Lea Pao, a professor of literature at Stanford University, has been experimenting with ways to get her students to learn offline. She has them memorize poems, perform at recitation events, look at art in the real world.

It’s an effort to reconnect them to the bodily experience of learning, she said, and to keep them from turning to artificial intelligence to do the work for them. “There’s no AI-proof anything,” Pao said. “Rather than policing it, I hope that their overall experiences in this class will show them that there’s a way out.”

It doesn’t always work. Recently, she asked students to visit a local museum, look at a painting for 10 minutes, and write a few paragraphs describing the experience. It was a purposefully personal assignment, yet one student responded with a sophisticated but drab reflection – “too perfect, without saying anything”, Pao said. She later learned the student had tried to visit the museum on a Monday, when it was closed, and then turned to AI.

As artificial intelligence has upended the way in which students read, learn and write, professors like Pao have been left to their own devices to figure out how to teach in a transformed landscape.

Last Friday, Nintendo joined thousands of companies suing the Trump administration to secure full refunds, plus interest, for billions in unlawful tariffs collected under the International Emergency Economic Powers Act (IEEPA).

In its complaint, Nintendo insisted that the Trump administration has already conceded that more than $200 billion in refunds are owed to hundreds of thousands of importers who paid tariffs, regardless of liquidation status.

However, Nintendo fears that the Trump administration may try to avoid paying refunds to certain companies whose tariff payments have already been liquidated, which means that the duties owed were finalized. The government has continually argued that it will only follow through on refunding all importers if a court directly orders refunds to be repaid in a way that requires reliquidation. Such an order would force officials to void all finalized tariffs and come as a relief to many companies in Nintendo’s position that remain uncertain if all their tariff payments can be clawed back.

Ultimately, Nintendo argued, it increasingly seems like the government plans to delay refunds until the court steps in. That leaves it up to the Court of International Trade to order Trump officials to do the right thing, Nintendo said. And in the gaming giant’s view, that’s to proceed with prompt refunds to make all importers whole.

"Could" is doing a lot of heavy lifting there.

Many parts of the world are experiencing a housing crisis, with demand in urban areas often outpacing supply, leading to soaring prices.

In countries including the UK and the US, an aging population of builders combined with a drive to fill the housing shortage means there is a need for more construction workers. The UK’s Construction Industry Training Board found that the country will need 250,000 more workers by 2028 to meet building targets but in 2023, more people left the industry than joined.

UK technology company Automated Architecture, or AUAR (pronounced “our”) believes it has a solution. It makes portable micro-factories that can produce the wooden framing of a house — the walls, floors and roofs. Co-founder Mollie Claypool says the micro-factories will be able to produce the panels quicker, cheaper and more precisely than a timber framing crew, freeing up carpenters to focus on the construction of the building.

Despite the focus on automation, Claypool insists she is not trying to put anyone out of work. “Automation isn’t replacing jobs. Automation is filling the gap,” she told CNN.

AI is everywhere. It’s in your phones, in your Internet searches, in defense software. And it’s expanding. The big tech giants—Alphabet, Microsoft, Meta and Amazon—are planning on spending nearly $700 billion this year alone on building out AI infrastructure.

And more recently, Thomas Germain, a tech reporter at the BBC, conducted a personal experiment into how an invested individual—or business—can get ChatGPT and Google Search’s “AI Overview” to spread lies. We talked to Thomas to find out just how easy it is to hack these common AI tools and what the consequences of that could be.

Pierre-Louis: Hi, Thomas. Thanks for taking the time to join us today.

Thomas Germain: Thanks for having me on.

Pierre-Louis: So my understanding is you hacked ChatGPT.

Germain: That’s right. So I got a tip a couple of weeks ago that manipulating the things that AI tools like ChatGPT or Google Gemini or the little, you know, “AI Overview” at the top of Google Search, apparently manipulating the things that they say to other people can be as easy as publishing an article on your own website, like a blog post, and apparently, people are doing this across the whole Internet.

I think this recent post by AI industry CEO Matt Shumer is worth a read. In it, he basically explains how quickly LLMs (large language models) are evolving to supplant many developers and programmers, and how that disruption is coming to other industries quickly. He also warns critics of AI to adjust their priors and realize the AI tools you mocked just six months ago, aren’t the ones in use today:

“I am no longer needed for the actual technical work of my job. I describe what I want built, in plain English, and it just… appears. Not a rough draft I need to fix. The finished thing. I tell the AI what I want, walk away from my computer for four hours, and come back to find the work done. Done well, done better than I would have done it myself, with no corrections needed. A couple of months ago, I was going back and forth with the AI, guiding it, making edits. Now I just describe the outcome and leave.”

While the post is interesting (with the understanding this is somebody making and selling automation software), you might notice something: absolutely nowhere in the blog post does he meaningfully acknowledge the widespread problems with existing AI use. Either because his financial self-interest doesn’t allow for honest acknowledgment of them, or because he simply doesn’t find those aspects all that interesting.

Maybe both.

A couple of timely blog posts remind us that RSS is alive, well, and can help you resist enshittification of the Web.

Last week, Caroline Crampton's blog post, The View from RSS really caught our attention, helped by its subtitle: What the web looks like when you subscribe to 2,000 RSS feeds. We were not the only ones who it grabbed: at the weekend, Cory Doctorow also picked up on it in a post called The web is bearable with RSS.

One of the snags of reporting on the tech sector is tackling the constant stream of announcements of radical new technology that is going to change everything. Another, of course, is trying to find out about them via websites in the 2020s, where even with an in-browser ad-blocker, plus a network one too, and an anti-cookie warning extension, many websites are still horribly cluttered.

So when someone following two thousand feeds tells you that an RSS reader can strip a lot of the cruft away, and a caped crusader of the blogosphere agrees ... well, this vulture sits up and pays attention.

The origins of the RSS system go back to the 1990s, and like the Markdown markup language we reported on earlier today, the RSS 1.0 standard was co-developed by the late Aaron Swartz when he was just 14 years old.

With enough RSS feeds, web search only becomes an occasional hell.

Meta’s approach to user privacy is under renewed scrutiny following a Swedish report that employees of a Meta subcontractor have watched footage captured by Ray-Ban Meta smart glasses showing sensitive user content.

The workers reportedly work for Kenya-headquartered Sama and provide data annotation for Ray-Ban Metas.

The February report, a collaboration from Swedish newspapers Svenska Dagbladet, Göteborgs-Posten, and Kenya-based freelance journalist Naipanoi Lepapa, is, per a machine translation, based on interviews with over 30 employees at various levels of Sama, including several people who work with video, image, and speech annotation for Meta’s AI systems. Some of the people interviewed have worked on projects other than Meta’s smart glasses. The report’s authors said they did not gain access to the materials that Sama workers handle or the area where workers perform data annotation. The report is also based on interviews with former US Meta employees who have reportedly witnessed live data annotation for several Meta projects.

The report pointed to, per the translation, a “stream of privacy-sensitive data that is fed straight into the tech giant’s systems,” and that makes Sama workers uncomfortable. The authors said that several people interviewed for the report said they have seen footage shot with Ray-Ban Meta smart glasses that shows people having sex and using the bathroom.

The core thesis of this paper is that the AI community needs to stop treating autonomous agents as just another text generation problem and start building comprehensive infrastructure to support closed loop learning. The authors argue that achieving reliable agentic behavior requires a full stack ecosystem that unifies data synthesis with sandboxed execution and specialized reinforcement learning. To prove this point they introduce the Agentic Learning Ecosystem which consists of an RL framework called ROLL alongside a sandbox manager named ROCK and an agent interface known as iFlow CLI. They believe that isolating models in static training environments is a dead end for solving complex real world workflows.

The team developed an open source model named ROME using a tightly integrated training pipeline with reproducible execution environments which allowed a relatively small 30 billion parameter model to rival or beat massive proprietary models exceeding 100 billion parameters on difficult software engineering benchmarks.

A big part of their argument rests on the idea that credit assignment in reinforcement learning needs to change. They propose a novel algorithm called Interaction Perceptive Agentic Policy Optimization which shifts the reward focus from individual text tokens to broader semantic interaction chunks. This chunk level optimization stabilizes the training process over long horizons and prevents the policy collapse often seen in complex tool use scenarios.

We're increasingly seeing a shift of priorities away from raw data scale and focus on the systematic infrastructure as the actual bedrock of next generation models.

About a year and a half ago, I wrote about my kid’s experience with an AI checker tool that was pre-installed on a school-issued Chromebook. The assignment had been to write an essay about Kurt Vonnegut’s Harrison Bergeron—a story about a dystopian society that enforces “equality” by handicapping anyone who excels—and the AI detection tool flagged the essay as “18% AI written.” The culprit? Using the word “devoid.” When the word was swapped out for “without,” the score magically dropped to 0%.

The irony of being forced to dumb down an essay about a story warning against the forced suppression of excellence was not lost on me. Or on my kid, who spent a frustrating afternoon removing words and testing sentences one at a time, trying to figure out what invisible tripwire the algorithm had set. The lesson the kid absorbed was clear: write less creatively, use simpler vocabulary, and don’t sound too good, because sounding good is now suspicious.

At the time, I worried this was going to become a much bigger problem. That the fear of AI “cheating” would create a culture that actively punished good writing and pushed students toward mediocrity. I was hoping I’d be wrong about that.

Turns out … I was not wrong.

I'm accused of being AI on other sites simply because I construct complex sentences with regularity -- and use emdashes.

Outside OpenAI’s headquarters, a handful of people gathered on Monday holding pieces of colorful chalk. They got down on their knees and started writing messages on the sidewalk. Stand for liberty. Please no legal mass surveillance. Change the contract please.

At issue was a business deal that the company recently signed with the Department of Defense, following the Pentagon’s sudden turn against Anthropic. OpenAI will now supply its technology to the military for use in classified settings, the sorts that may involve wartime decisions and intelligence-gathering—an agreement, many legal experts told me, that could give the government wide-ranging powers. “I would just really like to see OpenAI do the right thing and stand up for something, anything,” Niki Dupuis, an AI-start-up founder and one of the chalk protesters, told me.

In a widely leaked internal memo that Sam Altman sent last Thursday night, a copy of which I obtained, the OpenAI CEO said that he would seek “red lines” to prevent the Pentagon from using OpenAI products for mass domestic surveillance and autonomous lethal weapons. These were ostensibly the very same limits that Anthropic had demanded and that had infuriated the Pentagon, leading Defense Secretary Pete Hegseth to declare the company a supply-chain risk—a hefty sanction that would require anybody who sells to the Pentagon to stop using Anthropic products in their work with the military. Perhaps OpenAI was about to secure the very terms Anthropic had been denied.