Counterintuitively, the US has too *many* bus stops to make its bus services efficient. CC-licensed photo by Matt on Flickr.
You can sign up to receive each day’s Start Up post by email. You’ll need to click a confirmation link, so no spam.
A selection of 9 links for you. Not stopping here. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.
Why Greenland’s natural resources are nearly impossible to mine • The Week UK
Justin Klawans:
»
The island’s frigid, Arctic climate serves as the main culprit for challenging mining. Most of Greenland’s natural resources are “located in remote areas above the Arctic Circle, where there is a mile-thick polar ice sheet and darkness reigns much of the year,” said CNN. While people may understandably think neighboring Iceland is blanketed by ice, Greenland actually has the harsher climate; about “80% of Greenland is covered with ice,” and mining the “Arctic can be five to 10 times more expensive than doing it elsewhere on the planet.”
As a result, most of the efforts to mine Greenland’s minerals “generally haven’t advanced beyond the exploratory stage,” said The Associated Press. And beyond the weather playing a major factor, the remote areas where many of these elements are located also present a problem. Even in southern Greenland, where the island is “populated, there are few roads and no railways, so any mining venture would have to create these accessibilities,” said Diogo Rosa, an economic geology researcher with the Geological Survey of Denmark and Greenland, to the AP. There is also the question of power, as many of these remote areas don’t have consistent electricity.
As of now, Greenland only has one fully operational mine, which produces anorthosite and is “located deep inside a fjord system with no road access,” said Business Insider. All of the mine’s supplies, including the crew, “arrives by ship during the ice-free months or by helicopter when the fjord freezes over for months on end.” And there may not be another operational mine for a while; on average it “takes 16 years to develop a mine, right from the first idea to the actual mine,” Naaja Nathanielsen, Greenland’s minister responsible for natural resources, said to Business Insider.
It is clear that the “Trump administration might want to dominate the Arctic, not least to gain relative power over Russia and China,” Lukas Slothuus, a postdoctoral research fellow at the U.K.’s University of Sussex, said at The Conversation. But given the challenges with mining, any “natural resource extraction is unlikely to feature centrally.” If foreign powers did find a way to mine in Greenland, this would “reverberate in Copenhagen, as Greenland has a mining profit-sharing agreement with Denmark.”
«
Alternatively, as one person posted on Twitter: “Honestly this is a great premise for a sitcom:
In the icy wasteland of Greenland, a ragtag NATO detachment of 13 Germans, 2 Norwegians, 3 Swedes, 15 French and 1 lone Brit must “defend” the Arctic from… well, mostly boredom, polar bears, and each other’s national stereotypes.
Stationed at a remote outpost in Nuuk, their biggest enemies are cabin fever, frozen plumbing, the Danes who keep forgetting they’re there, and each other. They’re constantly prepping to counter the absurd threat of an American takeover with the total force smaller than a pub trivia team.
The show will have everything – petty national rivalries that project from history, offensive cultural stereotypes, frustration due to NATO bureaucracy, and because this is so European, the whole thing will be pretty gay. Oh and a recurring theme is that they will constantly be picking on the lone Brit.”
Apart from the “gay” bit (why??), I think this should be commissioned at once.
unique link to this extract
RAM shortage chaos expands to GPUs, high-capacity SSDs, and even hard drives • Ars Technica
Andrew Cunningham:
»
Big Tech’s AI-fueled memory shortage is set to be the PC industry’s defining story for 2026 and beyond. Standalone, direct-to-consumer RAM kits were some of the first products to feel the bite, with prices spiking by 300% or 400% by the end of 2025; prices for SSDs had also increased noticeably, albeit more modestly.
The rest of 2026 is going to be all about where, how, and to what extent those price spikes flow downstream into computers, phones, and other components that use RAM and NAND chips—areas where the existing supply of products and longer-term supply contracts negotiated by big companies have helped keep prices from surging too noticeably so far.
This week, we’re seeing signs that the RAM crunch is starting to affect the GPU market—Asus made some waves when it inadvertently announced that it was discontinuing its GeForce RTX 5070 Ti.
Though the company has since tried to walk this announcement back, if you’re a GPU manufacturer, there’s a strong argument for either discontinuing this model or de-prioritizing it in favor of other GPUs. The 5070 Ti uses 16GB of GDDR7, plus a partially disabled version of Nvidia’s GB203 GPU silicon. This is the same chip and the same amount of RAM used in the higher-end RTX 5080—the thinking goes, why continue to build a graphics card with an MSRP of $749 when the same basic parts could go to a card with a $999 MSRP instead?
…SSD prices have been climbing alongside RAM prices for months, but as of last month, the price increases for 500GB, 1TB, and 2TB drives were around twice as expensive as they had been in August 2025, compared to three or four times as expensive for DDR5 RAM kits. But things are moving in a bad direction, and high-capacity SSDs in particular are being hit hard.
If you want a 1TB internal M.2 SSD from a recognizable company, that will generally run you somewhere between $120 and $150—often a little higher than the $135 we recorded for a 1TB WD Blue drive back in January, but not absurdly so.
It’s still possible to find a decent 2TB drive like this SiliconPower model for around $230, the price we quoted for a 2TB Western Digital SN7100 in December. But currently, SSDs from big-name companies like Samsung and WD/SanDisk (currently in a state of flux, but the SanDisk-branded versions of these drives don’t appear to be at retail yet) are either out of stock at many major retailers or have shot up dramatically in price.
«
WhatsApp texts are not contracts, judge rules in £1.5m divorce row • The Times
Ed Halford:
»
A divorced painter who claimed her former husband signed over their £1.5 million home in an affluent north London area via WhatsApp has lost a High Court “test case” to keep the property.
Hsiao-mei Lin, 54, a Taiwanese-born British citizen, and the financier Audun Mar Gudmundsson, also 54, married in 2009 but separated seven years later.
During divorce proceedings, Lin was awarded their £1.5m house in Tufnell Park, an area known for its Victorian architecture and home to celebrities such as the actors Damian Lewis and Bill Nighy.
However, Lin, a graduate of the Royal Academy Schools, was unaware that only a week before their divorce, Gudmundsson, who ran a mezzanine finance company, was made bankrupt. It transpired that this was due to the financier owing a former friend and others £2.5m.
Despite a High Court ruling in 2024 declaring that the bankruptcy meant Lin only owned a 50% stake in their family home, she remained determined to prove that Gudmundsson had signed off the entire home to her before divorce proceedings. Lin’s proof, however, was not a watertight legal document but WhatsApp exchanges with her husband.
In a two-day appeal hearing last month, the trustees, Maxine Reid-Roberts and Brian Burke, disputed that these messages validly “disposed of” Gudmundsson’s interest in the house.
…Lin’s lawyers have claimed that because Gudmundsson’s name was in the header to the messages when they reached Lin’s phone they should be considered “signed”.
…Steve Fennel, acting for the trustees, argued that the messages were not signed and were therefore not legally binding. He said that if they were found to be so then the “result will be that a WhatsApp message in and of itself, without a ‘signature’ in the text, will in all cases count as signed for the purposes of all statutory requirements for signature”.
«
Tricky case. But does this mean that just because it came from your phone it isn’t legal? What does that mean for email?
unique link to this extract
Matt Damon: Netflix wants movies to account for viewers on their phone • Variety
Jack Dunn:
»
[The actor Matt] Damon pointed out that because viewers give a “very different level of attention” to a movie at home versus in a theatre, Netflix wants to push the action set pieces toward the front of the runtime. He also said there are behind-the-scenes discussions about reiterating “the plot three or four times in the dialogue” to account for people being on their phones.
“The standard way to make an action movie that we learned was, you usually have three set pieces. One in the first act, one in the second, one in the third,” Damon explained. “You spend most of your money on that one in the third act. That’s your finale. And now they’re like, ‘Can we get a big one in the first five minutes? We want people to stay. And it wouldn’t be terrible if you reiterated the plot three or four times in the dialogue because people are on their phones while they’re watching.'”
[Ben] Affleck then cut in, adding that the streamer formula for successful content isn’t the only way. He used Netflix’s recent limited series hit “Adolescence” as a shining example.
«
This was from an appearance they made on the Joe Rogan show. Can’t decide if it’s surprising or encouraging that all these years after they wrote and appeared in Good Will Hunting, they’re still bumping around together.
The point about restating the plot, though, is part of what’s noxious about streamer-made content. It’s also why some people complained about Pluribus, the new show on Apple TV+, being “too slow”: you have to wait for things to happen, and they don’t explain the plot every five minutes.
unique link to this extract
ChatGPT to start showing ads in the US • The Guardian
Robert Booth:
»
ChatGPT will start including advertisements beside answers for US users as OpenAI seeks a new revenue stream.
The ads will be tested first in ChatGPT for US users only, the company announced on Friday, after increasing speculation that the San Francisco firm would turn to a potential cashflow model on top of its current subscriptions.
The ads will start in the coming weeks and will be included above or below, rather than within, answers. Mock-ups circulated by the company show the ads in a tinted box. They will be served to adult users “when there’s a relevant sponsored product or service based on your current conversation”, according to OpenAI’s announcement. Ads will not be shown to users under 18 and will not appear alongside answers related to sensitive topics such as health, mental health or politics. Users will be able to click to learn about why they received a particular ad, according to OpenAI.
Previously, OpenAI’s CEO, Sam Altman, expressed reluctance to introduce ads to the chatbot: “I kind of hate ads just as an aesthetic choice.” His company has made commitments to spend more than $1tn on infrastructure supporting AI in the coming years. Altman has said that revenues are running at well over $13bn a year.
“Maybe there could be ads outside the [large language model] stream that are still really great, but the burden of proof there would have to be very high. And it would have to feel really useful to users and really clear that it was not messing with the model’s output,” Altman said recently. “I think it’d be very hard, we’d have to take a lot of care to get it right. People have a very high degree of trust in ChatGPT.”
«
As someone observed, Altman has finally achieved his goal of AGI: ad-generated income. Google started showing ads beside search results in October 2000, just two years after it was founded (September 1998). OpenAI is going to start showing adverts 10 years after its December 2015 founding, though only slightly more than three years after going public with ChatGPT in November 2022.
So, basically, it’s running on about the same timeline.
unique link to this extract
How Generative AI is destroying society • Marcus on AI
Gary Marcus, who is generally pessimistic about the big claims made for generative AI, is impressed by this paper:
»
This is one of the most lucid and powerful articles I have read in years. Here’s the opening paragraph.
»
If you wanted to create a tool that would enable the destruction of institutions that prop up democratic life, you could not do better than artificial intelligence. Authoritarian leaders and technology oligarchs are deploying AI systems to hollow out public institutions with an astonishing alacrity. Institutions that structure public governance, rule of law, education, healthcare, journalism, and families are all on the chopping block to be “optimized” by AI. AI boosters defend the technology’s role in dismantling our vital support structures by claiming that AI systems are just efficiency “tools” without substantive significance. But predictive and generative AI systems are not simply neutral conduits to help executives, bureaucrats, and elected leaders do what they were going to do anyway, only more cost-effectively. The very design of these systems is antithetical to and degrades the core functions of essential civic institutions, such as administrative agencies and universities.
«
In the third paragraph they lay out their central point:
»
In this Article, we hope to convince you of one simple and urgent point: the current design of artificial intelligence systems facilitates the degradation and destruction of our critical civic institutions. Even if predictive and generative AI systems are not directly used to eradicate these institutions, AI systems by their nature weaken the institutions to the point of enfeeblement. To clarify, we are not arguing that AI is a neutral or general purpose tool that can be used to destroy these institutions. Rather, we are arguing that AI’s current core functionality—that is, if it is used according to its design—will progressively exact a toll upon the institutions that support modern democratic life. The more AI is deployed in our existing economic and social systems, the more the institutions will become ossified and delegitimized. Regardless of whether tech companies intend this destruction, the key attributes of AI systems are anathema to the kind of cooperation, transparency, accountability, and evolution that give vital institutions their purpose and sustainability. In short, AI systems are a death sentence for civic institutions, and we should treat them as such.
«
«
The United States needs fewer bus stops • Works in Progress Magazine
Nithin Vejendla:
»
When people talk about improving transit, they mention ambitious rail tunnels and shiny new trains. But they less often discuss the humble bus – which moves more people than rail in the US, the EU, and the UK – and whose ridership has bounced back more quickly after Covid than rail.
The problem with buses is that they are slow. For example, buses in New York City and San Francisco crawl along at a paltry eight miles per hour, only about double walking speeds in the fastest countries. There are lots of ways to speed up buses, including bus lanes and busways, congestion pricing, transit-priority signals, and all-door boarding. But one of the most powerful solutions requires no new infrastructure or controversial charges and has minimal cost: optimizing where buses stop.
Buses in some cities, particularly those in the US, stop far more frequently than those in continental Europe. Frequent stopping makes service slower, less reliable, and more expensive to operate. This makes buses less competitive with other modes, reducing ridership. This is why, despite having fewer bus stops, European buses have a higher share of total trips than American ones.
Bus stop balancing involves strategically increasing the distance between stops from 700–800 feet (roughly 210–240 metres), common in older American cities or in London, to 1,300 feet, closer to the typical spacing in Western Europe, such as in Hanover, Germany. Unlike many transit improvements, stop balancing can be implemented quickly, cheaply, and independently by transit agencies. By removing signs and updating schedules, transit agencies can deliver faster service, better reliability, and more service with the same resources.
American bus stops are often significantly closer together than European ones. The mean stop spacing in the United States is around 313m, which is about five stops per mile. However, in older, larger American cities, stops are placed even closer. In Chicago, Philadelphia, and San Francisco, the mean spacing drops down to 223m, 214m, and 248m respectively, meaning as many as eight stops per mile. By contrast, in Europe it’s more common to see spacings of 300m to 450m, roughly four stops per mile. An additional 500 feet takes between 1.5 and 2.5 minutes to walk at the average pace of 2.5 to 4 miles per hour.
«
As you can imagine, having lots of bus stops slows a bus down, and means drivers (paid by the hour) travel less far, so you don’t carry as many people and they get where they’re going more slowly.
unique link to this extract
Remembering the time an Islington Street was renamed… backwards • Londonist
Matt Brown:
»
Does anywhere else in London do this?
You might well have wandered along Muriel Street. It’s the leafy residential road you’re forced to join, if you follow the Regent’s Canal east of King’s Cross (the canal heads into its Islington tunnel at this point).
Head north along Muriel Street and pass through a pedestrianised cutting, and you’ll emerge onto Leirum Street. It surely cannot be a coincidence that Leirum is Muriel backwards. What’s going on?
Turns out that both streets were once part of a united Muriel Street. However, the ‘pedestrian bit’ in the middle was causing confusion, and could lead to potential delays for emergency vehicles attending the wrong section.
In 2013, the Barnsbury Estate’s Tenant Management Association wrote to Islington Council requesting a change. Rather than invoking a new name, however, it was decided to simply run the letters backwards. The northern stretch thus became Leirum Street.
“It is ridiculous and it looks like a big balls up,” long-term resident Kath Wardely told the Evening Standard at the time. “I think they got the stencil the wrong way round and now they are trying to cover it up.”
The tenant association probably wanted to say: “That’s not how stencils work, Kath”, but instead responded to say that residents had been consulted in advance. Islington Council also chimed in to confirm that “Some residents thought it was a cock-up, but it was a quite deliberate decision.”
«
That’s rather clever. As long as you pick a road that’s pronounceable backwards.
unique link to this extract
To those who fired or didn’t hire tech writers because of AI • passo.uno
Fabrizio Ferri Benedetti:
»
You might think that the plausible taste of AI prose is all you need to give your products a voice. You paste code into a field and something that resembles docs comes out after a few minutes. Like a student eager to turn homework in, you might be tempted to content yourself with docs theatre, thinking that it’ll earn you a good grade. It won’t, because docs aren’t just artifacts.
When you say “docs”, you’re careful to focus on the output, omitting the process. Perhaps you don’t know how docs are produced. You’ve forgotten, or perhaps never knew, that docs are product truth; that without them, software becomes unusable, because software is never done, is never obvious, and is never simple. Producing those docs requires tech writers.
Tech writers go to great lengths to get the information they need. They write so that your audience can understand. They hunger for clarity and meaning and impact. They power through weeks full of deadlines, chasing product news, because without their reporting, most products wouldn’t thrive; some wouldn’t even exist. Their docs aren’t a byproduct: they tie the product together.
An LLM can’t do all that, because it can’t feel the pain of your users. It can’t put itself into their shoes. It lacks the kind of empathy that’s behind great help content. It does not, in fact, have any empathy at all, because it cannot care. You need folks who will care, because content is a hairy beast that can only be tamed by agents made of flesh and capable of emotions: humans.
«
This is certainly true about writing technical documents. You need people to have understood what’s going on. An LLM won’t do that.
unique link to this extract
| • Why do social networks drive us a little mad? • Why does angry content seem to dominate what we see? • How much of a role do algorithms play in affecting what we see and do online? • What can we do about it? • Did Facebook have any inkling of what was coming in Myanmar in 2016? Read Social Warming, my latest book, and find answers – and more. |
Errata, corrigenda and ai no corrida: none notified
The Overspill: when there's more that I want to say 