timelets

An even more bewildering use of drones took place in the early days of the Shanghai lockdown. The city’s top mental health official introduced an unexpectedly sparky phrase in an otherwise drab press conference on the course of the virus, demanding that Shanghainese “repress your soul’s yearning for freedom.”

“One night in April, as the lockdown swung into high gear, a drone carrying a megaphone began blasting that message into apartments full of huddling residents: “Repress your soul’s yearning for freedom,” with a woman’s voice played on loop while a light blinked from the drone. “Do not open your windows to sing, which can spread the virus.”

-- Daniel Wang. “Breakneck: China’s Quest to Engineer the Future.”

“[China] embraced a vision of technology radically different from Silicon Valley’s: the pursuit of physical and industrial technologies rather than virtual ones like social media or e-commerce platforms. In China, technology is not represented by shiny objects; rather, it is embodied by communities of engineering practice like Shenzhen, where technology lives inside the heads and in the hands of its workforce. ”
...
Chinese officials climbed over each other to host a Foxconn facility. They salivated at the number of jobs and amount of tax revenues the company could create for their jurisdiction, which could elevate them to higher office. Local officials promised to satisfy Foxconn’s extraordinary labor demands. In Chengdu, minor bureaucrats had to hit quotas on the number of workers to rustle up for factory work; those who failed might receive an order to work at assembly lines themselves.
...
A 2012 story in the New York Times reported that Apple needed to hire nearly nine thousand industrial engineers in the earlier days of iPhone production. The company’s analysts expected recruitment to last nine months to hire that many engineers in the United States. In China, they were able to do it in two weeks.

-- Daniel Wang. “Breakneck: China’s Quest to Engineer the Future.”

Right now, workers are potentially training AI how to make them obsolete. And they often don’t realize it.

The kind of AI used by companies, called an enterprise AI system, can capture everything you do at work and use that information to train itself. These systems can record your interactions within the platform—the prompts you write, the documents you create, the queries you run.

In other words, the company can potentially track—and claim ownership of—every keystroke you make within the system, every idea you document there, every tool you build using that platform.


This dynamic may fundamentally change the relationship between employer and employee. The stakes are so high and so urgent that both sides are rushing to position (or protect) themselves. Executives are rapidly implementing enterprise AI systems, seeking productivity gains and competitive advantage—and they often aren’t disclosing the implications for job security and privacy. Meanwhile, at least some employees are secretly adopting personal AI tools, sometimes violating corporate policies, so that their employers can’t capture everything they know and do.

Individual opt-out of AI is often impossible, so unions and professional associations need to pay attention. With collective bargaining, workers could demand transparency about the use of enterprise AI and demand fair compensation for the knowledge it gathers. Without collective power, individual employees will keep clicking “accept” on agreements that restructure their jobs simply because they have no alternative.

https://www.wsj.com/lifestyle/careers/ai-knowledge-capture-employees-a69a0e1c

Mossing and others, both at OpenAI and at rival firms including Anthropic and Google DeepMind, are ... studying them [LLMs] as if they were doing biology or neuroscience on vast living creatures—city-size xenomorphs that have appeared in our midst.

Anthropic and others have developed tools to let them trace certain paths that activations follow, revealing mechanisms and pathways inside a model much as a brain scan can reveal patterns of activity inside a brain. Such an approach to studying the internal workings of a model is known as mechanistic interpretability. “This is very much a biological type of analysis,” says Batson. “It’s not like math or physics.”

Anthropic invented a way to make large language models easier to understand by building a special second model (using a type of neural network called a sparse autoencoder) that works in a more transparent way than normal LLMs. This second model is then trained to mimic the behavior of the model the researchers want to study.

Creating a model that behaves in predictable ways in specific scenarios requires making assumptions about what the inner state of that model might be in those scenarios. But that only works if large language models have something analogous to the mental coherence that most people do.

And that might not be the case.

...
Another possible solution ... Instead of relying on imperfect techniques for insight into what they’re doing, why not build an LLM that’s easier to understand in the first place?

https://www.technologyreview.com/2026/01/12/1129782/ai-large-language-models-biology-alien-autopsy/

Elon Musk’s artificial intelligence startup xAI is burning through $1 billion a month as the cost of building its advanced AI models

https://www.bloomberg.com/news/articles/2025-06-17/musk-s-xai-burning-through-1-billion-a-month-as-costs-pile-up

“In 1812, the first edition of their anthology, comprising eighty-six stories, came out in an edition of 600, with an apparatus of notes running to hundreds of pages. It was not really intended to be read for pleasure at all by the children and households of its title; it was a learned work setting out to reconfigure the cultural history of Germany along lines that would emancipate it from the monopoly of classical and French superiority. Yet this collection—by the final, standard edition of 1857 the number of tales had grown to 210—was to become the most widely translated work in the world after the Bible and the Qur’an, rendered into more than 160 languages so far, including Xhosa and Tagalog, and still counting.”
...
"Germany wasn’t yet Germany. It was a congeries made up of dozens of principalities and archdukedoms, free Hanseatic ports and archbishoprics. History shows us that the modern nation-state develops long after a national culture and its language: think of Italy and of Dante, writing five hundred years before Italian unification. For centuries, “most of the peninsula had been under German or Austrian or Spanish rule, while the Renaissance, indisputably Italian, was influencing the whole world. The Grimms were living in a time of turmoil and bloodshed. ”
...
“One response to humiliation is to assert cultural riches and distinctiveness, even pre-eminence. ”

--- Warner, Marina;. “Once upon a Time."