The economics of modern playgrounds are surprising – and can keep them shut. CC-licensed photo by Dan Gaken 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. Use them wisely. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.
Another undersea cable damaged in Baltic Sea: Sweden launches probe, seizes suspected ship • France 24
»
Latvia said it had dispatched a warship on Sunday after damage to a fibre optic cable to Sweden that may have been “due to external factors”.
The navy said it had identified a “suspect vessel”, the Michalis San, which was near the location of the incident along with two other ships.
The Michalis San was headed for Russia, according to several websites tracking naval traffic.
Nations around the Baltic Sea are scrambling to bolster their defences after the suspected sabotage of undersea cables in recent months.
After several telecom and power cables were severed, experts and politicians accused Russia of orchestrating a hybrid war against the West as the two sides square off over Ukraine.
NATO earlier this month announced it was launching a new monitoring mission in the Baltic Sea involving patrol ships and aircraft to deter any attempts to target undersea infrastructure in the region.
“We have a warship patrolling the Baltic Sea around the clock every day and night, allowing us to quickly dispatch it once we learnt about the damage,” Latvian navy commander Maris Polencs said at a briefing Sunday.
Prime Minister Evika Silina said: “We have notified the Swedish authorities and are working together with them to assess the damage and its reason.”
«
That monitoring mission by NATO needs to step up its game. Also, what sort of threat would put off a captain under instruction from, presumably, Russia? Undersea cables have become a new, significant vulnerability.
unique link to this extract
An asteroid got deleted because it was actually Elon Musk’s Tesla Roadster • Astronomy.com
Mark Zastrow:
»
On Jan. 2, the Minor Planet Center at the Harvard-Smithsonian Center for Astrophysics in Cambridge, Massachusetts, announced the discovery of an unusual asteroid, designated 2018 CN41. First identified and submitted by a citizen scientist, the object’s orbit was notable: It came less than 150,000 miles (240,000 km) from Earth, closer than the orbit of the Moon. That qualified it as a near-Earth object (NEO) — one worth monitoring for its potential to someday slam into Earth.
But less than 17 hours later, the Minor Planet Center (MPC) issued an editorial notice: It was deleting 2018 CN41 from its records because, it turned out, the object was not an asteroid.
It was a car.
To be precise, it was Elon Musk’s Tesla Roadster mounted to a Falcon Heavy upper stage, which boosted into orbit around the Sun on Feb. 6, 2018. The car — which had been owned and driven by Musk — was a test payload for the Falcon Heavy’s first flight. At the time, it received a great deal of notoriety as the first production car to be flung into space, complete with a suited-up mannequin in the driver’s seat named Starman.
«
You might think (well, I did): sure, but if a Roadster were to hit the Earth it would make quite a mess, wouldn’t it? But it turns out that the car is on a path to Mars using a Hohmanm transfer orbit. It’s not going to rain cars.
unique link to this extract
10,000 AI tracks uploaded daily to Deezer, platform reveals, as it files two patents for new AI detection tool • Music Business Worldwide
Daniel Tencer:
»
France-headquartered music streaming service Deezer has launched a new AI detection tool – after filing two patent applications for the technology in December.
On Friday (January 24), the company revealed that its new tech has already discovered that roughly 10,000 ‘fully AI-generated tracks’ are being delivered to its platform every day.
That amounts to about 10% of the daily content delivered to Deezer.
Deezer CEO Alexis Lanternier also said on Friday the company plans to “exclude” fully AI-generated tracks “from algorithmic and editorial recommendation.”
“Generative AI has the potential to positively impact music creation and consumption, but its use must be guided by responsibility and care in order to safeguard the rights and revenues of artists and songwriters,” Lanternier said.
The company says it set out last year to develop an AI detection tool that “surpass[ed] the ability of available tools.”
“Tools that are on the market today can be highly effective as long as they are trained on data sets from a specific generative AI model, but the detection rate drastically decreases as soon as the tool is subjected to a new model or new data,” explained Aurelien Herault, Chief Innovation Officer at Deezer
«
Odd really, because Deezer claimed it was already detecting AI-generated music back in June 2023. Maybe it feels better about it now it has patents?
unique link to this extract
The end of the playground • The Value of Nothing
Martin Robbins:
»
This is a tale of two playgrounds. One is closing soon while the other – brand new – has stood empty for nearly a year, ringed with steel fencing to stop people from using it. Their stories aren’t the most important thing you’ll read today, but they illustrate something much bigger: the collapse and retreat of local government, and the profound effect it will have on our public spaces.
«
There’s no easy way to précis this, so you’ll have to take my word for it that this is worth reading. With a footnote about surly British teenagers as “hospitality” staff that’s worth the price of entry itself. As is the one about how developers near Shuttleworth placated the local NIMBYs.
Also, from later in the piece:
»
The preamble to Central Bedfordshire Council’s 2025-26 budget reads like a panic attack in written form. It’s not just bad, it’s completely unsustainable with no prospect of improvement. School transport costs have increased by over 100% – from £9m to £20m – in just four years. In a single year, residential care costs for children have increased by £2,000 per child… per week. (I’ve sort of glossed over it here, but I want to come back to these costs in a future post as some of these increases seem frankly bonkers.) Adult social care costs have risen a staggering 35% in the same period. The average cost per adult has increased 13% in that time, which is bad enough, but the number of older people seeking support has sky-rocketed, increasing “beyond any reasonable forecast based on previous trend data.” A key driver is that people who “previously would have paid for their care are now finding their finances depleted and seeking council support.”
«
Nvidia stock may fall as DeepSeek’s ‘amazing’ AI model disrupts OpenAI • Forbes
Peter Cohan:
»
America’s policy of restricting Chinese access to Nvidia’s most advanced AI chips has unintentionally helped a Chinese AI developer leapfrog U.S. rivals who have full access to the company’s latest chips.
This proves a basic reason why startups are often more successful than large companies: Scarcity spawns innovation.
A case in point is the Chinese AI model DeepSeek R1 — a complex problem-solving model competing with OpenAI’s o1 — which “zoomed to the global top 10 in performance”— yet was built far more rapidly, with fewer, less powerful AI chips, at a much lower cost, according to the Wall Street Journal.
The success of R1 should benefit enterprises. That’s because companies see no reason to pay more for an effective AI model when a cheaper one is available — and is likely to improve more rapidly.
“OpenAI’s model is the best in performance, but we also don’t want to pay for capacities we don’t need,” Anthony Poo, co-founder of a Silicon Valley-based startup using generative AI to predict financial returns, told the Journal.
Last September, Poo’s company shifted from Anthropic’s Claude to DeepSeek after tests showed DeepSeek “performed similarly for around one-fourth of the cost,” noted the Journal.
When my book, Brain Rush, was published last summer I was concerned that the future of generative AI in the U.S. was too dependent on the largest technology companies. I contrasted this with the creativity of U.S. startups during the dot-com boom — which spawned 2,888 initial public offerings (compared to zero IPOs for U.S. generative AI startups).
DeepSeek’s success could encourage new rivals to U.S.-based large language model developers. If these startups build powerful AI models with fewer chips and get improvements to market faster, Nvidia revenue could grow more slowly as LLM developers replicate DeepSeek’s strategy of using fewer, less advanced AI chips.
«
At some point the spending insanity has to end. DeepSeek may be exactly the shock that the arms race needs to end.
unique link to this extract
The Microsoft 365 Copilot launch was a total disaster • ZDNET
Ed Bott:
»
You’d think that Microsoft’s marketing team would have learned something after last year’s shambolic rollout of the Recall feature. Maybe, before trying another rollout, they might talk to a few customers, do some focus groups, even ask a few members of the press and analyst community for their advice.
But no.
Shortly after the New Year, someone in Redmond pushed a button that raised the price of its popular (84 million paid subscribers worldwide!) Microsoft 365 product. You know, the one that used to be called Microsoft Office? Yeah, well, now the app is called Microsoft 365 Copilot, and you’re going to be paying at least 30% more for that subscription starting with your next bill.
As far as I can tell, the response from customers has been overwhelmingly negative. I monitor Microsoft-focused online forums obsessively, and I read hundreds of complaints without seeing a single compliment. Seriously, the reaction to this rollout was an Excel #DIV/0 error.
«
A 30% rise for the AI element? It’s not that valuable.
unique link to this extract
How long do golden retrievers live? The answer could change our relationship with dogs • Slate
Isobel Whitcomb:
»
Inbreeding coefficients are commonly used by biologists to assess the health of an entire population of creatures. In human populations, an average inbreeding coefficient of 3% to 5% is considered unhealthy.
Studies suggest that in golden retrievers, that value, on average, hovers around 8%—not great. When Boyko and an international team of researchers analyzed the effects of inbreeding on longevity in golden retrievers, they found that dogs whose parents shared identical copies of the same genes lived shorter lives, on average, than those whose parents’ genes included less overlap.
The genetic mutations that erode dog lifespans can pop up seemingly out of nowhere, then spread rapidly through a population, like a spark exploding into a wildfire. Bernese mountain dogs, for instance, are plagued by a form of blood cancer called histiocytosis, said Ruple, the canine epidemiologist. In both humans and dogs, this cancer is associated with a mutation on one particular gene. While this cancer is incredibly rare in humans, 1 in 7 of these dogs dies of it. That wasn’t always the case: These gentle giants have existed for thousands of years, but it wasn’t until the 1970s that the first case of histiocytosis was described in a Bernese mountain dog. According to Ruple, it’s likely that the mutation happened in just one dog, was passed down to all of its puppies, then began causing cancer once those dogs were bred to one another.
In dogs, that can happen quickly. The average breeding-purebred male dog, called a sire, will father more than 100 puppies. That number can be much higher for particularly prolific sires—for instance, a male dog that wins a show. The tendency of one sire to spread a harmful mutation among its descendants even has a name, “the popular sire effect.” As it turns out, golden retrievers have the highest proportion of popular sires of any dog breed.
«
Genetic screening for dogs? Canine GATTACA? A 12-year US study into retrievers published in 2024 has found definite genetic causes for shorter lifespans.
unique link to this extract
Los Angeles shows cities can burn spectacularly. Vancouver is not exempt • Vancouver Sun
Dana Gee interviews John Valliant, an expert on fires:
»
JV: What L.A. is showing us is that major cities can burn spectacularly, and Vancouver is not exempt from that…we were capable of generating a heat dome. So, imagine if it had been 37ºC with 40 knot winds. The whole west side would have gone. All the Pacific Spirit Park would have gone, no problem. And that’s what L.A. shows us, is really how vulnerable we are, and it’s a very stern invitation to reevaluate the flammability of our neighbourhoods and how we engage with fire and how we measure fire risk, because it’s not 1990 anymore. We don’t live in a rainforest anymore. We’re living in something new, and it’s not as wet as it used to be.
DG: What can we do to help mitigate the risk?
JV: Get a heat pump. Get rid of your gas car…probably stop eating beef or eat less beef. A really great thing about Canada and about British Columbia (BC) is we have the FireSmart B.C. Program, which is run by the fire service. And you can have firefighters experienced, you know, flammability, people, experts on fire, come to your community, come to your neighbourhood, your cul-de-sac, to your backyard, and help you look at your garden, your back porch, your wood pile, through the lens of fire…I think we’re at a disadvantage in Vancouver because the houses are old. They’re wooden, and they’re built very close together, and that is a recipe for conflagration, as we saw in Pacific Palisades.And so it’s not like we’re gonna tear out every second house and create a space, but if you’re going to re-roof, re-roof with tin, don’t use vinyl siding. Think about having a sprinkler system, like a garden sprinkler, that you can mount on your roof to create a water curtain over your house…when embers start flying. You can reduce the flammability of your home and neighbourhood by double digit percentages…No one, as we saw in Los Angeles and as we saw in Fort McMurray, no one is stopping a firestorm. But firestorms are rare, but fires are going to be increasingly common…
«
Just in case you were wondering where the next human-aided climate fire calamity might happen.
unique link to this extract
How I use ChatGPT • The Ruffian
Ian Leslie:
»
My headline is slightly misleading. I use ChatGPT and Claude (the two leading LLMs, from OpenAI and Anthropic respectively) But I use Claude much more; it’s the only one I pay for.1 I cite ChatGPT above simply because it’s much better known. This is interesting in itself. I don’t think even those who prefer ChatGPT would argue it’s significantly better than Claude, and it has a clearly inferior brand name.
But OpenAI’s headstart on Anthropic, and everyone else, allied with their CEO’s talent for PR, has given their product a big advantage in brand awareness and market share, at least for now (it seems to be levelling out somewhat). ChatGPT has become the category generic: people say “ChatGPT” to mean “an LLM chatbot”, like they say “Jacuzzi” to mean “whirlpool bath”.
[A selection of the approximate prompts:]
• I’ve just bought a Logitech webcam and light and I’m having trouble setting them up on my Mac Air. [Description of problem]. Can you tell me what I’m doing wrong?
• How would you characterise the different movements of Mahler’s Third Symphony? Please use nontechnical language. I’m particularly interested in the emotional effects that Mahler conveys.
• Here’s a piece that I’ve written for my Substack. Can you check it for a) Spelling errors b) Grammatical errors c) Any problems with flow and clarity?
• Can I get to Belfast from London City Airport? What times are the flights?
• I live in London. I’m looking to buy a secondhand car. I’m looking for [list of things I want from a car]. Can you suggest which models I should look at, give me a rough idea of price, and how the payments might work out on different purchase schemes?
• I’m making an oxtail stew according to this online recipe [cut and paste]. I’ve just bought some Jerusalem artichokes, is it OK to throw them in? If so when should I add them?«
There are plenty more. I’m intrigued and puzzled by them: many seem like questions that could be answered by good search technique. Plus I’d constantly wonder if it was hallucinating. But: asking these questions may well be an efficient short cut to the right answer(s). So, a very efficient method after all. (And less digging through forums, says Leslie.)
One of the most fun questions, for a cryptic crossword: “that confounded pane in the neck (4)”. Neat. Though he doesn’t say if the LLM answered correctly.
unique link to this extract
Bluesky 2024 moderation report • Bluesky
Aaron Rodericks:
»
In late August, there was a large increase in user growth for Bluesky from Brazil, and we saw spikes of up to 50k reports per day. Prior to this, our moderation team handled most reports within 40 minutes. For the first time in 2024, we now had a backlog in moderation reports. To address this, we increased the size of our Portuguese-language moderation team, added constant moderation sweeps and automated tooling for high-risk areas such as child safety, and hired moderators through an external contracting vendor for the first time.
We already had automated spam detection in place, and after this wave of growth in Brazil, we began investing in automating more categories of reports so that our moderation team would be able to review suspicious or problematic content rapidly. In December, we were able to review our first wave of automated reports for content categories like impersonation. This dropped processing time for high-certainty accounts to within seconds of receiving a report, though it also caused some false positives. We’re now exploring the expansion of this tooling to other policy areas. Even while instituting automation tooling to reduce our response time, human moderators are still kept in the loop — all appeals and false positives are reviewed by human moderators.
Some more statistics: the proportion of users submitting reports held fairly stable from 2023 to 2024. In 2023, 5.6% of our active users [those who haven’t been suspended or deleted] created one or more reports. In 2024, 1.19M users made one or more reports, approximately 4.57% of our user base.
In 2023, 3.4% of our active users received one or more reports. In 2024, the number of users who received a report were 770K, comprising 2.97% of our user base.
«
Bluesky grew almost tenfold in 2024, from 2.89m to 25.9m users. So they grew the moderation staff to 100 people – and are still growing. In the year, they received 6.48m reports – up 17x from 2023. Odd how reports aren’t linear with growth.
As a view into a burgeoning social network, this is fascinating. We never got insight like this into Twitter’s early days.
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 