J. Arendt

@uwwoman

RVA librarian working at a local university and riding my bike there; opinions expressed are my own and not those of my employer or anyone else is the home of my blog / newsletter.

351 Siguiendo208 Seguidores443 Publicaciones

Instagram YouTube Bluesky

¿Es tu perfil?Reclámalo y uwwoman.dirty.social es tuyo.Reclamar perfil

J. Arendt@uwwoman· 17d

I'm used to university-provided news stories and press releases being mild, feel-good stories, but Christopher Whyte isn't doing that. news.vcu.edu/article/an-a...

What’s the takeaway here?
One lesson for AI companies and other organizations is the most likely failure around these frontier systems is socio-technical failure. AI safety cannot depend entirely on getting the model itself to behave properly – we also have to design the environment around these systems assuming that sometimes they will behave in ways we did not anticipate.

That’s obviously a pretty familiar cybersecurity problem, which makes this OpenAI failure a bit more damning than if the incident were truly without precedent. Organizations already use principles like least privilege, network segmentation, etc. because we assume that individual security controls will sometimes fail. These highly capable AI agents should have been treated in much the same way.

Going forward, models like these should only have access to the systems and information necessary for their tasks, and their ability to communicate outside controlled environments should be carefully constrained. Organizations need good visibility into what they are doing, so some version of forcing systems to explain themselves is critical.

J. Arendt@uwwoman· 24d

Generally I'm skeptical of this methodology. On the other hand, I like that the report describes how risks are borne by individuals, while the responsibility lies with AI creators and governors. It's a great comeback to, "Why do the kids hate AI?" cdn.prod.website-files.com/669550d38372...

Highlights
● Experts assessed at least 10% probability of catastrophic harm (e.g., more than 1 million
human deaths or more than USD $100B loss) from 18 of 24 AI risk domains under
business-as-usual trajectories over the next 5 years
● AI users and affected stakeholders were judged most vulnerable to AI risks; while generalpurpose AI developers, governments, regulators, and standards bodies were judged most
responsible for addressing AI risks

J. Arendt@uwwoman· 27d

I wonder what these economists think of protection rackets and lemon laws. www.nytimes.com/2026/07/16/t...

Excerpted from article: "While A.I.-assisted books received lower customer ratings than human-made ones, they deemed A.I.’s entry into the market a positive development, because the books were selling, if modestly. As economists, they told me, they’re less concerned with literary quality or customer satisfaction than revenue growth and market expansion."

J. Arendt@uwwoman· Jul 13

In counterpoint to the "teens these days aren't reading for pleasure anymore" narrative, it looks like the biggest drop is among elders. www.sciencedirect.com/science/arti...

Graph of percentage of U.S. teens (15 and over) and adults reading for pleasure, based on American Time Use survey data, from 2003 to 2023. For adults 66 and older, the percentage dropped from a bit over 50% to about a 33%. For young adults 15-24, the percentage was virtually unchanged. In the 25-65 age group, it dropped from about 25% to about 12%.

J. Arendt@uwwoman· Jul 10

The study was selective about the "human written" comparison texts. AI was easy to detect compared to published stories written by folks like "Joyce Carol Oates, Stephen King, Louis L'Amour, Charlotte Perkins, and Harlan Ellison." (Image from the Github page for the study)

Table of the genres of the source stories. The stories were drawn a little more than 2/3 from literary fiction and only 1/3 from genre fiction.

J. Arendt@uwwoman· Jul 9

Aside from Fs, students haven't been punished yet. The professor is working or butting heads with the university's Committee on Academic Code and the university's administrators. Their procedures and his patience for bureaucracy aren't built for cheating at scale. www.chronicle.com/article/the-...

I got no reaction from the provost — I continue to be completely ignored. The dean wrote back to say this is a wake-up call. He also said that I would need to submit for each individual student a separate memo. I said I’m not going to do that because it’s a huge amount of work. That is absolutely insane. In the preface of my memo to the Committee on the Academic Code, I said that I’m sending the case to them but they’re accustomed to dealing with individual cases or cases involving two or three students. I estimate about 50 students cheated in my class. So we need decisions from the top about how to deal with this. The committee never acknowledged that it received my case.

J. Arendt@uwwoman· Jun 20

A classic couple!

So I kindled the shavings; helped prop up the innocent little idol; offered him burnt biscuit with Queequeg; salamed before him twice or thrice; kissed his nose; and that done, we undressed and went to bed, at peace with our own consciences and all the world. But we did not go to sleep without some little chat.

How it is I know not; but there is no place like a bed for confidential disclosures between friends. Man and wife, they say, there open the very bottom of their souls to each other; and some old couples often lie and chat over old times till nearly morning. Thus, then, in our hearts’ honeymoon, lay I and Queequeg—a cosy, loving pair.

J. Arendt@uwwoman· Jun 17

Tesla didn't put out the report, so it's not quite the same as this description of AI companies. On the other hand, I laughed darkly at, "You're probably struggling to picture this scenario." www.nytimes.com/2026/06/17/o...

Imagine if the Ford Motor Company put out a report saying that it feared its popular F-150 trucks might soon start bursting into flames, but that there was nothing the company could do about it because automotive technology was too inevitable and important to slow down. You’re probably struggling to picture this scenario because no reasonable consumer product company would ever act like this.

J. Arendt@uwwoman· Jun 4

Did I read it right that the National PTA was offering astroturfing as a service? www.nytimes.com/2026/06/04/u...

"In a 2024 email pitching its services to Snap, the National PTA promised it could “help with sentiment” and create “more understanding and comfort” among parents. (Snap ultimately declined to offer funding.)

Exactly how much the National PTA has received from social media companies remains secret, but some details emerged in the documents. In 2024, a National PTA official told Snap executives that companies generally paid the organization $250,000 to $500,000 a year, and that a handful gave millions of dollars a year."

J. Arendt@uwwoman· May 27

This change was in application essays. Perhaps the students had already learned that homogenization could help their chances of getting in. www.nytimes.com/2026/05/27/o...

Even more problematic, Dr. Green’s research shows that A.I. has the largest homogenizing impact on students who are farthest from the mean and have unique perspectives, including neurodivergent students and those from racial and linguistic minorities.

J. Arendt@uwwoman· Apr 28

Periodically I go into a generative AI tool to see how much better it has gotten since the last time I checked. I keep reading that Gen AI is getting better and better and keep being disappointed when I check for myself.

Incomplete thoughts and an unreadable word in part of a NotebookLM infographic output.

J. Arendt@uwwoman· Apr 27

Ken's piece isn't behind a paywall. Just click on the link to find your answer.

A multi-sentence example of Ken quoting the (alleged) manifesto

J. Arendt@uwwoman· Apr 9

You might also be interested in some related RVA street art.

From a paper plastered to a lamppost: Photoshopped images of an Insane Clown Posse member asking how door handles work, while a Tesla is on fire.

J. Arendt@uwwoman· Apr 8

Why are you showing the study 1 results, when your method there excluded the worst fraction solvers from the control arm but not the AI arm? Study 2 is the better of the two studies.

The less dramatic results from Study 2, still showing (slightly) lower performance on the 3-question test for the AI participants than the control participants but no difference in skip rate.

J. Arendt@uwwoman· Apr 2

@gallup.com has confused me. Why would you weigh a nonprobability Web panel survey of college students to match the demographics of the U.S. population age 18-59 w/high school but not a college degree? Wouldn't you want to match the demographics of college students? www.gallup.com/file/analyti...

Methodology
Results for the Lumina-Gallup study are based on web surveys conducted Oct. 2-31, 2025, with a sample of 3,801 adults who are currently pursuing an associate or bachelor’s degree. All respondents were between the ages of 18 and 59 and had a high school diploma or equivalent but not an associate or bachelor’s degree. Respondents were interviewed via Dynata’s nonprobability web-based panel.
The data are weighted to match national demographics of gender, age, race, Hispanic ethnicity, education and region for the population of U.S. adults aged 18 to 59 with a high school diploma but without a college degree. Demographic weighting targets are based on the most recent American Community Survey figures.

J. Arendt@uwwoman· Mar 30

It's hard for me to infer much about this association when the use is so low. There's no decimal point on the left graph but there is on the right, and both graphs are expressing percent.

Based on Comscore data, the graph on the left shows that by Q4 of 2024, over 15% of households showed that ChatGPT had ever been used on the home computer. The graph on the right shows that about .4% of browsing duration in Q4 2024 on the home computer was on ChatGPT.

J. Arendt@uwwoman· Mar 30

It really is remarkable.

From data center coloring book: A child, viewed from behind, peddling a Big Wheel down a long hallway, with doors on either side, much like a scene in The Shining

J. Arendt@uwwoman· Mar 27

Aside from profiling someone who should have "content repackager" as his job title, this article covers the scope of the enslopification of the internet. www.wsj.com/business/med...

Much of the content people now consume online is generated by artificial intelligence, with some 9% of newly published newspaper articles either partially or fully AI-generated, according to a 2025 study led by the University of Maryland. The number of AI-generated articles on the web surpassed human-written ones in late 2024, according to research and marketing agency Graphite.

J. Arendt@uwwoman· Feb 12

And, it turns out not only are they using income as an indicator of cognitive enrichment for their composite, they are following in the footsteps of Oveisgharan, Wilson, Yu, Schneider, and Bennett, who used childhood socioeconomic status as an indicator of cognitive enrichment.

Three paragraphs describing the cognitive enrichment indicators, with key phrases being "Early-life enrichment included SES..." "Midlife cognitive enrichment included...income level at age 40..." and "late-life enrichment included...all sources of income at study enrollment..."

J. Arendt@uwwoman· Feb 3

Yikes! The administrators will be delighted when the D/W/F rate goes through the floor after this change.

Administrators point to evidence that students may learn better in an accelerated format. Old Dominion already offers 22 programs and certificates in the eight-week asynchronous model and reports that the rates of their students earning D’s, F’s, or withdrawals are 2.4 percent lower in asynchronous courses than in in-person ones, and 5 percent lower in eight-week asynchronous online courses than in 16-week in-person ones.

J. Arendt@uwwoman· Jan 29

Wow.

Hanover County voted 62% Trump, 36% Harris.

J. Arendt@uwwoman· Jan 25

When the NY Post has reporting about pumping rounds into a motionless body, it's bad. nypost.com/2026/01/24/u...

From the NY Post, "In various videos,the agent who fired the fatal shots could be seen shooting the armed protester at close range, then pumping at least 9 more rounds into his body as Pretti lay motionless on the ground."

J. Arendt@uwwoman· Jan 21

What ringing endorsements of generative AI from "pro-AI" faculty! dgmg81phhvh63.cloudfront.net/content/user...

PRO-AI VIEWS
AMONG FACULTY
“I have used it to
generate bad writing
samples and bad
computer code
for students to
practice fixing.”
“I like to write so I
wouldn’t use it for a
research paper. But
I’d consider using it to
create any of the endless
and mindless reports
that our admin wants
us to create.”

J. Arendt@uwwoman· Jan 20

Does anyone know of a reason to distrust last year's University of Melbourne / KPMG survey that AI use was ALREADY high outside wealthy countries, at least in countries where it was plausible that people had good enough internet access to use it in the first place? assets.kpmg.com/content/dam/...

There are notable differences across
countries in people’s adoption of AI, with
emerging economies leading the way
There is a distinct pattern of findings between
countries with advanced and emerging
economies, with the use of AI tools notably
higher in countries with emerging economies.
On average, four in five (80%) people in emerging
economies intentionally use AI tools on a regular
or semi-regular basis, compared to three in five
(58%) in advanced economies.
As shown in Figure 2, levels of AI use in most
emerging economies exceed 70 percent of the
population, with India and Nigeria reporting the
highest regular or semi-regular usage (92%). Two
emerging economies located in Eastern Europe—
Hungary and Romania—have notably lower AI
use compared to the other emerging economies.
In contrast, AI use levels in most advanced
economies fall below 70 percent of the population,
with the lowest usage reported in the Netherlands
(43%) and the highest in Singapore (73%).

J. Arendt@uwwoman· Jan 14

From The Masque of the Red Death "In an assembly of phantasms such as I have painted, it may well be supposed that no ordinary appearance could have excited such sensation. In truth the masquerade licence of the night was nearly unlimited; but the figure in question had out-Heroded Herod, and gone beyond the bounds of even the prince’s indefinite decorum. There are chords in the hearts of the most reckless which cannot be touched without emotion. Even with the utterly lost, to whom life and death are equally jests, there are matters of which no jest can be made. The whole company, indeed, seemed now deeply to feel that in the costume and bearing of the stranger neither wit nor propriety existed. The figure was tall and gaunt, and shrouded from head to foot in the habiliments of the grave. The mask which concealed the visage was made so nearly to resemble the countenance of a stiffened corpse that the closest scrutiny must have had difficulty in detecting the cheat. And yet all this might have been endured, if not approved, by the mad revellers around. But the mummer had gone so far as to assume the type of the Red Death. His vesture was dabbled in blood—and his broad brow, with all the features of the face, was besprinkled with the scarlet horror."

J. Arendt@uwwoman· Jan 8

Will the successor to "slop" for word of the year be "farts"? spectrum.ieee.org/ai-coding-de...

Article with intro that newer AI models are more prone to "silent but deadly" failures.

J. Arendt@uwwoman· Nov 19

Where have I heard that take before?

Page from the 2020 afterward to paperback edition of Niedermeyer's book, "Ludicrous: The Unvarnished Story of Tesla Motors." One quotation is, "When, if ever, will the fantasy falter and reality reimpose itself?"

J. Arendt@uwwoman· Nov 12

Relatedly, the first half of this recommendation sounds like it could bring in which high school the student attended to decide what math class the student can take in the first semester of college. www.insidehighered.com/news/quick-t...

The working group put forward a number of recommendations for addressing these shortcomings, including using a “math index” based on historical placement data and transcript-based variables to “predict students’ likelihood of placement into remedial math.” The group also recommended establishing feedback mechanisms with high schools and requiring math placement testing by June 1 for incoming students, among other things.

J. Arendt@uwwoman· Nov 4

Am I the only one who looked him up on Wikipedia and then clicked on his wife's name to see if there was a reason for her to have her own entry, aside from being his wife?

Another person just discovering his wife went to VCU

J. Arendt@uwwoman· Sep 19

How?