The British Board of Film Censorship has reccruited AI to classify content on HBO Max in the UK. CC-licensed photo by Rachel 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. Unmentionable. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.
The RAM shortage could last years • The Verge
Terrence O’Brien:
»
According to Nikkei Asia, even as suppliers ramp up DRAM production, manufacturers are only expected to meet 60% of demand by the end of 2027. The SK Group chairman has even said that shortages could last until 2030.
The world’s largest memory makers — Samsung, SK Hynix, and Micron — are all working to add new fabrication capacity, but almost none of it will be online until at least 2027, if not 2028. SK opened a fab in Cheongju in February, but that is the only increase in production among the three for 2026.
Nikkei says that production would need to increase by 12% a year in 2026 and 2027 to meet demand. But according to Counterpoint Research, an increase of only 7.5% is planned.
The new facilities will primarily focus on producing high-bandwidth memory (HBM), which is used in AI data centers. With the companies already prioritising HBM over general-purpose DRAM used in computers and phones, it’s not clear how much these new fabs will help alleviate the price crunch facing consumer electronics.
«
Are we going to see adaptations in software to reduced (or not expanding) availability of RAM in most computers in response to this stress, like the adaptation from fossil fuel-driven vehicles to electric vehicles in response to the stress of the closure of the strait of Hormuz? The last situation that resembled it was moving to smartphones, which had less processing capacity, less storage and less RAM. The 20-year cycle comes around.
unique link to this extract
Europe has “maybe six weeks of jet fuel left”, energy boss warns • BBC News
Theo Leggett and Jemma Crew:
»
Europe has “maybe six weeks of jet fuel left”, the head of the International Energy Agency (IEA) has warned.
Stocks would reach a tipping point in June if Europe was unable to replace at least half of its imports from the Middle East, the organisation said in a report this week.
The Strait of Hormuz, a key route for jet fuel out of the Gulf, has been effectively closed by Iran for more than six weeks in response to US and Israeli attacks, sending the price rocketing and prompting fears of shortages.
IEA executive director Fatih Birol told the Associated Press there could soon be flight cancellations if supplies remained blocked. In its monthly oil market report, the IEA- which advises 32 member countries on energy supply and security – said exports from the Gulf region were the largest source of jet fuel to the global market.
Refineries in other major exporting countries, such as Korea, India and China were themselves highly dependent on crude oil imports from the Middle East.
As a result, the crisis “has thrown a proverbial wrench into the inner workings of the aviation fuel markets”, it said.
A spokesperson for the UK government told the BBC that it was working with fuel suppliers and airlines to “ensure people keep moving and businesses are supported”.“UK airlines are clear that they are currently not seeing disruption to supply,” they said.
«
Not currently. But six weeks will pass very rapidly, and with the weekend having passed with minimal traffic through the strait of Hormuz – four cruise liners, not much else – this is not going to be good. It bears repeating.
unique link to this extract
Geelong fire: blaze at Australian oil refinery to affect petrol supplies • BBC News
Lana Lam and Tiffanie Turnbull:
»
A major fire at one of Australia’s two oil refineries has been extinguished, but the damage has deepened fears over the nation’s petrol supplies amid a global fuel crunch.
Emergency crews rushed to Viva’s Corio oil refinery in Geelong, south-west of Melbourne, just before midnight local time (1400 GMT) on Wednesday, after reports of explosions and flames.
The blaze was put out on Thursday after burning for 13 hours. No one was injured, with dozens of workers on site when it broke out evacuated safely.
The refinery – which produces 50% of Victoria’s fuel and 10% of the nation’s – is still partially operational but the government has warned of impacts to petrol production.
Energy Minister Chris Bowen said the fire was “not great timing” with Australia’s fuel supplies under pressure since war broke out in Iran, creating a global oil crisis.
The price of diesel in Australia has doubled in recent weeks, with fuel stations reporting shortages amid reports of panic buying, while airlines are cutting back some services as jet fuel costs rise.
“This is not a positive development, but obviously there’s a long way to go in terms of working out just what the impact is,” Bowen told Nine’s Today show on Thursday, adding that he was working closely with the company.
…Australia relies heavily on imported refined fuels – primarily from countries like Singapore, South Korea and Malaysia – which should mean any impact the fire has on domestic fuel availability is limited, experts say.
«
Limited, but definitely not zero.
unique link to this extract
Chinese trucks could go 100% electric, halving road transport oil use: industry • South China Morning Post
Dannie Peng:
»
Industry estimates suggest that heavy-duty cargo traffic in China is on track to become nearly 100% electric, a transition that could halve the country’s road transport oil consumption.
Speaking at a forum on intelligent electric vehicle development in Beijing on April 11, Liang Linhe, chairman of Sany Truck – a subsidiary of the Changsha-headquartered multinational Sany Group – said China’s heavy truck sector could eventually be almost entirely electrified, although he did not put a timeline on the transition.
Such a move, he noted, had the potential to cut fuel demand within the road transport sector by about half.
Liang said cost was the overriding factor behind the uptake of electric heavy trucks – vehicles that are engineered for hauling, towing and moving massive volumes of cargo, equipment or raw materials.
“As a means of production, economics is the core consideration,” Liang said. He estimated that much lower transport costs would drive market penetration to a point “even potentially approaching 100%, leaving little room for diesel trucks”.
Given that heavy trucks account for a substantial share of China’s petroleum and diesel consumption, the shift from conventional diesel to electric power is widely viewed as a crucial lever for reducing greenhouse gas emissions and advancing the clean energy transition in China.
Liang said the annual carbon emissions from a single diesel heavy truck were equivalent to those of about 100 petrol-powered cars.
…According to the General Administration of Customs, China imports 42% of its crude oil from Gulf nations. And, transport accounts for around half of China’s oil consumption.
«
Provide the right incentives, and everyone stampedes to the cheaper option. The irony of Trump, the anti-renewables guy, provoking this shift is on a planetary scale.
unique link to this extract
Whither Microsoft? A view from the neighborhood • GeekWire
Feroze Motafram:
»
I live in Sammamish, in the shadow of Microsoft’s looming presence. Microsoft employees are my neighbors, my social circle, the people I run into at weekend gatherings. Over time I noticed that conversations with them had a distinctive gravitational pull — always inward, toward reorgs, internal politics, who reports to whom now, who’s ascendant, who’s out. Customers were rarely part of the conversation. This usually means navigating the organization has become more consuming than building anything within it.
Microsoft’s stock decline and the softening of real estate in this corridor (both affecting me personally) were the prompts to write it down. The material was already sitting in front of me.
…Microsoft stock declined roughly 25% in Q1 2026, representing its worst quarterly performance since the 2008 financial crisis despite blockbuster results. The market may overreact, but it is not stupid.
A striking portion of Microsoft’s forward revenue backlog is tied to a single counterparty, OpenAI, an unprofitable startup that has since signed a landmark cloud agreement with Amazon, directly challenging the Azure exclusivity Microsoft had treated as a cornerstone of its AI strategy. Meanwhile, Microsoft is building its own internal AI model as a hedge, an expensive bet layered on top of an already expensive bet.
But the part that does not show up in an earnings report may be the more consequential story. That is what I want to offer here.
…A significant proportion of Microsoft’s engineering talent (and the engineering talent of the broader Seattle tech corridor) consists of H-1B visa holders. These are exceptional professionals: highly educated, deeply skilled, often carrying decade-long career investments in the United States. They have built lives here. Many have children born here. They have been, in many cases, the intellectual engine of the products Microsoft is depending on to compete in the AI era.
That population is operating under a level of personal anxiety that is, in my observation, without modern precedent. Travel advisories from their own employers. A $100,000 petition fee for new visa applications. Proposed rule changes touching birthright citizenship. A policy environment that sends a clear and unambiguous message: your presence here is conditional, negotiable, and subject to revision without notice.
The behavioral consequence of that anxiety is not visible in a quarterly earnings report. But it is real and consequential. People operating under existential personal uncertainty do not take professional risks. They do not champion the bold new initiative. They do not volunteer for the high-visibility project that could fail. They execute reliably on what already exists and protect their position.
«
British censors adopt AI to age-rate streamer favourites • The Observer
Vanessa Thorpe:
»
For four points, to what quiz question is “Natasha Kaplinsky” the unexpected right answer? There are, possibly, a few correct responses, but the one I have in mind asks: Which former news reader and Strictly Come Dancing winner now holds the position of Britain’s official film censor? Since 2022, as sharp-eyed cinema-goers will have spotted, it has been Kaplinsky’s signature on the ratings certificate that pops up before any public film screening.
From this month on, however, Kaplinsky will have a new resource to call upon when rating films, television shows or cartoons. Aside from her team of “compliance officers” at the British Board of Film Classification (BBFC), she has been joined by a freshly developed AI film censor.
The BBFC has confirmed the successful deployment of its first AI tool, helping it to classify the whole of HBO Max’s library of entertainment for the UK. The bot has already aided its officers in sticking on BBFC age ratings and drawing up “bespoke” content advice, along classification guidelines.
By using AI, the BBFC says, it has dealt with the streamer’s entire new UK catalogue, from Game of Thrones to Succession, in only six months. It says the same volume of content would probably have taken an officer 1,570 working days to process. David Austin, chief executive of the BBFC, sees it as “a major step forward in how we support families to make safe and informed viewing decisions”.
The AI tool works by highlighting each compliance issue in a film, show or documentary, including scenes of violence, nudity or bad language, so human officers can review the sequence. The BBFC is emphasising that the final age ratings and the advice it hands down will still be decided by trained officers, and that classification standards should not change.
«
Operator of slop farm that plagiarizes local journalism boasts that it’s attracting millions of readers • Futurism
Maggie Harrison Dupré:
»
On Tuesday evening, we published an original interview with a researcher who had recently coauthored an intriguing study about the effects of AI on users’ cognition.
A news site called National Today quickly sprang into action: by ten o’clock that night, it had published a piece that was obviously a reworded version of our story, including a direct quote from the interview we’d conducted. But instead of crediting us as the source of the information, as would be conventional, National Today made no mention of Futurism, and didn’t even link to our article. Instead, it presented the reporting as if it were the original source.
In other words, the National Today piece — which bears no byline — is blatant plagiarism. And this isn’t the first time this has happened.
…We’re not the only target. Once we started looking into National Today, we realized that it’s doing the same thing to countless other publications, ranging from top newspapers to local newsrooms across the country: stealing their original reporting and using it to publish a torrent of what appear to clearly be AI-generated articles, complete with bizarre errors and hallucinations. The scope is immense. We tried to count how many it published in a single day, but lost count around 300.
…In addition to the incredible speed at which it churns out all this slop, National Today is full of comically terrible errors that make it difficult to believe that any human is even skimming its articles before publication.
One issue is that the AI generating all the articles often seems to get confused and replace real people’s names with “Jane Doe.” In one recent piece, National Today reported that a NASA astronaut aboard Artemis 2 named “John Doe” had dedicated a Moon crater to his deceased wife, “Jane Doe.”
…According to a description on [branding and PR company] TOP’s site, National Today exists to help brands “Create Ownable Viral Moments for your brand” and “reach 10M consumers, 100K media outlets, and 10K influencers across traditional, digital, and social media.”
In other words, TOP seems to be saying, National Today is a marketing vehicle for its clients — made possible by the theft of local journalism on an almost incomprehensible scale.
«
They didn’t manage to get TOP to respond. According to the company’s website, it’s got offices in 18 locations, 14 of them in the US. Was it not possible to drop by one of them and ask some questions? Or find a telephone number for one of the employees, or even the chief exec Benjamin Kaplan?
unique link to this extract
Fire alert: the fake “Amazon TV stick” that opens the door to fraudsters • The Guardian
Shane Hickey, back in November:
»
The sticks are sold through social media and illegal streaming sites, often advertised as being jailbroken with promises of free TV shows, sports and films.
The potential for fraud happens the minute you connect the “dodgy stick” with a laptop or TV, according to Rob Shapland of Cyonic Cyber, an ethical hacker who shows companies where there are gaps in their security systems.
“Alongside the stream [of TV or sports] it will also install some malware on to your computer and give the criminal direct access to your computer so [they] could use it as if they were sitting there,” he says.
“Or they can install keyloggers which will record any password you are typing. So when you are accessing online banking, it will record your banking passwords. You are essentially volunteering to have your laptop hacked in many cases.”
Often the stick will come with instructions which appear to be how to install the software but are actually ways to bypass virus detectors.
When you connect the device with a TV, it can access other devices, such as laptops, through the home Wi-fi network they are all attached to, says Shapland.
“It might ask you to log in with your Google account and then you are giving your credentials to the app and that is then sent off to the criminals. Most people tend to reuse the same password so once they have one password, they have access to a hundred different things.”
Some sticks may ask for a small monthly subscription. Setting up the payment means criminals can use your credit card details in any way they want.
«
This article was mentioned on the latest episode of ATP, which also linked to this Brian Krebs article from the same period. Free or very cheap is tempting when streaming subscription prices keep going up, but the potential for these devices to be completely untrustworthy is sky-high.
Dozens of AI disease-prediction models were trained on dubious data • Nature
Mohana Basu:
»
Dubious data sets are being used to train artificial-intelligence models that are designed to predict people’s risk of stroke and diabetes, researchers report in a preprint on medRxiv. Some of the models seem to have been used in clinical settings, although it’s not clear whether this has led to flawed diagnoses. At least two journals are investigating studies that used these data sets.
Adrian Barnett, a statistician at the Queensland University of Technology in Brisbane, Australia, and his colleagues identified 124 peer-reviewed papers that report using one of two open-access health data sets to train machine-learning models that provide little information about where the data came from.
An analysis revealed multiple oddities that would not be expected for data from real people, leading Barnett and his colleagues to suspect that the data could have been fabricated. “It was an enormous surprise to come across something like that,” Barnett says.
At least two of the models have been used in hospitals, in Indonesia and Spain. One has also been documented in a medical-device patent application filed in 2024, and two are publicly available web tools that allow people to check their risk level by uploading information about themselves.
“Prediction models trained on provenance-unknown data have no place in clinical decision-making. They are intrinsically unreliable,” says Soumyadeep Bhaumik, a public-health researcher at the George Institute for Global Health in Sydney, Australia. If the tools do not use real-world data, they are likely to make incorrect predictions and lead clinicians to make inappropriate decisions, such as prescribing treatments unnecessarily or not prescribing them when they are needed, he says.
Institutions and funders must insist that researchers disclose the source of data used to train AI models for medical applications, and journals should reject papers that fail this requirement, says Bhaumik.
«
We’re having to update the models of how to use these almost in real time as AI spreads everywhere.
unique link to this extract
“Breakthrough” Alzheimer’s drugs unlikely to benefit patients, report suggests • BBC News
James Gallagher:
»
Influential analysis has concluded that “breakthrough” Alzheimer’s drugs are unlikely to benefit patients.
Researchers said the impact was “well below” what was needed to make a difference to dementia patients’ lives. However, their report has also provoked a vicious backlash from equally esteemed scientists who label it as fundamentally flawed.
The drugs have been licensed around the world. But the NHS in the UK won’t pay for them and an 18-month course of the drugs plus medical care would set you back a hefty £90,000 privately. They would be unaffordable for most so even if you had the money, are they worth paying for?
The drugs attack a sticky gunk – called beta amyloid – that builds up in the spaces between brain cells in Alzheimer’s disease. Antibodies – similar to those the body makes to attack viruses or bacteria – have been engineered to spot the amyloid and clear it from the brain.
For years the approach failed, but trials of two recent drugs called donanemab and lecanemab showed they could slow the pace of cognitive decline. This was a landmark moment as it was the first time any drug had slowed the destruction of the brain in Alzheimer’s disease.
The Cochrane Collaboration, which rigorously and independently analyses medical data, looked at 17 studies, involving 20,342 volunteers, of drugs that remove amyloid from the brain. Overall, they concluded the approach does slow Alzheimer’s disease, but not by enough to make a meaningful difference to patients. At the same time the medicines came with a risk of brain swelling and bleeding. They also need to be given every two to four weeks and at a high cost.
«
The positive side – minimally – is that people are being spared the expense of drugs that the review shows don’t really help.
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 