The ban on tetraethyl lead in petrol did make a difference, according to hair samples. CC-licensed photo by Steve Snodgrass 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 10 links for you. Breathe easy. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.
AI use in breast cancer screening cuts rate of later diagnosis by 12%, study finds • The Guardian
Tobi Thomas:
»
The use of artificial intelligence in breast cancer screening reduces the rate of a cancer diagnosis by 12% in subsequent years and leads to a higher rate of early detection, according to the first trial of its kind.
Researchers said the study was the largest to date looking at AI use in cancer screening. It involved 100,000 women in Sweden who were part of mammography screening and were randomly assigned to either AI-supported screening or to a standard reading by two radiologists between April 2021 and December 2022.
The AI system worked by analysing the mammograms and assigning low-risk cases to a single reading and high-risk cases to a double one by radiologists, as well as highlighting suspicious findings to support radiologists.
Mammography screening supported by AI reduced cancer diagnoses in the years after a breast screening appointment by 12%, according to the research, published in The Lancet. There were 1.55 cancers per 1,000 women in the AI-supported group compared with 1.76 cancers per 1,000 women in the control group.
More than four in five cancer cases (81%) in the AI-supported mammography group were detected at the screening stage, compared with just under three quarters (74%) in the control group, and there were also almost a third (27%) fewer aggressive sub-type cancers in the AI group compared with the control.
Dr Kristina Lång, from Lund University in Sweden and the lead author of the study, said that AI-supported mammography could help detect cancers at an early stage, but that there were caveats.
«
As it’s worth bearing in mind, the AI being used here is the worst it’s ever going to be. From here it’ll only get better. And AI is going to touch more and more lives – hopefully, improving them.
unique link to this extract
Spotlighting The World Factbook as we bid a fond farewell • Simon Willison’s Weblog
Simon Willison:
»
Somewhat devastating news today from CIA:
»
One of CIA’s oldest and most recognizable intelligence publications, The World Factbook, has sunset.
«
There’s not even a hint as to why they decided to stop maintaining this publication, which has been their most useful public-facing initiative since 1971 and a cornerstone of the public internet since 1997.
In a bizarre act of cultural vandalism they’ve not just removed the entire site (including the archives of previous versions) but they’ve also set every single page to be a 302 redirect to their closure announcement.
The Factbook has been released into the public domain since the start. There’s no reason not to continue to serve archived versions – a banner at the top of the page saying it’s no longer maintained would be much better than removing all of that valuable content entirely.
Up until 2020 the CIA published annual zip file archives of the entire site. Those are available (along with the rest of the Factbook) on the Internet Archive.
I downloaded the 384MB .zip file for the year 2020 and extracted it into a new GitHub repository, simonw/cia-world-factbook-2020. I’ve enabled GitHub Pages for that repository so you can browse the archived copy at simonw.github.io/cia-world-factbook-2020/.
«
Peculiar, but then the entire Trump administration – as mentioned here before – seems dedicated to tearing down absolutely everything built up over the past 50 years.
unique link to this extract
Why Waymo is having a hard time time stopping for school buses • The Verge
Mack DeGeurin, on how Waymos have 23 incidents where Waymo robotaxis haven’t fully stopped (as they’re legally required to) for school buses which are loading and unloading, including one low-speed collision with a child:
»
Navigating around school buses is one of the more dangerous aspects of driving, for both humans and robots alike. NHTSA attributed 61 fatalities to vehicles illegally passing school buses between 2000 and 2023, almost half of whom were pedestrians under the age of 18. That danger has less to do with the bus drivers themselves, who are typically licensed and careful, and more to do with the chaotic, improvisational nature of the situation. Buses are often double-packed, and kids, being kids, might not wait to cross the street when they are supposed to.
“Waymos are having an issue because every driver has issues around school buses,” Ju said
As a result, drivers navigating around buses need to rely on experience and intuition in addition to following the firm set of rules they learned in driver’s ed. That kind of common-sense logic, which comes naturally to skilled human drivers, Wu says, is particularly challenging for self-driving cars.
“There’s all these moments in time where you actually have to make a judgment call between different things that you’re supposed to do,” Ju said.
On a technical level, there may be more at play. According to George Mason University professor and director of the Mason Autonomy and Robotics Center Missy Cummings, the apparent spike in robotaxi safety issues involving school buses may be linked to what she describes as Waymo’s increasing shift away from traditional, modular machine-learning toward a greater emphasis on end-to-end learning, a technology she calls “faddish” and “still nascent.”
Waymo, publicly, says it uses a mixture of the two, but some speculate that that balance is shifting.
Earlier autonomous-vehicle systems relied on more conservative, layered architectures, with separate modules responsible for detecting objects, classifying them, and applying explicit safety rules governing how a vehicle should respond. End-to-end learning, by contrast, collapses much of that process into a single model that takes in all of the information gathered by a car’s sensors at once and produces driving decisions probabilistically, based on patterns learned from large swaths of human driving behavior. The result is something that can seem more “natural” and humanlike, though Cummings argues it can also introduce additional risk, especially in high-stakes scenarios like school bus stops.
The safety incidents demonstrate “all the hallmarks of problems when you change architecture,” Cummings said. “I suspect many [robotaxi companies] are doing it.”
«
New approaches to trolley problems bring new outcomes.
unique link to this extract
OpenAI is hoppin’ mad about Anthropic’s new Super Bowl TV ads • Ars Technica
Benj Edwards:
»
[OpenAI CEO Sam] Altman opened his lengthy post on X by granting that the ads were “funny” and that he “laughed.” But then the tone shifted. “I wonder why Anthropic would go for something so clearly dishonest,” he wrote. “We would obviously never run ads in the way Anthropic depicts them. We are not stupid and we know our users would reject that.”
He went further: “I guess it’s on brand for Anthropic doublespeak to use a deceptive ad to critique theoretical deceptive ads that aren’t real, but a Super Bowl ad is not where I would expect it.”
Altman framed the dispute as a fight over access. “More Texans use ChatGPT for free than total people use Claude in the US, so we have a differently shaped problem than they do,” he wrote. He then accused Anthropic of overreach: “Anthropic wants to control what people do with AI,” adding that Anthropic blocks “companies they don’t like from using their coding product (including us).” He closed with: “One authoritarian company won’t get us there on their own, to say nothing of the other obvious risks. It is a dark path.”
OpenAI CMO Kate Rouch posted a response, calling the ads “funny” before pivoting. “Anthropic thinks powerful AI should be tightly controlled in small rooms in San Francisco and Davos,” she wrote. “That it’s too DANGEROUS for you.”
Anthropic’s post declaring Claude ad-free does hedge a bit, however. “Should we need to revisit this approach, we’ll be transparent about our reasons for doing so,” Anthropic wrote.
OpenAI President Greg Brockman pointed this out on X, asking Anthropic CEO Dario Amodei directly whether he would “commit to never selling Claude’s ‘users’ attention or data to advertisers,’” calling it a “genuine question” and noting that Anthropic’s blog post “makes it sound like you’re keeping the option open.”
«
Very generous of OpenAI to take so much notice of Anthropic; the equivalent of standing in the street shouting “I’M NOT OFFENDED, YOU KNOW”. (Internet denizens might pick between a couple of @dril posts.)
However Brockman is surely right. Like Google before it (thanks @wendyg), Anthropic is taking a stance it won’t be able to keep.
unique link to this extract
Calls to halt UK Palantir contracts grow amid “lack of transparency” over deals • The Guardian
Robert Booth:
»
Labour should halt public contracts with the US tech company Palantir, opposition politicians have said, amid growing concern at the lack of government transparency over dealings with the company and Peter Mandelson.
Since 2023, Palantir has secured more than £500m in contracts with the NHS and the Ministry of Defence (MoD), while it employed Global Counsel, the lobbying firm founded by Mandelson. Emails released by the US Department of Justice show Mandelson sought help from Jeffrey Epstein to find “rich individuals” as clients.
The government has for months blocked attempts by MPs and campaigners to scrutinise Palantir’s deals. Requests for information about meetings between the company’s leadership with Keir Starmer and the former prime minister Boris Johnson were among those that have been refused.
With Palantir now expanding its AI-powered technology into British policing, the government is facing calls to freeze its involvement with the Denver-based company, which was co-founded by the Donald Trump-backing billionaire Peter Thiel, who also had a relationship with Epstein. It also provides its military technology to the Israel Defense Forces and to Trump’s ICE immigration crackdown.
On Thursday, Martin Wrigley MP, a Liberal Democrat member of the Commons technology select committee, called for a parliamentary debate on “the suitability of Palantir” as a supplier to critical national infrastructure. Wrigley told the Guardian: “I would halt any further contracts with Palantir until we have a clear picture of how these [existing contracts] came about.”
«
This is going to be impossible for the government to resist, surely. Peter Mandelson now has the reverse Midas touch, so anything he has been involved in will, or should, be viewed a priori as unsavoury. And as long as the government refuses to allow more transparency, the opposition (and some awkward government MPs) will keep punching the Mandelson bruise.
unique link to this extract
Does intermittent fasting live up to the hype? • The New York Times
Sally Adee:
»
In 2013, a British journalist and a doctor introduced an obscure dietary protocol to the wider culture. The idea was simple: Two days a week, eat almost nothing — fewer than 600 calories. The rest of the time, eat normally.
The writers, Dr. Michael Mosley and Mimi Spencer, claimed “The Fast Diet” could help you shed fat, reverse Type 2 diabetes and stave off age-related diseases of mind and body. Early studies had shown it had outsize benefits in lab mice, and scientists were enthusiastic about its prospects for humans.
…The most common claim about intermittent fasting is that it’s a better way to lose weight than other diets. Early mouse and rat experiments suggested that something interesting was going on beyond simple calorie restriction. The animals lost weight and stayed healthier than mice that ate normally, no matter how many calories they binged between fasts.
But in humans, the idea that intermittent fasts offer special weight loss benefits “really hasn’t been borne out by the data,” said James Betts, a professor of metabolic physiology at the University of Bath in the United Kingdom.
At first glance, it would appear that the research is mixed — some studies find that intermittent fasting is no more effective for weight loss than other diets, while others report a small additional benefit. But many of the latter are based on small, low quality studies, Dr. Betts said.
He and his colleagues have reviewed many studies published in the past few years and found that much of the pro-fasting research is corrupted by studies that make massive errors, like counting study subjects twice or misinterpreting their own data.
“There are only probably about 20 or 30 studies out there that are good,” Dr. Varady said. “It’s not bad, but it’s like all diets,” she added. “It can’t produce more than 5% weight loss.”
The diet’s major selling point is that it’s easier to stick to than simple calorie restriction, but even this may be overstated, according to two recent papers. In other words, Dr. Harvie said, people seem to get bored with intermittent fasting just as they get bored with other diets.
«
A nation turns its lonely eyes back to GLP-1s.
unique link to this extract
The coming crypto apocalypse • Nouriel Roubini
Nouriel Roubini:
»
As of this writing, Bitcoin is down 35% from its October peak, below where it was when Trump was elected, and the $TRUMP and $MELANIA meme coins are down 95%. Every time gold has spiked in response to trade or geopolitical ructions over the past year, Bitcoin has fallen sharply. Far from being a hedge, it is a way to leverage risk, showing a strong correlation with other risky assets like speculative stocks.
Calling Bitcoin or any other crypto vehicle a “currency” has always been bogus. It is neither a unit of account, a scalable means of payment, nor a stable store of value. Even though El Salvador made Bitcoin legal tender, it accounts for less than 5% of transactions for goods and services. Crypto is not even an asset, as it has no income stream, function, or industrial or real-world use (unlike gold and silver).
Seventeen years after Bitcoin’s launch, the one and only “killer app” in crypto is the stablecoin: a digital version of old-fashioned fiat money, which the financial and banking industry already digitalized decades ago. Yes, whether digital money and financial services should be on a blockchain (distributed ledger) or a traditional double-ledger platform remains a question.
But 95% of “blockchain” monies and digital services are blockchain in name only. They are private rather than public, centralized rather than decentralized, permissioned rather than permissionless, and validated by a small group of trusted authenticators (as in traditional digital finance and banking) rather than by decentralized agents in jurisdictions with no rule of law.
True decentralized finance will never reach scale. No serious government – not even the Trump administration – will ever allow full anonymity of monetary and financial transactions, because that would be a boon for criminals, terrorists, rogue states, non-state actors, human traffickers, assorted crooks, and tax dodgers.
Moreover, because digital wallets and regulated exchanges must be subjected to standard anti-money laundering and know-your-customer (AML/KYC) rules, it is not even clear that transaction costs through permissioned and private “blockchains” are any lower – especially now that traditional financial ledgers have improved with real-time settlement and faster clearing tools. The future of money and payments will feature gradual evolution, not the revolution that crypto-grifters promised.
«
Bitcoin’s been dropping further on Thursday, but I long since realised that – as the saying goes – the market can stay irrational longer than I can predict sensible things for it.
unique link to this extract
Flawed economic models mean climate crisis could crash global economy, experts warn • The Guardian
Damian Carrington:
»
As the world speeds towards 2C of global heating, the risks of extreme weather disasters and climate tipping points are increasing fast. But current economic models used by governments and financial institutions entirely miss such shocks, the researchers said, instead forecasting that steady economic growth will be slowed only by gradually rising average temperatures. This is because the models assume the future will behave like the past, despite the burning of fossil fuels pushing the climate system into uncharted territory.
Tipping points, such as the collapse of critical Atlantic currents or the Greenland ice sheet, would have global consequences for society. Some are thought to be at, or very close to, their tipping points but the timing is difficult to predict. Combined extreme weather disasters could wipe out national economies, the researchers, from the University of Exeter and financial thinktank Carbon Tracker Initiative, said.
Their report concludes governments, regulators and financial managers must pay far more attention to these high impact but lower likelihood risks, because avoiding irreversible outcomes by cutting carbon emissions is far cheaper than trying to cope with them.
“We’re not dealing with manageable economic adjustments,” said Dr Jesse Abrams, at the University of Exeter. “The climate scientists we surveyed were unambiguous: current economic models can’t capture what matters most – the cascading failures and compounding shocks that define climate risk in a warmer world – and could undermine the very foundations of economic growth.”
“For financial institutions and policymakers, it’s a fundamental misreading of the risks we face,” he said. “We are thinking about something like a 2008 [crash], but one we can’t recover from as well. Once we have ecosystem breakdown or climate breakdown, we can’t bail out the Earth like we did the banks.”
«
In the SF novelist John Brunner (him again!) book The Sheep Look Up, petrol cars are set upon by protesters who destroy them while demanding better environmental standards. The only problem is that everyone’s so comfortable with their petrol/diesel cars that you can’t imagine that happening; it would take too much willingness to tolerate self-sacrifice, and we have become worse and worse at that.
Maybe a sort of war footing is what’s needed. But you need to be submerged in the conflict for that to work.
unique link to this extract
Memory prices surge up to 90% from Q4 2025 • Counterpoint Research
»
Memory prices have risen by 80%-90% QoQ in Q1 2026 so far, according to the February issue of Counterpoint’s Memory Price Tracker, marking an unprecedented and record-breaking surge. The primary driver of this rise is the sharp price increase in general-purpose server DRAM. Furthermore, NAND, which was relatively quiet in Q4, is also seeing a parallel jump of 80%-90% in the first quarter. Combined with the rising prices of some HBM3e products, the market is witnessing a full-throttle upward trend across all segments.
In server-grade memory, for instance, the price of 64GB RDIMM has surged from a fixed contract price of $450 in the fourth quarter to over $900 in the first quarter, and it appears likely to surpass the $1,000 mark in the second quarter.
Senior Analyst Jeongku Choi emphasized, “For device manufacturers, this is a double whammy – rising component costs and weakened consumer purchasing power will likely slow the demand as the quarter progresses. This calls for OEMs to change procurement patterns or focus on premium models to justify the higher price by delivering more value to consumers.”
Smartphone manufacturers are reducing the DRAM content or substituting TLC SSDs with more cost-effective QLC alternatives. Simultaneously, there is a clear trend of declining orders for LPDDR4, which is currently in short supply, and increasing orders for LPDDR5, supported by the rollout of new entry-level chipsets that are compatible with the latest DRAM standard.
Choi further noted, “The memory profitability is expected to reach unprecedented levels. DRAM operating margins have already reached the 60% range in Q4 2025, marking the first time margins for general-purpose DRAM have surpassed those of HBM. The first quarter of 2026 is set to be the period where DRAM margins exceed their historical peaks for the first time.
«
More than doubling over the course of a few months. Anyone who didn’t have their RAM supply all tied up months ago is truly screwed now. In fact some memory manufacturers might be looking to simply break contracts with smaller buyers in order to serve more promising customers now. Hence PC makers looking for Chinese RAM.
unique link to this extract
A century of hair samples proves leaded gas ban worked • Ars Technica
Jennifer Oullette:
»
The Environmental Protection Agency (EPA) cracked down on lead-based products—including lead paint and leaded gasoline—in the 1970s because of its toxic effects on human health. Scientists at the University of Utah have analyzed human hair samples spanning nearly 100 years and found a 100-fold decrease in lead concentrations, concluding that this regulatory action was highly effective in achieving its stated objectives. They described their findings in a new paper published in the Proceedings of the National Academy of Sciences.
We’ve known about the dangers of lead exposure for a very long time—arguably since the second century BCE—so why conduct this research now? Per the authors, it’s because there are growing concerns over the Trump administration’s move last year to deregulate many key elements of the EPA’s mission. Lead specifically has not yet been deregulated, but there are hints that there could be a loosening of enforcement of the 2024 Lead and Cooper rule requiring water systems to replace old lead pipes.
“We should not forget the lessons of history. And the lesson is those regulations have been very important,” said co-author Thure Cerling. “Sometimes they seem onerous and mean that industry can’t do exactly what they’d like to do when they want to do it or as quickly as they want to do it. But it’s had really, really positive effects.”
An American mechanical and chemical engineer named Thomas Midgley Jr. was a key player in the development of leaded gasoline (tetraethyl lead) because it was an excellent anti-knock agent, as well as the first chlorofluorocarbons (CFCs) like freon. Midgley publicly defended the safety of tetraethyl lead (TEL), despite experiencing lead poisoning firsthand. He held a 1924 press conference during which he poured TEL on his hand and inhaled TEL vapor for 60 seconds, claiming no ill effects. It was probably just a coincidence that he later took a leave of absence from work because of lead poisoning. (Midgley’s life ended in tragedy: he was severely disabled by polio in 1940 and devised an elaborate rope-and-pulley system to get in and out of bed. That system ended up strangling him to death in 1944, and the coroner ruled it suicide.)
«
Maybe there is a God, with a very warped sense of humour.
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 