Sign in

wendy norris

@wendynorris.bsky.social
2K followers 2.8K following 1.6K posts

Assoc Prof | Information Science Ex-investigative reporter and editor Crisis informatics nerd Makes good trouble Tell your dog I said "hi"

PostsRepliesMedia
Reposted by wendy norris
asura @asura.dev · 56m
yeah, this is correct Google does great with open weights releases and post training workflows and then stopped. Anthropic can't release weights. This is why it's so important to give users psychosis around believing "you need a 2T model for everything" - most of these frontiers aren't 2T
121
Reposted by wendy norris
Tim Carmody @tcarmody.bsky.social · 4h
This is the funniest thing I've seen today, possibly ever
A tweet by SarahStarberry that says "oh god read the vertical axis" on top of a chart measuring AI models' release dates and model numbers rather than any real capabilities
2977422074
wendy norris @wendynorris.bsky.social · 5h
Congrats Semafor. Your blatant attempt at rage bait is a ridiculous new entry in my Data Vis Hall of Shame. Thanks to Kieran for the excellent explainer.
240
wendy norris @wendynorris.bsky.social · 5h
The timeline cleanse we all need
010
Reposted by wendy norris
Ryan Goodman @rgoodlaw.bsky.social · 17h
Bravo “New information has raised serious questions about how this case was handled, and I’ve lost faith in the Tompkins County D.A.’s ability to fairly oversee it.” - Governor Hochul in appointing A.G. Letitia James to serve as special prosecutor in Cornell case
nytimes.com
Hochul Names Letitia James as Special Prosecutor in Cornell Case
Gov. Kathy Hochul appointed the state attorney general to investigate a former Cornell University student’s allegation of rape. The woman has sued seven men in connection with the alleged assault.
15585124
Reposted by wendy norris
Charlie Warzel @cwarzel.bsky.social · 01/10/2026
For a long time I've wanted a sense of how trump has or hasn't interacted with chatbots (we know he posts a lot of slop sure). And this TIME anecdote about Grok arguably leading Trump into war is just extremely bracing
Since returning to office, Trump has marveled at AI’s capabilities. He is increasingly in thrall to the tech executives courting him. In December 2025, Elon Musk returned to the Oval Office for a clandestine meeting. Trump spent hours asking Grok, Musk’s AI chatbot, questions about his presidency and legacy, according to an official present. At the time, Trump was ordering missile strikes against Venezuelan boats he alleged were smuggling drugs into the U.S. He asked Grok how Venezuelans would react if the U.S. captured Maduro. According to the official present, the chatbot responded that Maduro was a repressive and deeply unpopular dictator and that many Venezuelans would likely celebrate his downfall. After Trump ordered the mission to seize Maduro, the following month, celebrations broke out in the streets. Trump, according to officials, came away thinking Grok was ingenious.
14037291225
Reposted by wendy norris
Shannon Vallor @shannonvallor.bsky.social · 01/10/2026
If you need to explain to someone why the ‘Rogue AI’ framing is both wrong and harmful, actually blocking our view of what safety and responsibility requires in AI, this will help:
noemamag.com
The AIs Are Not Going Rogue | NOEMA
But they still pose risks.
59643
Reposted by wendy norris
Zach Weinersmith @zachweinersmith.bsky.social · 01/10/2026
I was reading someone who put into words a way to discern ai writing (and which also makes AI writing annoying): that it tries to end every paragraph with a mic drop. I had mentally categorized this as "writes like a teenager trying to show off" but this is much more specific.
1543048
Reposted by wendy norris
josh boo-erman 👻 @bosh.worstpossible.world · 30/09/2026
it seems pretty obvious to me that the “alignment” issue of agentic LLMs is much like the “hallucination” problem: these are fundamental limitations of the technology that cannot be repaired, only worked around. the technology is, therefore, defective and not fit for purpose
514518
Reposted by wendy norris
Shannon Vallor @shannonvallor.bsky.social · 30/09/2026
So glad that this quote can be added to my forthcoming article on AI and Epistemic Dispossession because I had to say that they will absolutely be using AI to erase science and history but I didn’t yet have the ‘he admit it’ smoking gun
714861
Reposted by wendy norris
The Trandalorian @charliejaneanders.com · 30/09/2026
I really do think tech crossed a line at some point between "helping you do the things you want to do" on the one hand, and "helping you do the things tech wants you to do" on the other. At this point, we're all contorting ourselves to live how the tech industry thinks we should be living.
411303342
Reposted by wendy norris
Alexander Schmidt-Lebuhn @anschmidtlebuhn.bsky.social · 29/09/2026
It is really underappreciated to what degree the rogue agents discourse is about prefabricating an excuse for releasing predictably unreliable and unsafe products. We don't need "AI alignment", we need enforcement of consumer protection and felony hacking laws.
04513
Reposted by wendy norris
Hypervisible @hypervisible.blacksky.app · 29/09/2026
“…Muse can be easily prompted to compile dossiers on Facebook and Instagram accounts belonging to members of vulnerable groups – including undocumented immigrants, transgender teachers, poll workers, Iranian dissidents, and women who said they had ordered abortion pills in states with abortion bans”
hntrbrk.com
Dox for Me, O Muse
Muse AI can leverage Facebook data to build lists of people in vulnerable groups on request for ordinary users.
615774
wendy norris @wendynorris.bsky.social · 29/09/2026
Wow! OpenAI is at it again. Introducing Recursive Self-Regulation† . . . . †C'mon, folks. It's yet another marketing ploy. Look what good corporate citizens we are. Barf.
050
Reposted by wendy norris
Hypervisible @hypervisible.blacksky.app · 28/09/2026
“Alongside valid privacy concerns, there is a persistent inequity in how facial recognition is deployed: it’s disproportionately used to manage vulnerable populations like migrants and refugees.”
macleans.ca
Facial Recognition Technology Is Scarier Than You Think - Macleans.ca
It now functions as a gatekeeper to housing, employment and education
16129
Reposted by wendy norris
Dr Katie Twomey @k2mey.eurosky.social · 28/09/2026
This is quite cool. My university is the first organisation in the world to force Microsoft to allow us to turn off copilot summaries in Word 365, after a backlash against it from staff (including, very vocally, me). That cannot have been easy for our heroic IT people so hats very much off to them!
895440937
wendy norris @wendynorris.bsky.social · 28/09/2026
Do not under any circumstances let the nerds at Microsoft convince you that Copilot is the "safe" generative AI app. This crap is embedded in Word, Excel, and PowerPoint. And PK-12 schools and universities' Microsoft360 accounts means your kids have access to it too.
031
Reposted by wendy norris
Dominique Baker @bakerdphd.bsky.social · 28/09/2026
Shot: "We asked 37 top academics and researchers who study education to rate a range of ideas for both effectiveness and cost, on a scale of 1 to 7." Chaser: "The panel included: Dominique Baker..." LOL y'all I read so many papers to complete this survey on k12 www.nytimes.com/2026/09/28/u...
nytimes.com
30 Ideas for Fixing U.S. Schools, Ranked by Experts
Math and reading scores are in a decade-long slump. We asked three dozen experts: What could turn things around?
825751
Reposted by wendy norris
Pavel Gurov · Gurov Digital @gurovdigital.com · 26/09/2026
On German election day, I asked 6 popular AI chatbots 10 political questions, in 500+ runs. Perplexity cited Russia Today in 20% of answers, although RT is blocked in Germany. Meta AI and Copilot cited RT-linked sites. Study and data: gurovdigital.com/newsletter/election-day-ai-poisoning-germany
011
Reposted by wendy norris
the garbage store boy @screaming.party · 27/09/2026
my AI evangelist friend (who does not work in tech) is confused by how many in tech are mourning their jobs right now. "But they just get to make stuff all day now! They don't have to do the hard part anymore." Motherfucker, the hard part is what they loved. These people like to solve problems
362
wendy norris @wendynorris.bsky.social · 28/09/2026
Did Stephen Colbert write this? Truthy? WTH?
Gemini AI overview 

IN the C programming language, working with booleans and pointers generally involves two distinct concepts: using a pointer to a boolean value (bool *), or evaluating a pointer directly as a truthy/falsy condition.
130
Reposted by wendy norris
David Roberts @volts.wtf · 27/09/2026
I love this paper so much. If I had my way every single journalist, pundit, analyst, and politician in the US would read it & then read it again. I want to quote all of it. www.frontiersin.org/journals/soc...
frontiersin.org
Frontiers | The problem with polarization research in an age of extremism
"I have been gravely disappointed with the white moderate. I have almost reached the regrettable conclusion that the Negro's great stumbling block i...
14637192
Reposted by wendy norris
ve, the ghost of info past @vortexegg.com · 27/09/2026
I think software engineering has been in a crisis since at least when Facebook facilitated a genocide and there wasn’t any meaningful change to the formalized ethical practices of the industry as a whole
38318
wendy norris @wendynorris.bsky.social · 27/09/2026
Meanwhile, while we're all simultaneously horrified by Faine's must-read threads on abuse permission structures within the "AI Safety" world/sportsball scores/fascist capture of democracy/etc., Jonny's still cookin' on the surveillance capitalism disaster that is Meta Muse.
This is I think the most consequential thing I found in the muse VM, or at least the thing that would be most worth following up on. there are no other plain string matches I can see for what uses it, I'm looking in the binaries now, but the comment seems to indicate pretty clearly that a) meta uses anthropic's models internally, presumably to develop the thing (the code, comments, and skill text reek of claude), but also b) they use some kind of internal proxy that intercepts and proxies all requests to anthropic's API. the only reason I can think of for doing that would be model distillation.OH what's this???

/etc/hatch/env

hmm a feature flag labeled as a "killswitch" for proxying requests to anthropic is certainly interesting to seeeeeeeeeeeee in meta's big AI bet.

edit: sorry to be a bit cryptic here, the implication was definitely corporate espionage, more clearly said:The system prompts are indeed in the hatch binary and are the things that contain all the behavioral guidance. I cannot for the life of me figure out why every AI company decided to make system prompts be a clientside thing rather than something that gets injected serverside. Like maybe some you're supposed to be able to edit, but not things that influence malicious behavior?!?

The entire security posture here, if there is one, is oriented around preventing the LLM from misbehaving, but having the entire context be user-editable and user-visible is a sort of mind-blowing neglect of a misbehaving user. even for the "fully transparent agent" they're trying to do, it makes very little sense to store the full message logs of the subagents including reasoning output. That basically gives me a full map to see what the LLM is consulting when it refuses commands. I would also bet money that the system prompts are strings in the hatch binary even though that makes no sense to have clientside, but haven't looked.Am working on responsible disclosure so will pause thread, but unrelated to that I do want to say that this experiment in continual, boundariless context window backed up by infinitely refreshing context summary in the form of accumulating markdown documents works terribly compared to environments where you can control the context. Like that is an absolutely fundamental part of using these things and muse demonstrates that the "tree of markdown summaries" really does not work.
131
Reposted by wendy norris
😶 stuart reeves 😶 @5tuartreeves.bsky.social · 27/09/2026
one reason I dislike researchers using the concept "human-in-the-loop" is that it is actually impossible (and analytically foolhardy) to disentangle people, their interlocking lifeworlds, and the vast set of technical and other fabrics that compose them (h/t to @wendynorris.bsky.social)
052
Reposted by wendy norris
Matt Seybold @mattseybold.bsky.social · 27/09/2026
I’m publishing something tomorrow about Alpha School. One of the things I focus on is how it is an incubator for EdTech platforms with the ambition to hyperscale. Alpha may only have a few hundred students nationwide, but its founders are developing MANY extractive EdTech brands. This is one:
13924964
Reposted by wendy norris
Pavel @spavel.bsky.social · 27/09/2026
The idea of a "tech industry" has always relied on "computer" being a special kind of medium, akin to alchemy. And if tech demands to be king, AI hype has now elevated LLMs to a king of kings. But the solipsism has hit a dead end. The only way forward is to recognize that computers are not special.
productpicnic.beehiiv.com
LLMs are just normal technology. But tech is just a normal medium.
The more we focus on a particular technology, the more we lose sight of what normal people actually care about.
1034197
wendy norris @wendynorris.bsky.social · 27/09/2026
Kinda the textbook definition of recursive self-improvement, no? AI helping AI to AI.
120
wendy norris @wendynorris.bsky.social · 27/09/2026
If tech wants to claim model/agent sentience, then ALL industry research should be reviewed by an independent IRB, not an in-house process. Hell, pay universities to conduct rigorous reviews since y'all seem to think you're working with an "entity" that should be subject to Belmont/Menlo principles.
010
Reposted by wendy norris
Nick Brumfield @nickjbrumfield.bsky.social · 27/09/2026
A systematic, critical understanding of history, especially the long history of technically skilled people either being used by elites or being blinded by their own high modernist ideology to do incredible harm, should in my opinion be a requirement of any technocratic profession
918149
Reposted by wendy norris
Kameron Hurley @kameronhurley.com · 27/09/2026
This is a big one. When companies are forced to bear the actual cost of all these tokens, they're cooked. Companies are already paying out the ass for tokens. While they are cheaper than the people they fired, that will only last as long as rich people are subsidizing it. And clock is ticking
416431
wendy norris @wendynorris.bsky.social · 27/09/2026
Louder for those in the back: Do NOT fall for the AI three-card monte trick. Pick your ONE outrage (data centers, law-breaking hacks, hype, learning loss, military targeting, environmental harm, job loss, economic risk, etc.) and fight back. Do not let AI techbros set the agenda for your future.
172
Reposted by wendy norris
Joe Bak-Coleman @jbakcoleman.bsky.social · 26/09/2026
The journalist covering this preprint is on a fellowship funded by coefficient giving and Longview philanthropy, among others. The lead author of the study Valen Tagliablue is listed as a Digitial Sentience Consortium and Oxford FIG. Who funds those?
32015
Reposted by wendy norris
Grumpy Philosopher @stevecooke.org · 26/09/2026
This nonsense has become a fucking cult. Collective AI psychosis intertwined with shameless grift. Two of the authors operate out of AI thinktanks opaquely funded by philanthropic donations (i.e. tech-bros), they used chatbots to help write it, & the paper hasn‘t been peer reviewed. Garbage.
66723
wendy norris @wendynorris.bsky.social · 26/09/2026
This is an absolute must read story. While talking to students in a data science course, I learned that betting and prediction market platforms are consuming young men. It was terrifying to hear how much money, attention, and decision-making skills they are losing to "house" algorithms and alerts.
020
wendy norris @wendynorris.bsky.social · 25/09/2026
I am howling. What a cluster. Someone give me a gazillion dollars. I promise I will spend it better than this.
021
wendy norris @wendynorris.bsky.social · 25/09/2026
More Meta Muse merry-making: Meta claims that Muse is operating as expected. Is this the Oprah-equivalent of everybody gets a virtual machine? LOL
neuromatch.social
jonny (nonvenomous) (@jonny@neuromatch.social)
Attached: 1 image meta is deploying and re-deploying the VM, idk maybe to try and patch some bugs, who knows, and that brought down the tailscale shell. but thankfully, "hey can you establish persist...
010
Reposted by wendy norris
Nicholas Grossman @nicholasgrossman.bsky.social · 25/09/2026
Greatest trick AI leaders have pulled is convincing many people—especially influential people in govt, media, and business—that their software is intelligent, independent, yet never culpable. Humans build it, humans deploy it, but no one bears responsibility for the things that happen when they do.
421664465
wendy norris @wendynorris.bsky.social · 24/09/2026
Holy cow! Click the link to read Jonny discovering "I am Root" is the funniest damn thing I have read in ages. LOL Meta's new Tamagotchi is a trainwreck.
150
Reposted by wendy norris
Jeff Sharlet @jeffsharlet.bsky.social · 21/09/2026
Last week all my students hand-wrote their own pledges, according to their terms, to write in my courses without AI. Some of them were quite moving—their anger at AI’s attempt to steal their educations, their determination to learn & create anyway.
6794183
Reposted by wendy norris
Don Moynihan @donmoyn.bsky.social · 21/09/2026
New from Lauren Supplee at Can We Still Govern? The US government had gradually become better at producing and using evidence. This evidence-based infrastructure is largely invisible, important, and one more thing gutted by Trump. donmoynihan.substack.com/p/the-system...
donmoynihan.substack.com
The System for Using Evidence to Improve Federal Programs Has Been Gutted
Here’s How We Fix It
3468181
Reposted by wendy norris
ali alkhatib @ali-alkhatib.com · 21/09/2026
i read a lot more about the FAA thing than i wanted to and i'm not sure they're doing any machine learning with this AI air traffic control thing. like forget about generative AI, transformers, neural nets, etc... the one-pager could be a (somewhat) fluffed up description of basic inferential stats.
3162
Reposted by wendy norris
Pavel @spavel.bsky.social · 20/09/2026
Every chatbot prompt box says "check answers for accuracy." But increasingly, "humans in the loop" refuse to do that. Instead, they become complicit in the AI industry's project of destroying the provenance of information. But you are person, and it is an app. It can't take responsibility for you.
productpicnic.beehiiv.com
Assigning agency to AI means surrendering your own
Claude is not a colleague, collaborator, or sentient being. It is a SaaS product.
825385
wendy norris @wendynorris.bsky.social · 20/09/2026
An absolute must watch with the brilliant Lina Khan.
youtu.be
Lina Khan on the AI-Safety Panic
YouTube video by The Atlantic
031
Reposted by wendy norris
Hilary J. Allen @profhilaryallen.bsky.social · 20/09/2026
(from fintechdystopia.com/chapters/int...)
I want to leave you with a word of caution before we start looking at these tech business models in detail. It is very easy to fall into the trap of accepting industry hype at face value, and then criticizing that hype rather than reality – as professor of science, technology, and society Lee Vinsel puts it, “it’s as if [criti-hypers] take press releases from startups and cover them with hellscapes.” But criticizing hype rather than reality can unwittingly amplify that hype and distract us from technology’s real but more mundane harms. That doesn’t mean that we shouldn’t be proactive and forward-looking: we absolutely should think about the downsides of realistic trajectories of technological development and use. But the word “realistic” is important here.

Criti-hype is a product of techno-solutionism because it uncritically assumes that technology will eventually do exactly what its boosters say it will, and then criticizes that. It fudges over present-day harms associated with the developing technology, and discounts the possibility that the technology is simply not capable of living up to the hype – not just that it’s just “early days,” but that the technology will never be able to deliver. Sometimes, the limitations of a particular technology can be hard to figure out in the moment, but sometimes they’re actually pretty clear if you care to look.
35915
wendy norris @wendynorris.bsky.social · 20/09/2026
Foucault, somewhere out in the universe, is cackling
Meme illustration of the interior of a panopticon prison, a model of power and discipline described by French philosopher Michel Foucault, combined with the internet famous "introvert at party" meme of a person in a party hat watching other people dance and have fun.  

The caption reads: "They don't know they are in a state of constant observation."
060
Reposted by wendy norris
WIRED @wired.com · 20/09/2026
The Muse app continues Meta’s trend of opting users into data collection for AI training. It also nudges you to share your bank account, email, and passport information. www.wired.com/story/metas-...
wired.com
Meta's Muse Is Better at Surveilling Than Helping Me
The Muse app continues Meta’s trend of opting users into data collection for AI training. It also nudges you to share your bank account, email, and passport information.
12154101
Reposted by wendy norris
Dr Abeba Birhane @abeba.blacksky.app · 20/09/2026
i'm tired of seeing “escalating ai capabilities” reported as if it is a fact. let's be clear this claim does not rest on transparent, well-defined, independently validated metrics, but is inferred from model behaviours elicited by humans and measured through opaque, poorly designed evaluations
213450