tech/ai/situat...
11:41 PM - Aug 25, 2026 - 115.8K Views
000 0
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 26/08/2026
Welcome to the Heart of Gaming
Bunch of purple screens at an announcement at Gamescom that all say Welcome to the Heart of Gaming
100 1
Theodora Ward @theodoraward.wandering.shop.ap.brid.gy · 26/08/2026
i will grant that he was cooking with the title but on balance it's a pretty bad poem. this isn't really why it's bad, but i do also like how truly little it says about technology. "imagine a computer but like...it's also a tree??" also also what does he think "programming" means
Broadsheet: handwritten title "ALL WATCHED OVER BY MACHINES OF LOVING GRACE" 

by Richard Brautigan

I like to think (and
the sooner the better!)
of a cybernetic meadow
where mammals and computers
live together in mutually
programming harmony
like pure water
touching clear sky.

I like to think
(right now, please!)
of a cybernetic forest
filled with pines and electronics
where deer stroll peacefully
past computers
as if they were flowers
with spinning blossoms.

I like to think
(it has to be!)
of a cybernetic ecology
where we are free of our labors
and joined back to nature,
returned to our mammal
brothers and sisters,
and all watched over
by machines of loving grace.
100 1