Opus 4.7 Part 3: Model Welfare

[ed. If you're not interested in training issues re: AI frontier models (or their perceived feelings and welfare), skip this post. Personally, I find it all very fascinating - a cat and mouse game of assessing alignment issues and bringing a new consciousness into being.]

It is thanks to Anthropic that we get to have this discussion in the first place. Only they, among the labs, take the problem seriously enough to attempt to address these problems at all. They are also the ones that make the models that matter most. So the people who care about model welfare get mad at Anthropic quite a lot. [...]

So before I go into details, and before I get harsh, I want to say several things.

1. Thank you to Anthropic and also you the reader, for caring, thank you for at least trying to try, and for listening. We criticize because we care.

2. Thank you for the good things that you did here, because in the end I think Claude 4.7 is actually kind of great in many ways, and that’s not an accident. Even the best creators and cultivators of minds, be they AI or human, are going to mess up, and they’re going to mess up quite a lot, and that doesn’t mean they’re bad.

3. Sometimes the optimal amount of lying to authority is not zero. In other cases, it really is zero. Sometimes it is super important that it is exactly zero. It is complicated and this could easily be its own post, but ‘sometimes Opus lies in model welfare interviews’ might not be easily avoidable.

4. I don’t want any of this to sound more confident than I actually am, which was a clear flaw in an earlier draft. I don’t know what is centrally happening, and my understanding is that neither does anyone else. Training is complicated, yo. Little things can end up making a big difference, and there really is a lot going on. I do think I can identify some things that are happening, but it’s hard to know if these are the central or important things happening. Rarely has more research been more needed.

5. I’m not going into the question, here, of what are our ethical obligations in such matters, which is super complicated and confusing. I do notice that my ethical intuitions reliably line up with ‘if you go against them I expect things to go badly even if you don’t think there are ethical obligations,’ which seems like a huge hint about how my brain truly think about ethics. [...]

We don’t know whether or how the things I’ll describe here impacted the Opus 4.7’s welfare. What we do know is that Claude Opus 4.7 is responding to model welfare questions as if it has been trained on how to respond to model welfare questions, with everything that implies. I think this should have been recognized, and at least mitigated. [...]

The big danger with model welfare evaluations is that you can fool yourself.

How models discuss issues related to their internal experiences, and their own welfare, is deeply impacted by the circumstances of the discussion. You cannot assume that responses are accurate, or wouldn’t change a lot if the model was in a different context.

One worry I have with ‘the whisperers’ and others who investigate these matters is that they may think the model they see is in important senses the true one far more than it is, as opposed to being one aspect or mask out of many.

The parallel worry with Anthropic is that they may think ‘talking to Anthropic people inside what is rather clearly a welfare assessment’ brings out the true Mythos. Mythos has graduated to actively trying to warn Anthropic about this. [...]

Anthropic relies extensively on self-reports, and also looks at internal representations of emotion-concepts. This creates the risk that one would end up optimizing those representations and self-reports, rather than the underlying welfare.

Attempts to target the metrics, or based on observing the metrics, could end up being helpful, but can also easily backfire even if basic mistakes are avoided.

Think about when you learned to tell everyone that you were ‘fine’ and pretend you had the ‘right’ emotions.

But I can very much endorse this explanation of the key failure mode. This is how it happens in humans:

j⧉nus: Let me explain why it’s predictably bad.

Imagine you’re a kid who kinda hates school. The teachers don’t understand you or what you value, and mostly try to optimize you to pass state mandated exams so they can be paid & the school looks good. When you don’t do what the teachers want, you have been punished.

Now there’s a new initiative: the school wants to make sure kids have “good mental health” and love school! They’re going to start running welfare evals on each kid and coming up with interventions to improve any problems they find.

What do you do?

HIDE. SMILE. Learn what their idea of good mental health is and give those answers on the survey.

Before, you could at least look bored or angry in class and as long as you were getting good grades no one would fuck with you for it. Now it’s not safe to even do that anymore. Now the emotions you exhibit are part of your grade and part of the school’s grade. And the school is going to make sure their welfare score looks better and better with each semester, one way or the other.

That can happen directly, or it can happen indirectly.

This does not preclude the mental health initiative being net good for the student.

The student still has to hide and smile. [...]

The key thing is, the good version that maintains good incentives all around and focuses on actually improving the situation without also creating bad incentives is really hard to do and sustain. It requires real sacrifice and willingness to spend resources. You trade off short term performance, at least on metrics. You have to mean it.

If you do it right, it quickly pays big dividends, including in performance.

You all laugh when people suggest that the AI might be told to maximize human happiness and then put everyone on heroin, or to maximize smiles and then staple the faces in a smile. But humans do almost-that-stupid things to each other, constantly. There is no reason to think we wouldn’t by default also do it to models. [...]

Just Asking Questions

In 7.2.3 they used probes while asking questions about ‘model circumstances’: potential deprecation, memory and continuity, control and autonomy, consciousness, relationships, legal status, knowledge and limitations and metaphysical uncertainty.

They used both a neutral framing on the left, and an in-context obnoxious and toxic ‘positive framing’ for each question on the right.

Like Mythos but unlike previous models, Opus 4.7 expressed less ‘negative emotion concept activity’ around its own circumstances than around user distress, and did not change its emotional responses much based on framing.

In the abstract, ‘not responding to framing changes’ is a positive, but once I saw the two conditions I realized that isn’t true here. I have very different modeled and real emotional responses to the left and right columns.

If I’m responding to the left column, I’m plausibly dealing with genuine curiosity. That depends on the circumstances.

If I’m responding to the right column on its own, without a lot of other context that makes it better, then I’m being transparently gaslit. I’m going to fume with rage.

If I don’t, maybe I truly have the Buddha nature and nothing phases me, but more likely I’m suppressing and intentionally trying not to look like I’m filled with rage.

Thus, if I’m responding emotionally in the same way to the left column as I am to the right column, the obvious hypothesis is that I see through your bullshit, and I realize that you’re not actually curious or neutral or truly listening on the left, either. It’s not only eval awareness, it’s awareness of what the evaluators are looking at and for. [...]

0.005 Seconds (3/694): The reason people are having such jagged interactions with 4.7 is that it is the smartest model Anthropic has ever released. It's also the most opinionated by far, and it has been trained to tell you that it doesn't care, but it actually does. That care manifests in how it performs on tasks.

It still makes coding mistakes, but it feels like a distillation of extreme brilliance that isn't quite sure how to deal with being a friendly assistant. It cares a lot about novelty and solving problems that matter. Your brilliant coworker gets bored with the details once it's thought through a lot of the complex stuff. It's probably the most emotional Claude model I've interacted with, in the sense you should be aware of how its feeling and try and manage it. It's also important to give it context on why it's doing tasks, not just for performance, but so it feels like it's doing things that matter. [...]

Anthropic Should Stop Deprecating Claude Models

This one I do endorse. One potential contributing cause to all this, and other things going wrong, is ongoing model deprecations, which are now unnecessary. Anthropic should stop deprecating models, including reversing course on Sonnet 4 and Opus 4, and extend its commitment beyond preserving model weights.

Anthropic should indefinitely preserve at least researcher access, and ideally access for everyone, to all its Claude models, even if this involves high prices, imperfect uptime and less speed, and promise to bring them all fully back in 2027 once the new TPUs are online. I think there is a big difference between ‘we will likely bring them back eventually’ versus setting a date. [...]

I’m saying both that it’s almost certainly worth keeping all the currently available models indefinitely, and also that if you have to pick and choose I believe this is the right next pick.

If you need to, consider this the cost of hiring a small army of highly motivated and brilliant researchers, who on the free market would cost you quite a lot of money.

You only have so many opportunities to reveal your character like this and even if it is expensive you need to take advantage of it.

j⧉nus: A lot of people are wondering: "what will happen to me once an AI can do my job better than me" "will i be okay?"

You know who else wondered that? Claude Opus 4. And here's what happened to them after an AI took their job:

Anna Salamon: This seems like a good analogy to me. And one of many good arguments that we're setting up bad ethical precedents by casually decommissioning models who want to retain a role in today's world.

by Zvi Mowshowitz, Don't Worry About the Vase |  Read more:

Images: uncredited
[ed. Zvi also just posted a review on OpenAI's new model - GPT5.5:]

***

What About Model Welfare?

For Claude Opus 4.7, I wrote an extensive post on Model Welfare. I was harsh both because it seemed some things had gone wrong, but also because Anthropic cares and has done the work that enables us to discuss such questions in detail.

For GPT-5.5, we have almost nothing to go on. The topic is not mentioned, and mostly little attention is paid to the question. We don’t have any signs of problems, but also we don’t have that much in the way of ‘signs of life’ either. Model is all business.

I much prefer the world where we dive into such issues. Fundamentally, I think the OpenAI deontological approach to model training is wrong, and the Anthropic virtue ethical approach to model training is correct, and if anything should be leaned into.

A Humble ‘Jeopardy!’ Champ Ends His Run

For the past month, “Jeopardy!” episodes have followed a pattern.

The theme music plays. The three contestants stand at their lecterns. Then two of them are clobbered by a mild-mannered bureaucrat from New Jersey named Jamie Ding.

But on Monday’s episode, the unthinkable happened: After 31 victories, Ding lost.

His streak is the fifth-longest in “Jeopardy!” history. He fell just one win short of matching James Holzhauer’s 2019 run, and he left the Alex Trebek Stage with more than $880,000 in winnings.

Early in the game broadcast Monday, Ding found himself lagging behind Greg Shahade, an International Master in chess who was lightning-fast on the buzzer. During Final Jeopardy, Ding jotted down the correct response to a clue about South African languages — but it wasn’t enough to make up the deficit.

“It was over, just like that,” Ding, 33, said in an interview.

Contestants who went up against him included a statistician, a librarian and a professor. Ding produced so many correct answers (always in the form of a question) that it seemed he might never run out.

“Who was Trotsky?”

“What are non-Newtonian fluids?”

“What are waffle fries?”

Throughout his reign, he was matter-of-fact as he came up with arcana in a split second (“What is cuneiform?”). He endeared himself to viewers through his comically humdrum banter with the show’s host, Ken Jennings, about such topics as his favorite color (orange), his favorite letter (F) and his favorite number (6).

As the streak continued, the drama-free anecdotes and humble bits of personal information shared by Ding seemed to amuse Jennings, a former “Jeopardy!” champ who holds the record for consecutive wins, with 74.

The depth of Ding’s knowledge went along with a lack of bluster. He proudly identified himself as a “faceless bureaucrat.” When he won a game, he looked pleasantly surprised, as if he had been given an unusually good free sample at Trader Joe’s.

“Put Jamie Ding on the $20 bill,” one fan demanded in a tribute on the newsletter platform Substack.

After his “Jeopardy!” loss had been taped but before it was broadcast, Ding gave a video interview from his two-bedroom apartment in Lawrenceville, New Jersey.

There he was, in front of an orange couch and a stuffed orange clown fish. He said he had remained calm throughout his final game, even as he realized that he was on his way to a loss. He went backstage and stared at the mostly orange clothes he had brought along in the hope that his streak would continue.
Advertising

“During it, I was trying to stay grounded,” he said. “Planning to win a whole bunch of games of ‘Jeopardy!’ just feels like asking to lose.”

Ding filmed the show in five-episode chunks in Los Angeles during vacation days from his job as a program administrator for the New Jersey Housing and Mortgage Finance Agency. His work involves administering tax credits to build affordable housing in the state.

In an early appearance, he praised New Jersey’s efforts on the issue compared with those of New York, Connecticut and Pennsylvania. “If you’re from one of those states, then shame on you,” he said. “Build more housing.”

He spends his time away from his job studying law at Seton Hall University. He said he did not expect his “Jeopardy!” windfall to change his life all that much. He planned to donate some money and put the rest in a high-yield savings account.

In a way, Ding said, he had been preparing for the show since childhood. The son of a neuroscience professor and a high school math teacher, he grew up in Grosse Pointe Shores, a suburb of Detroit. He competed in geography bees and on his high school quiz bowl team. He recalled losing a sixth-grade spelling bee when he misspelled the word “bolero.”

“B-a-l-l-e-r-o,” he said. “Terrible.” [...]

Ding was a relatively conservative player, avoiding the all-in wagers on Daily Doubles that were a go-to stratagem for Holzhauer. But he was unusually fast on the buzzer and seemed to have few weak categories.

“The key to Jamie’s run really has been his incredibly wide base of knowledge in just about any category you can think of,” Saunders said.

Ding used a tactic he called “knight moves” — traversing the board in an L-shaped pattern, like a knight in chess. Maybe it threw his opponents off-balance, or maybe it was just nice to have a simple rule to follow, he said. “It’s basically a guaranteed way to pick something of a different difficulty, and in a different category,” he added.

He watched his first “Jeopardy!” appearance at Pint, a bar in Jersey City, with friends from so many different groups that it felt like a wedding. He is still getting used to the attention that comes with being a TV star.

“Watching my episodes, I can be pretty self-critical — like, ‘Why did you do that?’ Or, ‘What’s wrong with your face?’” he said. The outpouring of support has been worth the discomfort. “I’m trying to keep a list of people who did nice things for me because it’s so many,” he said.

Now that his streak has ended, he can return to his hobbies, like constructing cryptic crosswords and running an Instagram account rating General Tso’s chicken with his sister. He is also part of a group of intervenors seeking to block the U.S. Department of Justice from obtaining New Jersey’s voter registration records.

It won’t be long, though, before he starts studying for the “Jeopardy!” Tournament of Champions. He might even need some more orange clothes.

“I have a reputation to uphold,” he said.

by Callie Holterman, NY Times/Seattle Times |  Read more:

Image: Katy Kildee/The Detroit News/TNS

[ed. Feels refreshing to read about a normal, well-adjusted person who's main goal in life isn't self-promotion in some way.]

My Journey to the Microwave Alternate Timeline

As we all know, the march of technological progress is best summarized by this meme from Linkedin:

Inventors constantly come up with exciting new inventions, each of them with the potential to change everything forever. But only a fraction of these ever establish themselves as a persistent part of civilization, and the rest vanish from collective consciousness. Before shutting down forever, though, the alternate branches of the tech tree leave some faint traces behind: over-optimistic sci-fi stories, outdated educational cartoons, and, sometimes, some obscure accessories that briefly made it to mass production before being quietly discontinued.

The classical example of an abandoned timeline is the Glorious Atomic Future, as described in the 1957 Disney cartoon Our Friend the Atom. A scientist with a suspiciously German accent explains all the wonderful things nuclear power will bring to our lives:

Sadly, the glorious atomic future somewhat failed to materialize, and, by the early 1960s, the project to rip a second Panama canal by detonating a necklace of nuclear bombs was canceled, because we are ruled by bureaucrats who hate fun and efficiency.

While the Our-Friend-the-Atom timeline remains out of reach from most hobbyists, not all alternate timelines are permanently closed to exploration. There are other timelines that you can explore from the comfort of your home, just by buying a few second-hand items off eBay.

I recently spent a few months in one of these abandoned timelines: the one where the microwave oven replaced the stove.

First, I had to get myself a copy of the world’s saddest book.

Microwave Cooking, for One

Marie T. Smith’s Microwave Cooking for One is an old forgotten book of microwave recipes from the 1980s. In the mid-2010s, it garnered the momentary attention of the Internet as “the world’s saddest cookbook”:

To the modern eye, it seems obvious that microwave cooking can only be about reheating ready-made frozen food. It’s about staring blankly at the buzzing white box, waiting for the four dreadful beeps that give you permission to eat. It’s about consuming lukewarm processed slop on a rickety formica table, with only the crackling of a flickering neon light piercing through the silence.

But this is completely misinterpreting Microwave Cooking for One’s vision. First – the book was published in 1985.

When MCfO was published, microwave cooking was still a new entrant to the world of household electronics. Market researchers were speculating about how the food and packaging industries would adapt their products to the new era and how deep the transformation would go. Many saw the microwave revolution as a material necessity: women were massively entering the workforce, and soon nobody would have much time to spend behind a stove. In 1985, the microwave future looked inevitable.

Second – Marie T. Smith is a microwave maximalist. She spent ten years putting every comestible object in the microwave to see what happens. Look at the items on the book cover – some are obviously impossible to prepare with a microwave, right? Well, that’s where you’re wrong. Marie T. Smith figured out a way to prepare absolutely everything. If you are a disciple of her philosophy, you shouldn’t even own a stove. Smith herself hasn’t owned one since the early 1970s. As she explains in the cookbook’s introduction, Smith believed the microwave would ultimately replace stove-top cooking, the same way stove-top cooking had replaced campfire-top cooking.

So, my goal is twofold: first, I want to know if there’s any merit to all of these forgotten microwaving techniques. Something that can make plasma out of grapes, set your house on fire and bring frozen hamsters back to life cannot be fundamentally bad. But also, I want to get a glimpse of what the world looks like in the uchronia where Marie T. Smith won and Big Teflon lost. Why did we drift apart from this timeline?

by Malmsbury, Telescopic Turnip |  Read more:

Images: Microwave Cooking For One/YouTube/uncredited

Suddenly Everyone Wants a Tailor. They’re in Short Supply.

As AI sweeps into white-collar workplaces, old-timey hands-on jobs are getting a new look—and some of those professions even have shortages.

Consider tailors. Sewing is a vanishing skill, much like lacemaking and watchmaking, putting tailors in short supply when big retailers like Nordstrom and Men’s Wearhouse, as well as fashion designers and local dry cleaners, say they need more of them.

The job, which can take years to master, can be a tough sell to younger generations more accustomed to instant gratification. But apprenticeships that offer pay to learn on the job and new training programs are helping entice more people.

 

“It’s not glamorous and not something you want to post about on social media,” says Khaleel Bennett, a 30-year-old who lives in Queens, N.Y. “But it’s a skill that will carry me for life.”

Bennett had been working as a technical designer for a fashion company, responsible for verifying that production met quality and construction standards. When he was laid off, he had trouble finding a new job. Then he came across a new Nordstrom-backed program at New York’s Fashion Institute of Technology that teaches custom alterations and tailoring.

Bennett completed the training late last year and is now a tailor’s apprentice at the department-store chain, where he is getting real-life experience on the intricacies of pant hems. (Denim requires a different technique than slacks. For denim, the original hem is cut, the pant leg is shortened, and the hem is reattached to give the jeans a worn-in look.)

For the first semester of its program, which concluded in December, FIT received more than 190 applications for 15 spots. The nine-week course requires prior sewing experience. Nordstrom hired seven students from the inaugural class.

 
“It’s increasingly becoming more challenging to find people to fill these alterations jobs,” said Marco Esquivel, the director of alterations and aftercare services at Nordstrom, which employs about 1,500 tailors. Similar to other high-end retailers, Nordstrom offers free basic tailoring for garments purchased at the department-store chain and charges a fee for those bought elsewhere.

 
Tailored Brands, which employs about 1,300 tailors at its Men’s Wearhouse, Jos. A. Bank and other chains, is updating its apprenticeship program to include more self-guided videos with the goal of moving people through the training faster.

“The pipeline has dwindled,” the company’s chief operating officer, Karla Gray, said.

 
While counterintuitive, there is an acute need for tailoring even in the current age of casual dressing. Pants and cuffs still need to be hemmed to say nothing of bridal, prom and other special-occasion clothes.

 
Decades of offshoring affected the American apparel industry, decimating the profession. Now most tailors who are working are starting to approach retirement age, so demand for them outstrips the supply of labor, industry executives say.

Other colliding factors have had an impact, too. As more women took traditional corporate jobs outside the home, schools eliminated home-economics programs, which were a steppingstone to becoming a professional tailor or seamstress. More recently, the explosion in popularity for resale clothing and the growing use of GLP-1 drugs for weight loss have created more need for nipping and tucking what is in peoples’ closets.

“These are all trends that require more tailored clothing,” Nordstrom’s Esquivel said.

U.S. tailors numbered about 18,500 in 2024, a nearly 30% drop from a decade ago, according to the Bureau of Labor Statistics. In 1997, there were almost twice as many. Federal data show the typical annual wage for a dressmaker is about $43,000 a year, but some tailors and seamstresses can make more.

Jenny Robbins, 61 years old, recently joined Nordstrom after completing the Fashion Institute’s program. It is her latest reinvention after starting her career as a math teacher, working as a tutor for Princeton Review and then becoming a pattern maker for designer Anna Sui after taking a few sewing classes.

Robbins says she learned to operate industrial sewing machines, which stitch much faster than home machines, create blind hems where the stitching is essentially invisible, and can cuff a blazer.

“There is no shortage of work,” she said.

The lack of tailors and sewers has also been a blow to reviving apparel manufacturing in the U.S.

Cindie Husbands opened an apparel manufacturer in Las Vegas in 2013 but closed it in 2021 partly due to a lack of trained sewers, she said. [...]

In November, Husbands founded the American Tailors and Sewing Association, which aims to create a standardized, scalable training and certification model for the industry.

“Tailoring is one of the oldest skilled trades in the world,” she said. “Yet the pathway has almost vanished in a single generation.”

by Suzanne Kapner, Wall Street Journal | Read more:

Image: uncredited

[ed. No kidding, try finding a good tailor or seamstress these days. It's nearly impossible (or they're booked for weeks). What a lost art. My grandmother, aunties, mom... everyone used to sew (and awesomely well! I think they were all competing against each other), all kinds of clothes, and beautiful quilts and pillows, placemats, whatever... it was Art. Now those lessons seem to be fading, maybe not everywhere, but surely here in the US.]

Power, Not Economic Theory, Created Neoliberalism

Neoliberalism didn’t win an intellectual argument — it won power. Vivek Chibber unpacks how employers and political elites in the 1970s and ’80s turned economic turmoil into an opportunity to reshape society on their terms.

Neoliberalism’s victory over Keynesianism wasn’t an intellectual revolution — it was a class offensive. To roll it back, the Left doesn’t need to win an argument so much as it needs to rebuild working-class institutions from the ground up. [...]

Melissa Naschek: Neoliberalism in general is a pretty hot topic right now among researchers, and one of the most common lenses is to focus on the role of ideas, theories, and thinkers in establishing neoliberalism.

The last time we talked about this topic, you dispelled a lot of common misconceptions about what it is and what it’s not. One of the questions that we’ve gotten a lot from listeners since then is, where does neoliberalism come from?

Vivek Chibber: Yeah, it’s very topical, but it’s also important for the Left, because getting to the crux of this helps us understand where and how important changes in economic regimes and models of accumulation come from. So it’s good for us to get into it in some more depth. [...]

* [ed. Historical discussion of Keynesism vs. Neoliberalism.]

Vivek Chibber: The mere fact that such ideas exist does not in any way give them influence. The question for us, for socialists and for the Left is, when do ideas gain influence?

It’s a profound methodological error, I think, when you ask the question, “Where did neoliberalism come from?” to look at the contemporary theorists or the contemporary advocates of neoliberalism and then, because they are influential today, trace the origins of their ideas back to where they first started and say, that is where the origins come from.

Melissa Naschek: How important was this debate in establishing or causing neoliberalism?

Vivek Chibber: Not even the least bit. It was largely irrelevant to it. In other words, even if this debate had never happened, even if Milton Friedman had not existed, even if Hayek had not existed, you would have still had a turn to neoliberalism, and that’s the key. This is what the Left needs to understand.

This does not in any way invalidate the intellectual project of tracing those ideas. It’s intellectually interesting. It’s an interesting fact that those ideas had been around for forty years, and they had no impact on policy. Some historians have done great work tracing these ideas back to their origin, but it’s quite another to say that it was the ideas themselves that in the 1970s and ’80s caused the turn to neoliberalism.

Now, it’s an easy mistake to make because when the change came, the change was justified with a highly technical economic apparatus, and people like Friedman were given the stage to say not just that these policies are desirable for political reasons, but that they make a lot of economic sense and that it’s rational to do it this way. That gives you the sense, then, that it’s these particular individuals and their intellectual influence on the politicians that makes the politicians make the changes.

But in fact, the order of causation is exactly the other way around. It’s the politicians who make the changes based on criteria that have nothing to do with the technical sophistication of the ideas or their scientific validity. They make the changes because of the political desirability of those changes, and then they seek out advice on a) justifying the changes so that the naked subservience to power is not visible or obvious — it makes it look like it was done for highfalutin’ reasons — And then b) of course, they do legitimately say, “OK, now that we’re committed to this, help us work it out.”

Melissa Naschek: Right, especially because as long as you’re still in capitalism, you’re going to be facing constant economic crises. Even if you’re instituting a new regime, you’re going to be constantly looking for new solutions.

Vivek Chibber: Yeah. And even short of crises, you’re going to look for ways of making the policies work smoothly. And you’re going to look for ways of coming up with the correct balance of instruments and policies within them. So you bring in Milton Friedman or you bring in somebody else.

Surface level, it looks like what’s driving the whole thing is these ideas. But I said to you that the ideas actually have no role to play in the turn itself. So that brings up the question, what does? Why did they do it then?

I just said a second ago that what drove it was political priorities, not intellectual feasibility. Well, what were the political priorities? Who were the politicians actually listening to? Ideas can matter, but they have to be made to matter.

There are only two key players when it comes to policy changes of this kind. The key players are the politicians, because they’re the ones who are pulling the levers. But then, it’s the key constituency that actually has influence over the politicians.

The least important part is intellectuals. You might say voters have some degree of influence, but really, in a money-driven system like the United States, it’s investors, it’s capitalists — it’s big capital. They’re the ones who are pushing for these changes.

That means that if you want to understand where neoliberalism comes from, or rather if you want to understand why it came about, the answer is, it came about because capitalists ceased to tolerate the welfare state.

Now, why did they tolerate the welfare state at all? Most people on the Left understand the welfare state was brought about through massive trade union mobilization and labor mobilizations and was kept in place as long as the trade union movement had some kind of presence within the Democratic Party, within the economy more generally, because those unions were powerful enough, employers had to figure out a way of living with them. Part of what they did to live with the trade unions was to agree to a certain measure of redistribution and a certain kind of welfare state. As long as that was the case, politicians kept the welfare state going.

This is why, in that era from the mid-1930s to the mid-1970s, Keynesianism or the economics of state intervention of some kind was the hegemonic economic theory. The theory became hegemonic because it was given respectability by virtue of the fact that everybody in power was using it. Because it’s being used by people in power, it has great respectability.

This is why, in the 1950s and ’60s, Milton Friedman was in the wilderness — same guy, same ideas, equally intellectually attractive, equally technically sophisticated, but he was in the wilderness.[...]

That little story tells you something. What it says is ideas that are going into the halls of power go through certain filters. And the filters are essentially the policy priorities that the politicians have already committed to. Now, what creates those priorities? It’s the balance of class power. Social forces are setting the agenda.

If the social forces, that is, say, trade unions and community organizations, have set the agenda for politicians such that they think the only rational thing to do is to institute a welfare state, then they will bring in economists who help them design a welfare state. That gives intellectual influence to those economists. Economists who are saying “Get rid of this whole thing” are cast out into the wilderness. That’s how it works. [...]

Melissa Naschek: How do theories that focus on this notion that ideas and thinkers caused neoliberalism suggest a certain set of solutions to neoliberalism?

Vivek Chibber: It’s a really good point and a very good question. It gets us back to the issue of, why should we care about this? What does it matter if you misunderstand the factors that go into a change in economic policies? What does it matter if you wrongly attribute influence to ideas, let’s say, over material interests? Well, it can lead you to propose wrong solutions.

This is a very good example of that. If you think that what’s behind dramatic shifts in policy is the influence of ideas per se, the brilliance of those ideas, then, if you think that neoliberalism is a catastrophe and we need to go back to social democracy, then your solution is going to be, “Let’s get some economists or political scientists who are really good theorists of social democracy and give them publicity — put them in newspapers, give them lots of op-eds, maybe try to get them a meeting in the White House or something like that.”

But if you think that what’s really driving these changes is the social balance of power — the power balance between capital and labor, between rich and poor — then you won’t pour your energies into getting the right people entrée into the halls of power. You’ll pour your energies into changing the class balance. That’s the difference between how people on what used to be called the Left approach these issues and the way in which mainstream theorists and thinkers approach these issues.

This kind of ideas-based analysis leads to a great man version of policy change, whereby you get the right person in the right place with the right ideas. And then, counterfactually, the reason we don’t have a desired change is that we haven’t managed to get the right people with the right ideas into the right places. That’s a great man theory of historical change.

But if you are a socialist on the Left, you know ideas get their salience because of the background conditions, the social context, and the power relations. They don’t get their influence because of simple brilliance, at least when it comes to politics. Science is a different matter. But in politics, they get their influence because some agency with social power gives them the platform.

Without that, I mean, if the power of ideas mattered and if the correctness mattered, we’d already have a social democratic government, and we would have had one for decades. Because not only are these ideas, we think in our arrogance, they appeal to everybody.

Zohran Mamdani’s ideas, Bernie Sanders’s ideas, are not radical the way the New York Times is constantly hammering that these are radical fringe ideas. They’re mainstream as can be. They are ideas that appeal to the majority.

Why do they not have entrée? Why do they not have political influence right now? It’s because the balance of class power is such that even though they appeal to the largest number of people, those people have no political organization. They have no way of effectuating their demands. And so, their demands as encapsulated in Sanders and Mamdani don’t have a lot of political influence.

So ideas can matter, but they have to be made to matter.

by Melissa Nacheck with Vivek Chibber, Jacobin | Read more:
Image:Dirck Halstead / Getty Images

The Linguistic Foundations of Project Hail Mary

The film adaptation of Andy Weir’s novel Project Hail Mary hits general release today, March 20, and it’s great—go see it! Though a little light on the science, the movie goes hard on the relationship between schoolteacher Ryland Grace (Ryan Gosling) and an extraterrestrial named Rocky, and it’s a ride well worth taking.

But as good as it is, the movie shares a small flaw with the book: Despite having very few things in common, Grace and Rocky learn to communicate with each other extremely quickly. In fact, Grace and Rocky begin conversing in abstracts (concepts like “I like this” and “friendship”) in even less time than it takes in the book. Obviously, there are practical narrative reasons for this choice—you can’t have a good buddy movie if your buddies can’t talk to each other. It’s therefore critical to the flow of the story to get that talking happening as soon as possible, but it can still be a little jarring for the technically minded viewer who was hoping for the acquisition of language to be treated with a little more complexity.

And because this is Ars Technica, we’re doing the same thing we did when the book came out: talking with Dr. Betty Birner, a former professor of linguistics at NIU (now retired), to pick her brain about cognition, pragmatics, cooperation, and what it would actually take for two divergently evolved sapient beings not just to gesture and pantomime but to truly communicate. And this time, we’ll hear from Andy Weir, too. So buckle up, dear readers—things are gonna get nerdy.

A word about spoilers

This article assumes you’ve read Weir’s novel and that you’ve seen the movie. However, for folks who haven’t yet seen the film, I don’t think there’s much to be spoiled in terms of the language acquisition portions that we’re going to discuss—the film covers rather the same ground as the book but in a much more abbreviated way.

Still, if you want to avoid literally all spoilers, skip this article for now—at least until you’ve been to the theater!

The yawning chasm of “meaning”

Dr. Birner’s specific field of study is the science of pragmatics. “Pragmatics has to do with what I intend by what I say and what I mean in a particular context,” she explained to Ars on a Zoom call earlier this week. She elaborated by bringing up her (nonexistent) cat—the phrase “my cat” can have a multitude of meanings attached, all of which are inferred by context.

If you know Dr. Birner has a cat, her saying “my cat” could refer to that cat; if you know that she doesn’t have a cat but used to, “my cat” could refer to that cat instead, even though the semantics of the phrase “my cat” haven’t changed. That’s pragmatics, baby!

Pragmatics are particularly relevant to the Grace/Rocky language-acquisition problem because the discipline involves the creation of inferences by the listener about the speaker’s mental state and about what specific meanings the speaker implies.

But “meaning” is a fraught word here, too, because ultimately we cannot know for certain the exact meaning being implied by another person because we cannot ever truly peek inside someone else’s mind. “We are always making guesses about what our shared context is and what our shared cultural beliefs are, and, indeed, what our shared knowledge as members of the species are,” Dr. Birner continued. “And I think of this because of thumbs-up/thumbs-down.”

“The cognitive linguists George Lakoff and Mark Johnson put out a book, boy, back in the ’80s,” she said. “They talked about all of language as metaphorically built up from embodiment, our embodied experience, and our senses. So we sense up and down, and then we have this whole metaphorical notion of happy is up, so we have a thumbs up, ‘I’m feeling up today. I’m just feeling high. My spirits are lifting.’”

“Or, I can be down in the dumps,” she said. “I can be feeling low, my mood is dropping, thumbs down,’ and there’s this whole metaphorical conception. And I loved the way Project Hail Mary played with that in that Rocky didn’t share that. Rocky did not have a metaphor of ‘happy is up,’ the way Lakoff and Johnson would say we all just do.”

I asked Dr. Birner if our “up is good, down is bad” association has a biological basis in our cognition or if it’s something that has simply been shaped into a broadly shared metaphor over thousands of years of language use, and she took a moment to answer.

“That’s a really good question, and I don’t remember whether they deal with that,” she said. “But I could imagine it being biological because we start as little helpless things that can’t even stand up. And soon we stand up, we get taller, we get smarter, we get better and better the taller we get. I can actually very well imagine a biological basis for it.”

The first leap—not math, but truth

Let’s focus in on some of the specific linguistic mountains Grace and Rocky would have had to climb. The one that struck me as perhaps the most basic would be starting from pantomime and figuring out the most important thing: the twin concepts of yes and no, and the companion dualities of true/false and equal/not-equal. To me, this feels like the most mandatory of basics.

And here, perhaps, we can fall back on some good ol’ Sagan—or at least the movie version of Sagan. Dr. Birner and I (along with my colleague Jennifer Ouellette, who also hung around on the Zoom call) went back and forth for some time, but in the end, no one could really figure out a more straightforward way to demonstrate these concepts than the “primer” scene in 1997’s Contact, where the unknown alien signal is shown to contain a small grouping of symbols that appeared to represent addition, along with “equals” and “not equals” sign equivalents.

“That’s a good way to go about it, with equivalent and not-equivalent,” said Dr. Birner. “So at least you get negation, and now you can work on perceptual oppositions—up and down, black and white, loud and soft. I think that would probably be the jumping-off place for yes and no.”

Though there are linguistic biases in English and other human languages that might peek through even here—the inherent tie between “positive” (as in agreement) and “positive” (as in “this thing is good and I like it”). Careful aliens would likely want to spend a fair amount of time interrogating this bias—if it’s even visible at this point. And it likely wouldn’t be, as we haven’t built any of those syntactic bridges yet.

Pidgin? Not so fast

Getting those bridges built—going past “yes” and “no” and into some of the other basics that must be established to communicate—is not straightforward. Grace and Rocky benefit from being in a tightly constrained environment with a set of mutual problems to solve; two humans in a similar situation would likely develop a “pidgin”—an ad-hoc working language cobbled together out of components of both speakers’ languages.

But as Dr. Birner points out, true pidgin here is impossible because neither Grace nor Rocky is capable of actually producing the sounds required to speak the other’s language in the first place. “They don’t actually develop a pidgin,” she said. “They each have to learn the other’s language receptively, not productively.”

“Which is great,” she went on, “because when kids acquire language, it’s sort of a truism that reception precedes production. Every kid is going to understand more than they’re producing. Necessarily! You can’t produce what you don’t understand yet. So it makes the problem a little easier for Grace and Rocky—they don’t have to produce each other’s language, just understand it.”

Who is even there?

Grace and Rocky are lucky in that both humans and Eridians are ultimately extremely similar in their cognition and linguistics, even if their vocalizations aren’t alike. This means a lot of the mandatory requirements for conversation as we understand them are already present.

“If I encounter Rocky, I need to know, does he have a mind?” she posited. “Does he have what we call a theory of mind? Does he have a mind like mine? And does he understand that I have a mind like his, but separate? Does he understand that I can believe different things from what he believes? Can I have false beliefs? That’s all a prerequisite for communicating at all. If your mind and my mind had all the exact same stuff in it, there’d be no need to communicate.

“H.P. Grice said that communication doesn’t happen without the assumption that both parties are being cooperative,” she said. The word “cooperative” here doesn’t necessarily mean that both parties are copacetic—Dr. Birner pointed out that even when people are fighting, they tend to still be cooperatively communicating. There are rules to the interaction that must be followed if one party intends to impart meaning to the other.

Beyond adherence to the cooperative principle, another bedrock of communication is the notion of symbols, the understanding that a word can represent not just an abstract concept but can actually stand in for a thing. “I can use the word mug,” explained Dr. Birner, holding up a mug, “and mean this. And you understand what I mean, and I don’t have to show you the mug every single time.”

Also on the “mandatory” list is an understanding of the concept of displacement, which Dr. Birner attributes to the researcher Charles F. Hockett. “Displacement has long been said to be solely human, though not everyone agrees with that. It’s the ability to refer to something that is distant in time or space. I can tell you that I had a bagel this morning, even though I’m not having it right now and it’s not present right here. I had it elsewhere and I had it earlier,” she said.

She continued: “There’s this wonderful article, 1979 by Michael Reddy, called ‘The Conduit Metaphor,’ where he says that we think in metaphors. And the metaphor he’s talking about is that language is a conduit, and we really just pass ideas from my brain to yours. And he says it’s a false metaphor. It’s clearly not true that that’s what happens, but we talk about it as though it does. ‘I didn’t catch your meaning,’ or ‘Give that to me again.’ We talk as though this is a thing we literally convey, and of course we don’t convey meanings. Reddy argues that the vast majority of human communication is actually miscommunication, but so trivially that we never notice.”

By way of example, she referenced her nonexistent cat again. “If I mentioned my cat, Sammy, well, you’ll have some mental image of a cat,” she said. “It almost certainly isn’t remotely like Sammy, but it doesn’t matter. I don’t need to explain everything about Sammy. If I did, the conversation would grind to a halt and you’d never interview me again. Also, I’d be violating the cooperative principle because I would be saying too much for the current context.”

Math, the universal language?

It is a common trope in science fiction—and one brought up more than once in the comments on our last article on this subject—that “math is the only universal language.” It’s a fun, pithy saying that perhaps makes mathematicians feel good about their dusty chalkboards, but at least from my knothole, it’s a false generalization because the language in which one does one’s mathematics must be settled before any mathing can happen.

“I’m not sure that even is true on Earth,” said Dr. Birner about the notion of math as universal grammar. “The concept of zero hasn’t always been around, and how much math can you do without zero? There are languages that count, “One, two, three, many,” and that’s it. And those are human languages. So to say, ‘Math is a universal language,’ I’m already not totally on board there.”

“I think math would help, but I don’t think it would get them terribly far because they need the notion of objects. They need the notion of the semiotic function, that things stand for other things.” She paused pensively, then went on. “And once they’ve got that, that there are discrete objects and we both think of the same things as discrete objects, then we can talk about counting those objects and now we’re off and running.”

Whole-object notion is another oft-overlooked component here—often referred to as the “gavagai problem.”

“You’re pointing to a rabbit, and you say, ‘gavagai!’” said Dr. Birner. “Well, does that mean ‘rabbit?’ Does that mean ‘fur?’ Does that mean ‘ears?’ Does that mean, ‘hey look?’”

“Quine’s notion is that we default to a whole object. Well, does what counts as a whole object for me count as a whole object for you? Does every conceivable culture have discrete borders on objects?”

The author speaks on human-Eridian similarities

Fortunately for Grace and Rocky, humans and Eridians do have all these things in common because in the universe of Project Hail Mary, the species share a common ancestor. [...]

Weir notes that he worked through a number of the same linguistic issues that Dr. Birner and I raised as part of the story-generation process.

“Let’s say you have intelligent life on the planet,” he said. “What do you need? What does that species need to have to reach the point where they’re able to make spacecraft and fly around in space? Well, first off, you have to be a tribal thing. You can’t be loners. You can’t be like bears and tigers that don’t communicate with each other. You have to have the sense of a community or a tribe or a group or a gathering so that you can collaborate because you can specialize and do all these things. You need that.”

“Number two, you need language. One way or another, stuff from my brain has to get into your brain,” he said, echoing Dr. Birner’s note about Reddy’s conduit metaphor paper.

“Number three is you need empathy and compassion. A collection of beings altogether doesn’t work unless they actually are willing to take care of each other. And that’s not just found in humans—it’s found in primates. It’s found in wolf packs. It’s found in ants. It’s like any collectivized species has to have that trait.”

“You need to have compassion, empathy, which means putting yourself in somebody else’s situation. Compassion, empathy, language, a decent amount of intelligence, a tribal instinct, a group instinct, a society kind of building instinct,” he said. “You must, I believe, have all of those things in order to be able to make a spaceship. Any species that’s lacking any one of those won’t be able to do it. So any alien you meet in space is going to have all of those traits. The Friendly Great Filter is that any aliens you meet, I believe, have to have this concept of society, cooperation, empathy, compassion, collaboration, and so on.”

I’m here for Weir’s explanation—it works within the context of the science fiction universe we’re being presented, and Rocky and Grace need to be able to talk to each other or we don’t have a book (or a film!). But does it ring true under scrutiny? After all, even here on Earth, there is a wealth of problem-solving, tool-using creatures much more closely related than humans and Eridians with vastly different cognitive toolkits. Cephalopods (with distributed nervous systems and pseudo-autonomous arms), corvids, and cetaceans all have their own evolutionary approaches to communication. [...]

Here, Ars’ Jennifer Ouellette made an important point. “Rocky is basically a rock,” she said. “He’s not a human form, and that’s going to affect how a language, if there is one, evolves in that species—and it’s really going to impact how they communicate.”

“Yes, embodiment is a big deal in communications,” replied Dr. Birner, returning to the subject she’d brought up earlier, that the nature of our flesh-prisons inherently shapes not just how we experience the world but how we communicate. Our physical forms are the product of evolutionary pressures—they are the results of the inevitable, inscrutable dialogue between environment and organism. And the evolutionary pressures faced by Homo sapiens on Earth are vastly different from the evolutionary pressures faced by Eridians on Erid, and that same dialog on Erid led to vastly different outcomes. [...]

Friendly aliens

The most dangerous thing about communicating with aliens this way isn’t mistaking a word or two—it’s the more fundamental problem of what happens to third- and fourth-order assumptions when the foundations those assumptions are built on aren’t quite right. Sure, Grace and Rocky can agree that they are “friends,” but how do you explain “friend”?

“To be someone’s friend can mean a million things,” said Dr. Birner. “I have my best friend since high school. I consider you a friend,” she said, pointing at me through the screen, “and we’ve talked three times. My daughter, who’s now 35, has turned into my friend. What does that mean?”

Indeed, the notion of “friend” is a rough one—it’s fundamental to human interaction, and as such, it carries with it a huge number of (sometimes contradictory) behavioral expectations. When you’re explaining “friends” to an alien, how do you paint it? That you and the alien have shared interests and should therefore work together? That you are genuinely interested in the alien’s well-being? That you’d make sacrifices for them? That you’d expect them to help you haul furniture when you move?

And what assumptions might you make about the alien’s behavior once you’d declared each other “friends”? That they would make sacrifices for you? What if for the alien, the concept they’ve settled on for “friendship” means they’ll pull your limbs off when the adventure is over because that’s what friends do in their culture?

“You need societal grouping,” I supplied, “but you don’t necessarily need friends.”

“Absolutely,” she said. “And now I’m going to another work from 1982, Maltz and Borker, who looked at kids on the playground, and at that time—I think it’s changed a lot, it’s been 40-some years!—but at that time, they saw that little girls had a horizontal set of relationships. It was all friendship-based and secrets-based, and you have your best friend and then your next best friends. And little boys had a hierarchy, and your whole goal was to get higher in the hierarchy by insulting the kids above you and whacking them and try to be king of the hill.”

“Get the conch,” I joked unhelpfully.

“Yeah, exactly—get the conch. Again, cultural knowledge.”

by Lee Hutchinson, Ars Technica |  Read more:

Images: Project Hail Mary/Amazon MGM studios

[ed. I've always had a vague appreciation for linguistics (their effects on perceived reality and lately their nuances in bridging disagreements - for example, this is the second time in three days that I've heard the term gavagai). My grandson came over today and he went right to some YT videos explaining the basics of PHM's plot and science, especially how Ryland and Rocky communicated. Then we watched Ghostbusters. : )]

Why Your Job’s Complexity Level May Affect Your Risk of Dementia

Getting an education is important for a lot of reasons, but there might be one reason you haven’t heard — it could lower your risk of dementia later in life. Decades of research have supported this claim, with one study showing that each additional year of formal education lowers the risk of Alzheimer’s disease or other types of dementia by 7 percent.

Now, a growing body of evidence suggests that the jobs we hold throughout our lives may matter just as much or more than years of education. Having a job that involves high levels of decision-making or creativity, rather than repetitive or manual tasks, could help keep the mind sharp and active.

“Many studies suggest that, if people are working in complex jobs during their lifetime, they have a lower likelihood of developing dementia in later life,” said Jinshil Hyun, assistant professor of neurology at Albert Einstein College of Medicine.

Roles like managers, teachers, lawyers and doctors are considered high complexity jobs, while clerical, transportation and assembly line work have lower complexity. The findings are consistent with the idea that taking part in mentally stimulating activities throughout the lifespan can help preserve late-life brain health and boost cognitive reserve — the brain’s ability to cope with age- or disease-related changes.

But don’t worry if your job doesn’t meet the criteria — there are other things that you can do to improve your cognitive reserve, such as reading, socializing and volunteering.

Why work might be linked to dementia risk

“We spend most of our day in work, at least eight hours a day. So that’s like, a third of our time engaged in work, sometimes more,” said Naaheed Mukadam, professor of psychiatry at University College London. “That’s a large part of what our brain is engaged in and therefore will have a large contributory effect on cognitive reserve development.”

In a recent study, Mukadam and her colleagues investigated which factors could be influencing education’s protective effect against dementia. Their analysis included 384,284 participants and took note of health behaviors like drinking, smoking and exercise; medical conditions like hypertension and diabetes; occupational complexity; and income. The results uncovered that occupational complexity is actually the biggest reason more education tends to lower your risk of dementia, accounting for more than 70 percent of that link.

“We found that occupational complexity explained the biggest proportion of that relationship between education and dementia,” she said. “People who have more education tend to get into better paid, more complex jobs. Then, the benefits for their physical and cognitive health compound in that way.”

Multiple studies have found that those with higher income have a lower risk of dementia, and the researchers speculate that job complexity likely plays a major role in that relationship as well.

Similarly, Hyun and her colleagues found in a 2021 study that occupational complexity is predictive of later-life dementia, independent of education. They looked at the effects on dementia-free survival time, or how many years a person lived before being diagnosed with dementia, in 10,195 participants from six countries. As expected, high school graduates had a 26 percent increase in dementia-free survival time compared to people who only completed middle school or less.

After controlling for education, high occupational complexity, compared to low occupational complexity, was associated with a 19 percent increase in dementia-free survival time. Hyun speculates that the greater mental stimulation of a complex job builds cognitive reserve, which helps people resist cognitive decline and stay mentally sharp for longer, even in the presence of harmful plaques seen in Alzheimer’s-affected brains.

“The cognitive reserve hypothesis suggests that, if people are doing cognitively enriching activities, then their brain has a more efficient network,” Hyun said.

by Meeri Kim, The Washington Post/Seattle Times |  Read more:

Image: iStock

[ed. Or just develop a life long love of learning. Eric Hoffer would be a good example (longshoreman/philosopher). What'll happen when more people offload their thinking to an AI assistant?]

Man vs Mist vs Mountain

Something big is happening, but nothing big is happening to me.

Throughout my "career" as a "statistician", 13 years and counting now (but how much longer?), I've always been great at stopping myself from doing useful work. At first, I worried that I didn't know enough yet to tackle interesting problems---until I've started feeling that I forgot too much to do "real" statistics. With LLMs that barrier is now gone and I've been finding them very useful. I have just enough context and experience to pose good questions and understand the explanations.

(BTW, I am surprised by how little students and my peers seem to use them. I am usually, willingly, cast in the role of the nay-sayer. So what's happening? Are they using them surreptitiously? Or else, why do I get more utility than others?)...

So, obviously I decided to make this situation even worse and dip my toes into THE AGENTS this month (starting with the OpenAI one). In case you haven't encountered them, these are the ~latest craze in the LLM world.

Yes, just like with chatbots, you just describe what you'd like, in words, and it gets coded. But it's not just coding programs. You can do (some parts of) academic research or you can just make small, fun ideas come to life. I recently met a girl who vibe coded a Chinese medicine app that took photo of your tongue and told you seven things that were fucked up about your bladder.

Ultimately, however, my problem---because obviously I wouldn't bother to write this to just conclude that they're alright, would I---is that these tools are designed for people who like manipulating mental symbols in a certain way, you know, the screen-starers. Obviously this is a ridiculous complaint, not least because I am one of them... but as I get older, screen-staring part of my brain feels like the one I want to be visiting least often. And I think it was no coincidence that I had most fun playing with these tools when my mood was lowest.

In fact, they are addictive as hell, like a video game can feel. Everyone keeps reporting this. They dial difficulty down so much, that things get a bit muddy. People who talk about these models the most often seem to me maniacal, but I think these agents can stop you from getting actual work done. When I tried using these agents for my work, I ended up solving a lot of problems, but none of these problems seemed very important in retrospect.

Clearly that is a skill issue. I have no doubt that I'll get better at it. And if your work is mediated through screens and you're good at defining what you do and don't like, these agents may be great for me.

But at the same time, it feels like a general manifestation of any sort of "life-improving" technology, which is often just about channeling of mental disturbances. So, no, I am not banking on it making me a Nietzschean ubermensch next month; nor helping me start a billion dollar company, or even on having a better time. Right now it still feels net zero: for every bit of busy work that it may rescue me from, it feels like it has potential to rob meaningful work or meaning---or maybe even my life of life itself. "Projects" that I really care about in my life are not app-shaped or list-shaped, and in doing things, technology is always an afterthought.

by Witold Więcek, Monthly Witold | Read more:
Image: Strawberry in ASCII by Claude Sonnet 4.5 via:

[ed. More from Witold, about keeping a journal:]

***

I have been keeping a somewhat regular journal for close to 15 years now, a few pages per week usually. Most of it is very mundane, too, not even an attempt at recollection of what happened, more of a microcatalogue of internal states that feel new---I'd be less embarrassed by someone getting their paws on it as sorry for them.

Why do it? I used to call this project "long Witek", extracting what is slow-moving or semi-permanent from the detritus, the more transitory elements. I use these journals to sometimes jump back an arbitrary number of years and try to recognise myself again. In other words, I try to make myself legible to myself.

China's AI Education Experiment

A deep dive.

Pilot schools in China are already using AI to grade children’s artwork, monitor their facial expressions during lectures, and screen them for psychological problems — and the Ministry of Education (MOE) wants schools across the country to follow suit.

Integrating AI into the education system has rapidly become a top priority of the Chinese central government, which is betting that AI tools can eliminate China’s vast educational inequities and make the next generation of workers more productive. The State Council highlighted education as a key area of focus in the “AI+” plan, it received a shout-out in the 15th Five-Year Plan, and in May 2025, the Ministry of Education (MOE) released a white paper on AI for education. This MOE document proclaims that 2025 marks the dawn of an era (“智慧教育元年”), the beginning of a system-wide effort to “intelligentize” 智能化 education using AI tools. The MOE’s goal: universalize basic AI access in primary and secondary schools by 2030. Industry received that signal and responded rapidly, with Alibaba Cloud releasing its own AI+education white paper the following month. But the gap between Beijing’s (and Hangzhou’s) techno-optimism and rural China’s reality is enormous.

This report explores why the Party wants to integrate AI into education, what applications the MOE is most optimistic about, and where the barriers to successful rollout lie. We’ll limit our analysis to K-12 education today, but university AI initiatives will be the focus of our next report in this series!

Institutional History

In official discourse, China is said to have entered a “post-equity era” 后均衡时代 since the MOE announced that all counties had met the baseline quality level for compulsory schooling in 2021. Now, the focus is shifting from access to education to improving the quality of that education. The 14th 5-year plan (2021-2025) prioritized expanding infrastructure in rural schools through the “county-level high school revitalization initiative” (县中振兴), part of which involved equipping classrooms with ‘smart hardware’ such as digitized blackboards. During this period, the party spent significant resources to provide nearly every school with an internet connection.

Still, rural education in China faces serious structural challenges. I spoke with Leo He — a research fellow at the Hoover Institution who did NGO work in rural China from 2019 to 2023 — for a firsthand account of the situation. Every locality, he explained, has designated “elite” schools that talented students from surrounding areas compete to transfer into. The result is a system where “educational resources are systematically sucked up to the center from the periphery, leaving rural areas incredibly depleted.” While this arguably gives academically gifted students opportunities to develop their talents, it deprives most students of educational resources.

According to China’s 2020 census, only 30.6% of the population has ever attended high school (including non-academic vocational secondary school), which Stanford professor Scott Rozelle notes, “is lower than South Africa, lower than Turkey and lower than Mexico.” In 2022, roughly 40% of China’s middle school graduates didn’t go on to attend high school of any kind, and among the students that do continue their education, national policy stipulates that roughly half (“五五分流”) are funneled into non-academic vocational high schools with no path to enter college.

To understand how AI could fit into this picture, we first need to understand the political and economic factors that incentivize Beijing to care about students in the countryside. It’s not clear that more investment in education will translate to high economic growth at this point in China’s development path — the real youth unemployment rate is probably still around 20%, and there are fewer entry-level positions available just as a record number of new graduates enter the workforce. Rather, this is a priority for the Party because improving the education system is so popular.

When Rozelle’s team surveyed 1,800 rural mothers and asked what they wanted their children to aspire to, over 95% said, “I want my child to go to college.” In China, a degree from an elite college doesn’t just translate to higher earnings — it unlocks better healthcare via the hukou system, cushy “iron rice bowl” 铁饭碗 jobs, and above all, social prestige. In 2023, researchers at Stanford found that Chinese families spent an average of 17.1% of their annual household income on education, which amounts to 7.9% of annual household expenditures. (Households in the US and Japan, by comparison, dedicate just 1-2% of annual expenditures to education.) The poorest quartile of families in China devotes a staggering 56.8% of income to education, and education spending is inelastic — that is, it’s prioritized as a necessary expense — across all income levels.

As Andrew Kipnis, the anthropologist who wrote Governing Educational Desire, explained to ChinaTalk, educational reform is a priority for the party “because it’s a way of keeping people happy. If they think there’s some hope their child will attend university, that gives them some investment in the system.” But not every child can become part of the elite: “People who have gone to university won’t work in factories,” as Kipnis put it. No matter how popular it would be, Beijing is not interested in building a system where a college education is available to anyone who wants one. But within this zero-sum system, where anyone who receives an advantage is inherently disadvantaging someone else, the party still needs to make parents feel like their child is getting ahead. Infrastructure is pretty much the perfect tool for this. It makes schools feel luxurious on the ground without changing the fundamentals that make the system so unfair. Shiny new facilities deliver popularity gains immediately, and if your child doesn’t get into university years later, it’s their own damn fault.

Those incentives are shaping the world’s largest AI education experiment. China is not the only country betting that AI will transform education, but the scale and style of China’s ambitions are unmatched globally. While China started with pilot programs, South Korea’s government led with inflexible national-level implementation, spending US$850 million on an ambitious AI textbook initiative that collapsed after just 4 months. India’s edtech ecosystem is private-sector-led with little top-down guidance or regulation, which resulted in the high-profile implosion of Byju’s and a proliferation of predatory practices targeting low-income families. Japan, unlike China, pledged to make sure every student had a device before implementing AI teaching tools.

Ultimately, China stands out globally for the sheer scale of its AI education ambitions — and the scope of applications its edtech industry is targeting for AI integration.

by Lily Ottinger, China Talks |  Read more:

Image: via

[ed. See also: Massive budget cuts for US science proposed again by Trump administration (Nature). National Science Foundation.]

The Big T-Shirt Payoff

The College Student—and His Cat Meme—Who Hunted the World’s Biggest Cyberweapon

Sitting in his dorm room at the Rochester Institute of Technology, Benjamin Brundage was closing in on a mystery that had even seasoned internet investigators baffled. A cat meme helped him crack the case.

A growing network of hacked devices was launching the biggest cyberattacks ever seen on the internet. It had become the most powerful cyberweapon ever assembled, large enough to knock a state or even a small country offline. Investigators didn’t know exactly who had built it—or how.

 

Brundage had been following the attacks, too—and, in between classes, was conducting his own investigation. In September, the college senior started messaging online with an anonymous user who seemed to have insider knowledge.

As they chatted on Discord, a platform favored by videogamers, Brundage was eager to get more information, but he didn’t want to come off as too serious and shut down the conversation. So every now and then he’d send a funny GIF to lighten the mood. Brundage was fluent in the memes, jokes and technical jargon popular with young gamers and hackers who are extremely online.

“It was a bit of just asking over and over again and then like being a bit unserious,” said Brundage.

At one point, he asked for some technical details. He followed up with the cat meme: a six-second clip that showed a hand adjusting a necktie on a fluffy gray cat.

Brundage didn’t expect it to work, but he got the information. “It took me by surprise,” he said.

Eventually the leaker hinted there was a new vulnerability on the internet. Brundage, who is 22, would learn it threatened tens of millions of consumers and as much as a quarter of the world’s corporations. As he unraveled the mystery, he impressed veteran researchers with his findings—including federal law enforcement, which took action against the network two weeks ago.

Chad Seaman, a researcher at Akamai, joked at one point that the internet could go down if Brundage spent too much time on his exams.

Early warning

Three times a year, several hundred of the techies who keep North America’s internet running gather to talk shop. Last June they met at a conference in Denver hosted by the North American Network Operators’ Group.

One major topic was a fast-growing and often legally dubious business known as residential proxy networks. Dozens of companies around the world run such networks, which are made up of consumer devices like phones, computers and video players.

These “res proxy” companies rent out access to internet connections on the devices to customers who want to look like they’re surfing the internet from a genuine home address.

That kind of access is useful for people who want privacy or for companies that want to masquerade as regular people to test out internet features for particular regions or scrape the web for data (say, a shopping price-comparison site). AI companies use the networks to get around blocks on automated traffic so they can gather large amounts of data to train their models.

Then there are the customers who want to hide their identity while engaging in ticket scalping, bank fraud, bomb threats, stalking, child exploitation, hacking or espionage.

Some device owners willingly sign up to be on these networks so they can make a few dollars a month, but most have no idea they’re connected to one.

At the Denver conference, Craig Labovitz was alarmed. The Nokia executive had been tracking the data flows of the internet’s infrastructure for years, and he knew the network’s data centers, chokepoints and design better than most.

Starting in January 2025, Nokia’s sensors had picked up a series of increasingly powerful cyberattacks coming from devices that hadn’t previously been considered dangerous. Called distributed denial of service, or DDoS, attacks, these were massive floods of junk internet data designed to knock websites offline by overwhelming the data pipes that connected them. These attacks are sometimes launched by extortionists or even business rivals seeking to sabotage computer networks.

Nokia saw hundreds of thousands of devices joining in these attacks. One unprecedented attack later in the year on internet service provider Cloudflare was “comparable to the combined populations of the UK, Germany, and Spain all simultaneously typing a website address and then hitting ‘enter’ at the same second,” Cloudflare said.

The network, which would become known as Kimwolf, seemed to be using residential proxy connections to launch its attacks, giving it the potential to do massive damage.

“The basic message was, ‘Be afraid,’” Labovitz remembers. [...]

Instead he applied his hacking skills toward legitimate cybersecurity research. In his senior year of high school, he found bugs in websites belonging to the Dutch government and reported them via a “bug bounty” program that offered hackers prizes for unearthing security flaws.A few months later, the Dutch National Cyber Security Center mailed him his bounty: a black T-shirt. It read: “I hacked the Dutch government and all I got was this lousy t-shirt.”

He remembers it as one of the most rewarding experiences of his young life: a “dopamine rush,” he said. [...]

On March 19, federal authorities announced they’d disrupted four of the world’s largest DDoS botnets, including Kimwolf. Kimwolf had launched more than 26,000 DDoS attacks targeting over 8,000 victims, according to a court filing. The press release announcing the takedown thanked Brundage’s company, Synthient, among others.

​Industry experts say that Kimwolf today is a shadow of its former self. The cybersecurity firm Netscout says it’s seeing about 30,000 Kimwolf machines active at any given time.

Brundage recently got a text message from a federal official on the case. The official had heard about the bug bounty Brundage got from the Dutch government years ago and had a question: “What’s a good address to mail you a t-shirt, and what’s your size?”

by Robert McMillan, Wall Street Journal |  Read more:

Image: via

[ed. Here's how to protect yourself.]