The mathematics of queueing shows how even an apparently well-staffed service can rapidly be overwhelmed by demand. CC-licensed photo by oatsy40 on Flickr.
A selection of 9 links for you. Waiting. 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 UK paid £4.1 million for a bookmarks site • Mahad Kalam
Mahad Kalam:
»
The UK Government recently unveiled its “AI Skills Hub“, which wants to provide 10 million workers with AI skills by 2030. The main site was delivered by PwC for the low, low price of.. £4.1m (~$5.6m).
It is not good. Like, at all – the UI is insanely bad and it’s clear that this was just a vibecoded site. To be fair, this is the AI Skills Hub, but c’mon, where is the pride in your work? I would be ashamed to even release this as a prototype!
PwC didn’t even write any of the course content! The only thing the Skills Hub does is link out to external pages, like Salesforce’s free Trailhead learning platform:
Note that I’m fairly certain this course already existed before the contract was even awarded, so all the site does is.. link out to other sites?
PwC itself also admits that the site does not properly meet accessibility standards. Even for those without a disability, the lack of here in this regard means that the site can be very confusing and buggy as a result.
The site has a course on “AI and intellectual property”. One thing it mentions is fair use. Except that fair use is not a thing in the UK – that’s a US concept! The UK uses what’s known as “fair dealing”, which is more restrictive than fair use, so the details here are plain wrong.
The interface for this website has also not been clearly thought out – one glaring example is the process of actually enrolling in a course. On the course page, the “Enroll Now” button is tiny, and if you don’t see it and try scrolling down to the bottom, you will find yourself nothing but a comment section! [The correct word is “enrol”; one is enrolled, but the active verb only has only L – Overspill Ed]
Then you have other bugs too, like the “Skills & Training Gap Analysis” – which is linked at the top of the site! – apparently being closed off to the public for no reason:
To be honest, seeing this made me angry.
I’m angry at the sheer wastefulness of the UK Government here. Our public services are collapsing – while £4m is admittedly chump change for the UK government, there are real people behind these numbers – families waiting months for NHS appointments, children in crumbling schools, vulnerable people not getting the care they need. The waste feels particularly galling when you realise that almost no one will actually use this site!
«
Absolutely right. This is a shameful waste of money, and the AI technology minister Kanishka Narayan (appointed in September) should be tearing people apart today for these failures.
unique link to this extract
Amazon cuts another 16,000 jobs • AP News
Kelvin Chan:
»
Amazon is slashing about 16,000 corporate jobs in the second round of mass layoffs for the ecommerce company in three months.
The tech giant has said it plans to use generative artificial intelligence to replace corporate workers. It has also been reducing a workforce that swelled during the pandemic.
Beth Galetti, a senior vice president at Amazon, said in a blog post Wednesday that the company has been “reducing layers, increasing ownership, and removing bureaucracy.”
The company did not say what business units would be impacted, or where the job cuts would occur.
The latest reductions follow a round of job cuts in October, when Amazon said it was laying off 14,000 workers. While some Amazon units completed those “organizational changes” in October, others did not finish until now, Galetti said.
She said U.S.-based staff would be given 90 days to look for a new role internally. Those who are unsuccessful or don’t want a new job will be offered severance pay, outplacement services and health insurance benefits, she said.
«
Completely unrelated: Amazon has spent $75m on access fees ($40m) and marketing ($35m) for the new “Melania” documentary, which is a complete and utter flop from the start. A typical documentary of this sort would cost $5m and $3.5m, respectively. Probably worth about 15,000 jobs.
unique link to this extract
How British queues got out of hand • Tim Harford
Tim Harford:
»
when bottlenecks feed into bottlenecks, some strategic thinking is required to fix the system. There is often more than one bottleneck in a congested system and opening that bottleneck will sometimes mean the same queue builds up somewhere else.
Second, the optimum queuing time probably isn’t zero. In most cases, demand arrives at irregular intervals and it is likely to be impractically wasteful to have so much capacity (so many doctors, so many ambulances, so many crown courts) that even after a sudden surge in demand, nobody has to wait.
That said, the optimum queuing time should probably be kept quite short. Imagine a situation where an emergency doctor can see four patients an hour and patients arrive every 15 minutes. At first everything is fine: every patient can be seen immediately. Then something goes wrong. Perhaps there’s a sudden rush, when five patients unexpectedly arrive together. Perhaps the doctor takes an hour off for lunch. The waiting time suddenly increases from nothing to an hour, even though the doctor is still seeing four patients an hour and four patients an hour are still arriving.
The moral of this very simple story is that even if the capacity of the system is equal to the demand for it, queues can grow and then stay at unpleasant lengths. What’s needed is a little extra capacity to work through the inevitable queues that build up from time to time. Unfortunately, systems under intense pressure rarely have a little extra capacity hanging around.
Third, it can be hard to increase the capacity of a system. Let’s say that we have one million nurses and each nurse trains for two years before working for 20. Arithmetically, that requires 100,000 nurses to be in training at any given moment. What if it is decided that we need 1.1 million trained nurses and we need them as soon as possible? That would require an immediate recruitment boost, doubling the number of nurses in training.
Would that be possible? Even though the expansion in nursing personnel seems modest, it requires nursing courses to double in size and then to shrink again after a couple of years. An even more dramatic expansion will be needed at the advanced training colleges at which the teachers of nursing are themselves trained. It might be easier to persuade nurses to stay a little longer in the profession or to recruit from the Philippines.
«
A neat musing on the challenge facing the NHS, and other public services, in trying to clear huge backlogs. Also explains why the unit of queueing load is the erlang.
unique link to this extract
As we breach 1.5 °C, we must replace temperature limits with clean-energy targets • Nature
Kwesi Quagraine, Mark Lynas and Erle Ellis:
»
On 28 October 2025, United Nations secretary-general António Guterres acknowledged that the totemic goal of the Paris climate agreement is going to be missed: “The truth is that we have failed to avoid an overshooting above 1.5°C in the next few years”.
Guterres was merely stating the obvious. In 2024, Earth’s global mean surface temperature averaged 1.55°C above pre-industrial levels, and the average for 2023–25 is 1.48°C, perilously close to the limit. Keeping to the Paris target now looks impossible by any realistic measure. Yet this moment should not invite despair. Instead, it demands an urgent reframing of how climate progress is measured and mobilized.
The world today looks very different from that in 2015 when the Paris goal was framed. Although emissions are still rising and global actions on climate change are slow, a lot of progress has been made. Clean energy is expanding rapidly and decarbonization, not fossil fuels, is the new ‘business as usual’. In the first three quarters of 2025, growth in clean electricity generation outpaced that in energy demand for the first time, implying that fossil fuels are being displaced.
We argue that the main focus of climate action in 2026 and beyond should be on accelerating the clean-energy revolution. And the rate at which clean energy displaces fossil fuels in the global economy should become the key measure of climate progress. Here we describe how such progress can be tracked and incentivized using a metric we call the clean-energy shift. Unlike chasing intangible temperature targets, cleaning up the energy sector is a more-focused battle that the world can win.
To move forwards, climate scientists and policymakers must first accept that the Paris 1.5°C target has outlived its usefulness. Although initially valuable as a unifying focus for international efforts to increase mitigation, continuing to emphasize a failed temperature target might produce more harm than good.
«
This is absolutely right. For years, the problem with warnings about climate change/global warming has been that they leave one asking: OK, but what do you expect me to do about it? How will we know if our government is doing the right thing?
A clean energy target – this proportion of production from renewables, that proportion of fossil fuel use removed – is far more comprehensible. People can even get involved by getting their own renewables.
unique link to this extract
Metaphors for biology: time • Asimov Press
Sam Clamons:
»
The interior of a cell is densely packed with millions of molecules vibrating, jostling, and moving about. Sugar molecules fly through a cell at 250 miles per hour, ricocheting off of ribosomes, organelles, cytoskeletal fibers, and enzymes. Indeed, every protein in the cell is hit by about 1013 water molecules each second. This chaos makes biology seem hopelessly convoluted. With everything moving so quickly, how can we begin to understand biomolecules?
As with other hard-to-intuit quantities in science, one could look up biological rates using resources like PubMed or BioNumbers, to discover facts like “water flows through aquaporin at 100 million molecules per second,” or “yeast transcribes RNA at 0.12 molecules per minute.” But knowing a number doesn’t necessarily give one a feel for it. Are those rates… fast? How do they compare to protein folding? Or enzymatic activity? Or squeezing a muscle?
In short, how fast do things in a cell happen, from the perspective of the molecules it’s made from?
We can answer this question with a quantitative metaphor, by visualizing the most important goings-on of a typical cell slowed down to speeds that are still accurate relative to one another, but matched to what we experience in the everyday world. The slowdown factor we pick should make it easy to understand the molecular machines that run our cells — proteins. Ideally, we would scale the fastest functionally important protein event to match the shortest unit of human perception.
As a representative “fastest functionally important protein event,” I’ve picked the opening of a membrane-bound ion channel — specifically, a potassium-gated ion channel.1 This protein channel opens and closes to allow potassium ions into the cell. This conformational change must happen very quickly. So, for the sake of our metaphors, let’s imagine slowing down the opening of this channel 10,000x, so it takes as long as the blink of an eye. If this were the case, then…
«
This is utterly mindblowing. The speed of ATP production (essential to cell energy) is surprisingly slow, relatively. It’s rather like the illustrations of “life as seen by a computer CPU” – where what feel like aeons are spent waiting for the human to press a key, any key. Your word processor thinks you are the most slothful thing in the universe.
unique link to this extract
Apple acquires Israeli audio AI startup Q.ai • Reuters
Stephen Nellis:
»
Apple on Thursday said it has acquired Q.ai, an Israeli startup working on artificial intelligence technology for audio.
Apple did not disclose terms of the deal for Q.ai, which was backed by venture capital firms Matter Venture Partners, Kleiner Perkins, Spark Capital, Exor and GV, formerly known as Google Ventures. The Financial Times reported it was worth nearly $2bn, a figure Reuters could independently verify.
Apple did not say how it will use Q.ai’s technology but said the startup has worked on new applications of machine learning to help devices understand whispered speech and to enhance audio in challenging environments.
Q.ai last year filed a patent application to use “facial skin micromovements” to detect words mouthed or spoken, identify a person and assess their emotions, heart rate, respiration rate and other indicators.
Q.ai’s 100 employees, including CEO Aviad Maizels and co-founders Yonatan Wexler and Avi Barliya, will join Apple, the companies said.
Maizels founded three-dimensional sensing firm PrimeSense and sold it to Apple in 2013. The PrimeSense deal eventually helped Apple move away from fingerprint sensors on its iPhones and toward facial recognition technology.
In a statement, Maizels said, “Joining Apple opens extraordinary possibilities for pushing boundaries and realizing the full potential of what we’ve created, and we’re thrilled to bring these experiences to people everywhere.”
Apple has been putting new AI features into its AirPods earbuds, last year introducing technology that allows them to translate speech between languages.
«
Apple’s second-largest (in terms of value) acquisition ever, after Beats – though of course the acquisitions of NeXT ($404m) and PA Semi ($278m) probably transformed the company more than any other. But you can see the breadcrumbs of future products by looking through the list of acquisitions. Typically it takes at least 18 months for an acquisition to show up in a product.
unique link to this extract
What is Nipah virus? Key things to know about the disease amid cases in India • The Guardian
Hannah Ellis-Petersen:
»
Airports across Asia have been put on high alert after India confirmed two cases of the deadly Nipah virus in the state of West Bengal over the past month.
Thailand, Nepal and Vietnam are among the countries screening airport arrivals over fears of an wider outbreak of the virus, which can spread from animals to humans and has a high fatality rate.
The Indian health ministry has confirmed two cases in the state of West Bengal since December but said there had been a “timely containment” of the virus.
The government did not give details on the infected patients but said almost 200 close contacts had been tested and no further outbreaks had been detected.
Nipah virus is primarily transmitted to humans from animals such as pigs and fruit bats, either by direct contact or through their secretions.
It can incubate in the body for a period of four to 14 days. The initial symptoms of the virus are often high fever, nausea, vomiting and respiratory problems, which can then develop into pneumonia. In severe cases it causes a dangerous swelling of the brain that can lead to neurological symptoms such as drowsiness and seizures.
While human to human transmission is low, it is seen by the World Health Organization as a high risk for epidemics because there is no vaccine. It has a high fatality rate of 40% to 75%, far deadlier than Covid-19.
«
Yikes. Transmissible, highly deadly, no vaccine? Definitely putting a watching brief on this one. Those damn fruit bats, eh.
unique link to this extract
Tesla scraps models in pivot to AI as annual revenue falls for first time • Financial Times
Kana Inagaki and Stephen Morris:
»
Tesla plans to scrap two models and invest $2bn in Elon Musk’s xAI, as the electric vehicle pioneer accelerates a charge into robotics and artificial intelligence following its first drop in annual revenue.
In the clearest sign yet of where Musk is steering Tesla, the company said it would end production of the premium S and X models next quarter and convert its California factory into a manufacturing hub for its Optimus robots.
The announcements came as Tesla’s fourth-quarter results laid bare the damage to the carmaker from a year dominated by the Trump administration’s ending of EV tax incentives, Musk’s polarising politics and the continued rise of low-cost Chinese rival BYD.
Fourth-quarter revenues fell 3% to $24.9bn, Tesla said after US markets closed on Wednesday. That pushed revenues for the year down 3% to $94.8bn, in a blow to a carmaker that over the past decade became a symbol for the arrival of the EV era.
Announcing the decision to ditch the premium S and X models, Musk said: “That is slightly sad, but . . . it’s part of our overall shift to an autonomous future.”
The world’s richest man has gambled Tesla’s future on self-driving Cybercabs and AI-enabled humanoid robots. The group has begun calling itself a “physical AI company”.
«
Tesla sells the Model S, 3, X, Y and Cybertruck. Its financials say that the 3 and Y make up 97% of deliveries, so this is less shocking than it might appear. The Cybertruck is a probably rounding error on the 3%, so doing something else with it – as Musk said he would – makes sense.
The robots and self-driving stuff is all fluff, though. Musk has literally been promising it for a decade, and still not got near it.
unique link to this extract
What is Nick Shirley? • The Verge
Mia Sato:
»
The violent federal occupation of Minneapolis — and the subsequent killings of two residents at the hands of immigration agents — began with a vlog. Nick Shirley, a roving 23-year-old with a smartphone and a taste for outrage, made a YouTube video with unfounded allegations of fraud at daycares operated by the local Somali American community. Like so much partisan media in history, he was trying to rile up the right-wing base. But he was also playing to another audience: the algorithm.
When I wrote about Shirley in early January, I described him as an influencer — a catch-all term that could be applied to a wide range of people, anyone from Joe Rogan to a 20-something woman sharing Shein hauls on TikTok. Shirley exhibits many shared behaviors: He has a following with parasocial tendencies. His style and sensibilities are finely tuned to what will play well online. He hawks merch at every turn. His literal influence reaches into the highest offices of the US government (Vice President JD Vance has sung his praises). But Shirley and his ilk are not just content creators with a right-wing twist — they’re algo hounds. And he was not just making plain old propaganda; he was making internet slop.
We mostly talk about it in the context of AI-generated material, but slop does not need to be synthetic — AI slop is just a subgenre of a larger type of content that is made quickly and cheaply and poorly. The same lukewarm financial advice peddled by thousands of literal talking heads on Instagram Reels is slop. Falsehoods and oversimplifications about breaking news or contentious celebrity drama that snowball to millions of views is slop. Engagement bait is slop. The president’s social media posts are slop. The main function of slop is to take something from you: your time, your attention, your trust. It is passive in that it requires nothing from viewers but to sit back and consume it. Slop is boring, repetitive, and often inexpensive to make — the natural evolution of an internet built for scale and ruthless optimization.
«
Sato slightly misses the point that Shirley’s accusations against daycares were direct repeats of investigations that the New York Times did in 2022, and which had been formally investigated by the Minnesota Attorney General. There’s a Wikipedia page, “2020s Minnesota fraud scandals” which points to those 2022 investigations.
In that sense, Shirley was pushing on an open door once Trump got in. The algorithms were ready and waiting to amplify him.
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
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.
The Overspill: when there's more that I want to say 