Ships stranded in the strait of Hormuz are succumbing to barnacles and jellyfish. CC-licensed photo by Curtis Gregory Perry 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. Scraping by. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.
Waymo pauses Atlanta service as its robotaxis keep driving into floods • TechCrunch
Sean O’Kane:
»
Waymo has now paused service in two cities because its robotaxis are struggling to deal with heavy rain and flooded roads, a problem that already prompted the company to issue a recall last week.
One of Waymo’s robotaxis was spotted driving through a flooded street in Atlanta, Georgia, on Wednesday before it ultimately got stuck for about an hour, according to local news reports. The vehicle was recovered and removed from the scene, Waymo told TechCrunch. Waymo says it paused service in the city, just like it has in San Antonio, Texas, while it figures out a solution.
“Safety is Waymo’s top priority, both for our riders and everyone we share the road with. During a period of intense rain yesterday in Atlanta, an unoccupied Waymo vehicle encountered a flooded road and stopped,” the company said in a statement.
Waymo admitted that it hadn’t finished developing a “final remedy” for avoiding flooded areas when it issued its software recall last week. Instead, the company said that it shipped an update to its fleet that placed “restrictions at times and in locations where there is an elevated risk of encountering a flooded, higher-speed roadway,” according to documents released by the National Highway Traffic Safety Administration (NHTSA).
But even those precautions apparently were not enough to stop the Waymo robotaxi from entering the flooded intersection in Atlanta. Waymo told TechCrunch on Thursday that the storm in Atlanta produced so much rainfall that flooding was happening before the National Weather Service had issued a flash flood warning, watch, or advisory. The company said those alerts are part of a larger set of signals it relies on to prepare the vehicles for poor weather.
«
Even humans struggle with how to drive well in floods, so that’s not entirely shocking. Even so, it demonstrates how big the gap is between ambition and reality for self-driving cars.
unique link to this extract
Incoming Ofcom chair vows to take on ‘tech bros’ • The Guardian
Michael Savage and Dan Milmo:
»
Ofcom’s incoming chair has vowed to take on the “tech bros”, as he conceded there was now a perception the regulator had been complacent and slow over concerns about online safety.
Ian Cheshire, the former Channel 4 chair who has secured the job overseeing the technology and media regulator, also told MPs he had personal concerns about the impact of social media on under-16s.
During a hearing before the science, innovation and technology select committee, Cheshire was asked directly about whether he would take on the powerful tech companies that dominate the online world.
“Yes,” he said, adding: “It is the area I want to probe and understand, because I think there is clearly a perception that it has been either complacent or slow or both.”
However, he suggested Ofcom needed to be clear about what it could and could not achieve in terms of policing tech platforms. He said he wanted the platforms themselves to come together and demonstrate they wanted to do more.
He said: “I think there are some questions about the practicality of what speed to do … because I think there are slightly more constraints.
“If expectations are up here and the delivery is here, I think Ofcom has to take it on the chin to work out how to communicate that and say: ‘What’s the maximum we could do?’”
While he said limiting social media for children was a matter for the government, he said: “I am personally – as a parent and grandparent – very nervous about social media under 16 personally, but I wouldn’t want to impose that as a political or an Ofcom view.”
«
OK, but how is he going to take on the tech bros without the support of the government? How will he do it without a proper budget to take them on in the courts, and at the internet level, and at the legislative level? Talk is cheap; proper regulation takes multi-agency effort, as the failure of the Environment Agency to tackle fly-tipping and water pollution (separately) shows.
unique link to this extract
Meta begins laying off 8,000 employees as it transforms around AI • WSJ
Meghan Bobrowsky and Raffaele Huang:
»
Meta Platforms began laying off thousands of employees Wednesday morning and reassigning thousands of others to AI-focused roles, according to an internal memo and people familiar with the matter.
Meta’s chief people officer, Janelle Gale, told staff last month that the coming layoffs would affect 10% of the company, or roughly 8,000 employees, and that the company would also cancel plans to hire for 6,000 open roles. In a follow-up memo on Monday, she said it would also move a separate 7,000 staffers into new AI-focused roles and transition a number of managers to individual contributor roles as part of the reorganization efforts.
Individual employees in Asia and Europe began receiving notifications that they would be affected early in their mornings, followed hours later by employees in the Americas, the people said.
In a memo to staff Wednesday morning, Meta chief executive Mark Zuckerberg thanked laid off employees for their contributions and called AI “the most consequential technology of our lifetimes.”
“This is the most dynamic I have seen our industry. I’m optimistic about everything we’re building,” he said. “But success is not a given.”
Meta is full steam ahead into a gargantuan effort to reimagine its workforce and become more nimble to compete with AI-native startups. The company has flattened teams and started tracking employees’ keystrokes and mouse clicks to help train its AI models on how to use computers.
«
Of course the stock went up, because why not. Replacing humans with machines and watching every keystroke of those who remain? What’s not to like, from the stock market’s view?
unique link to this extract
GLP-1 use has more than quadrupled since 2021 as obesity rates continue to show signs of decline • Epic Research
Epic Research:
»
Semaglutide (Wegovy, Ozempic) received FDA approval for chronic weight management in 2021, followed by tirzepatide (Zepbound) in 2023, both demonstrating substantial weight loss in clinical trials.2,3 Demand has grown rapidly, raising questions about real-world uptake, prescribing patterns across clinical indications, and whether population-level BMI distributions are beginning to shift in response.
The GLP-1 utilization tracker reports prescriptions per 100,000 patients for each GLP-1, and the BMI tracker reports the distribution of U.S. adults across BMI categories.
GLP-1 prescribing has grown substantially across the tracker’s window, with prescriptions per 100,000 U.S. adults rising from 1,884 in the second quarter of 2021 to 8,819 in the first quarter of 2026. The composition of GLP-1 prescribing changed markedly over the period.
In 2021, the most prescribed agents were semaglutide (about 680 per 100,000) and dulaglutide (about 790 per 100,000). By the first quarter of 2026, tirzepatide had become the most frequently prescribed agent at roughly 4,700 prescriptions per 100,000 patients, and semaglutide had grown to around 3,900 per 100,000. Use of older agents declined as the newer drugs displaced them.
BMI distribution has shifted modestly toward lower weight categories over the same period, with the change concentrated in the second half of the period. The share of U.S. adults classified as obese was 42.3% in the second quarter of 2021 and remained near that level through 2022 before declining to 40.7% by the first quarter of 2026. Over the same window, the healthy weight share rose from 25.1% to 25.6%.
«
There’s a graph showing usage taking off on a straight line. These drugs are going to be everywhere in a decade.
unique link to this extract
AI tool has “saved a lot of aircraft” in Epic Fury, AFSOC chief says • C4isrnet
Hope Hodge Seck:
»
The head of Air Force Special Operations Command revealed that an AI-powered intelligence collection and transfer system has been in use since the “first day” of Operation Epic Fury to help large attack drones and manned aircraft avoid Iranian threats.
Speaking at a hearing before the Senate Armed Services Committee Tuesday, Lt. Gen. Michael Conley said “necessity had been the mother of invention” in spurring the service to apply available machine learning tools to combat operations.
“On the first day, they realized that our MQ-9 [Reaper drones] and other aircraft were at risk in a very hostile environment, and they were able to take some smart people, use artificial intelligence tools, and put humans on the loop, instead of in the process the whole time, and move Top Secret national-level intel,” Conley said, referring to his “small team” of AI and tech specialists.
The process of moving intelligence would have taken human operators “20 to 30 minutes to get that to a crew, into a cockpit, or into a ground control station,” he said.
Instead, Conley said, nine crew members were able to use AI “bots” to convert applicable data to a Secret clearance level and make it available in the cockpits of aircraft — all within two to three seconds.
«
Good that the US is demonstrating the flexibility it needs. Still not sufficient, though: at least one F-35 was shot down. And the US is clearly worried about the potential of Iran’s drones.
unique link to this extract
Barnacles and jellyfish infest ships trapped in the Gulf • Financial Times
Alice Hancock:
»
Ships trapped in the Gulf are accumulating barnacles, algae and jellyfish as the Middle East conflict drags on and temperatures begin to rise, impeding their ability to eventually exit the region.
At least 800 merchant ships are still stranded in the Gulf following the outbreak of fighting on February 28 with about 20,000 seafarers left on board to perform routine maintenance duties.
But the Gulf’s shallow sandy seabed and warm waters have put ships at anchor or adrift there at risk of sand and sea creatures clogging up gratings that protect the vessel’s internal pipework. Seafarers are also struggling to get hold of critical parts when systems have broken down.
“What happens is that if you don’t move, if the vessels stay out of speed just basically drifting and these are warm waters, you have a lot of fouling growing and that is probably happening for the moment for our vessel,” said Lasse Kristoffersen, chief executive of Wallenius Wilhelmsen, the car carrier company that currently has one ship stuck in the Gulf.
Rolf Habben Jansen, chief executive of Hapag-Lloyd, told the shipping line’s podcast recently that the one vessel it managed to transit out of the strait of Hormuz, which has been in effect shut to most shipping traffic since the beginning of March, had to travel much slower because of the drag created by barnacles.
“The main surprise was the amount of fouling we had on the ship because the ship had been in water of about 30 degrees C for six to eight weeks so then you see there was a lot of stuff that gets attached to the ship that you really don’t want to have attached to the ship,” he said.
«
So many side effects. These will need some sort of drydock when this clears.
unique link to this extract
Context rot: the emerging challenge that could hold back LLM progress • Understanding AI
Timothy B. Lee:
»
relatively soon, we’re going to bump up against fundamental limitations of the attention mechanism underlying today’s leading LLMs.
With attention, an LLM effectively “thinks about” every token in its context window before generating a new token. That works fine when there are only a few thousand tokens in the context window. But it gets more and more unwieldy as the number of tokens grows into the hundreds of thousands, millions, and beyond.
An analogy to the human brain helps to illustrate the problem. As I sit here writing this article, I’m not thinking about what I ate for breakfast in 2019, the acrimonious breakup I had in 2002, or the many episodes of Star Trek I watched in the 1990s. If my brain were constantly thinking about these and thousands of other random topics, I’d be too distracted to write a coherent essay.
But LLMs do get distracted as more tokens are added to their context window — a phenomenon that has been dubbed “context rot.” Anthropic researchers explained it in a September blog post:
»
Context must be treated as a finite resource with diminishing marginal returns. Like humans, who have limited working memory capacity, LLMs have an “attention budget” that they draw on when parsing large volumes of context. Every new token introduced depletes this budget by some amount, increasing the need to carefully curate the tokens available to the LLM.
This attention scarcity stems from architectural constraints of LLMs. LLMs are based on the transformer architecture, which enables every token to attend to every other token across the entire context. As its context length increases, a model’s ability to capture these pairwise relationships gets stretched thin, creating a natural tension between context size and attention focus.
«
The blog post went on to discuss context engineering, a suite of emerging techniques for helping LLMs stay focused by removing extraneous tokens from their context windows.
Those techniques are fine as far as they go. But I suspect they can only mitigate the underlying problem. If we want LLMs to reason effectively over much longer contexts, we may have to fundamentally rethink how LLMs work.
«
In effect, this is scaling – the database thing – for LLMs. As token length grows, the computing demand grows exponentially rather than linearly. So you need something that will go linearly, at best.
unique link to this extract
2026 World Cup: empty rooms & Fifa cancellations mean US hotels fear washout • BBC Sport
Dale Johnson:
»
The World Cup was supposed to provide a tourism boom for the US, but now the fear is it may never materialise.
A report produced by the American Hotel & Lodging Association (AHLA) has found that bookings are well below expectations in almost every host city.
The AHLA said this does not align with Fifa’s statement that more than five million tickets have been sold and it creates a risk that “the anticipated economic lift may fall short”.
The AHLA is the largest hotel association in the US, representing more than 32,000 properties and over 80% of all franchised hotels.
Its report partially puts the blame at the door of Fifa, accusing world football’s governing body of block-booking far too many rooms for its own use and creating false demand.
This, the AHLA said, led to artificially high pricing which, after Fifa cancelled a large number of rooms, has been replaced by a vacuum of availability.
Fifa said it does not recognise this accusation.
Hotels said high match ticket pricing, local transport and tax costs, and the political backdrop have put visitors off.
For the hotels, this World Cup could fall flat.…Up to 70% of rooms reserved by Fifa in Boston, Dallas, Los Angeles, Philadelphia and Seattle have been cancelled, the AHLA said.
In a statement Fifa rejected the AHLA’s claims and said it had followed agreements made with hotel chains.
«
Fifa predicted it would create 185,000 jobs and add $17.2bn in GDP. Looks like that’s not going to happen, with just three weeks left to the first kickoff.
unique link to this extract
All those AI note takers? They’re making lawyers very nervous • The New York Times
Sarah Kessler:
»
Productivity powered by artificial intelligence is all the rage. Skipping meetings and sending an A.I. note taker instead has been called “the latest office power move.” Wallet-size recorders that use AI to log live interactions have become a product category. And at least one CEO has endorsed the idea of adding an AI board member. (Maybe one programmed to behave like Warren Buffett?)
But to lawyers like [Jeffrey] Gifford, inviting an A.I. bot to meetings introduces a ticking time bomb of legal risk.
AI-generated transcripts, which some video call apps allow users to turn on by default, preserve all sorts of things — offhand comments, quickly corrected statements, jokes — that humans would rarely write in the meeting minutes. And they show up in meetings that would otherwise not be recorded.
In a lawsuit or an investigation, that can make every word uttered discoverable. Even worse, say corporate lawyers: Sharing the meeting with an AI bot may void attorney-client privilege, making conversations that would not otherwise be subject to discovery fair game in a lawsuit.
The New York City Bar Association issued a formal opinion on AI note takers last year, urging lawyers to “consider whether recording, transcribing and summarizing is tactically well advised in the particular circumstances” and to advise clients using such tools “of the disadvantages of doing so.”
One concern is accuracy. An AI transcript could, for example, record “does matter” as “doesn’t matter.” If that sentence comes up in court years later, the mistake may be difficult to remember.
Corporate lawyers also worry about AI note takers’ lack of context and discretion. For example, recording every word of a board meeting, no matter how tangential the remark, could be legally perilous.
“You want to make sure that the minutes, if they get into a courtroom, are going to not only be accurate but also are going to have the emphasis that the board would like,” said Doug Raymond, a partner at Faegre Drinker Biddle & Reath.
«
That loss of priviliege is potentially colossal.
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 