Farmed salmon suffer from sea lice – but fortunately there’s a laser-armed submarine AI robot to zap them off. CC-licensed photo by Sam Beebe 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. Fishy. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.
Nvidia stock plunges after Jensen Huang resigns as CEO • The New York Times
Aruni Soni, Eshe Nelson and Jason Karaian:
»
Nvidia’s shares plummeted Thursday, sending tremors through global markets, as the company’s chief executive, Jensen Huang, stepped down from his role.
Shares in Nvidia dropped as much as 14% when markets opened for trading on Thursday, extending a decline that has wiped out more than $1 trillion from the company’s market value this month alone. The stock began to reverse most of its losses by late morning.
The broader market rose slightly, too, as investors began to conclude that the fallout of Mr. Jensen’s departure would not fundamentally undermine the artificial intelligence trade. Yields on the 10-year US Treasury rose above 5% on Thursday for the first time in weeks. Yields go up when Treasury prices go down. [Translation: people sold off US Treasury bonds.]
Nvidia has emerged as central to the AI economy because its chips — or graphic processing units — help train and power models across the industry. It has also become a central bank of sorts for several AI companies, because its finances are deeply intertwined with those of everyone from frontier A.I. labs to neoclouds, which rent out computing power.
“This is an Nvidia problem before it is an AI problem,” said Mark Malek, chief investment officer at Siebert Financial. But he added that Nvidia’s woes could trickle through to other companies. “Where I would be careful is with the smaller AI cloud companies that leaned on Nvidia’s checkbook,” he said. “If the new regime gets more conservative about financing its own customers, that is where it shows up first.”
…Among the hardest hit stocks on Thursday were cloud computing providers that have strategic partnerships with Nvidia. Coreweave sank nearly 8%, and Nebius plunged nearly 17%. Nvidia’s rivals, on the other hand, ticked higher. AMD rose nearly 3% and Broadcom gained more than 4% on Thursday.
«
Once it turned out that Jensen Huang had not one, not two, but three affairs his goose was truly cooked. Imagine though how amazingly effective Nvidia could be if its chief executive isn’t having to buy three sets of flowers and scented candles every day.
unique link to this extract
Jensen Huang is out. His successor has the toughest job in tech • The Economist
»
Mr Huang was one of Donald Trump’s favourite businessmen, and used his access to lobby hard for looser curbs on selling chips to China, often clashing with hawks in Congress. The scandal has left Nvidia’s standing in the capital badly bruised. On November 18th, a day after the earnings call, Mr Huang has been asked to appear at a closed-door congressional briefing to answer questions about his relationship with Alice Huang, an executive at Megaspeed, a Singaporean customer of Nvidia, whom American officials reportedly suspect of being a Chinese agent. Democrats, who take control of Congress in January, may add more pressure. One employee concedes that Mr Huang’s combative style in Washington made Nvidia enemies, and that it will need to work with policymakers rather than against them. [Chief financial officer Colette] Kress will need to change course.
The hardest task of all may be keeping Nvidia at the centre of the ai economy. Mr Huang’s aggressive dealmaking has made the company the ai-infrastructure boom’s most important player, and boosted demand for its chips. Over the past three years Nvidia has pledged more than $70bn to startups, making about 90 investments last year alone and at least 60 so far this year. It has also backstopped ai clouds to the tune of up to $36bn, promising to buy computing capacity they cannot sell to others, and guaranteed the debts for a giant data-centre project in Ohio intended for Openai. On top of that, it has signed $20bn of data-centre leases that it plans to pass on to third parties, and orchestrated a financing programme with Wall Street lenders that could direct up to $500bn of further investment into data centres. Many of these deals rested on Mr Huang’s word. Ms Kress must now convince Nvidia’s partners that they still stand.
However deftly she handles all this, Ms Kress is unlikely to keep the job for long. She lacks a technical background; Bloomberg reports that the board has been working with an outside recruiting firm to find a permanent replacement. Among insiders, Ian Buck, who runs Nvidia’s data-centre computing business, is popular with engineers but untested at running a giant company. Lisa Su of amd and Rene Haas of Arm, a former Nvidia executive, may be options. Another could be Johny Srouji, Apple’s hardware-chip chief. Nvidia may yet find that the answer to its succession problem lies at the company that showed it how succession should be done.
«
Johnny Srouji would be an interesting – and logical – pick. He was given the salve of being promoted to “Chief Hardware Officer” at Apple when John Ternus got the top job (there are no other Chief Anything Officers at Apple). He has helped make Apple’s CPUs the envy of the business. But the culture at the two companies couldn’t be more different. That’s the dilemma he will face when the headhunters surely call him.
(Gift article.)
unique link to this extract
Trump backed Nvidia chip controls aimed at China three times as Lutnick stalled • Axios
Marc Caputo and Maria Curi:
»
On Thursday [last week], as the spy scandal rocked the administration, [Commerce Secretary Howard] Lutnick met in the Situation Room with Treasury Secretary Scott Bessent, Defense Secretary Pete Hegseth, U.S. Trade Representative Jamieson Greer and National Security Adviser and Secretary of State Marco Rubio.
During the meeting, Bessent confronted Lutnick about why his department has resisted implementing the export controls, particularly on Megaspeed, the company at the center of the spy scandal. The group sided with Bessent.
Lutnick appeared to agree, but then muttered under his breath, “I need to check with the president.”Lutnick’s seemingly innocuous comment raised concerns, the sources said, because officials have repeatedly believed the Commerce Department would impose the export controls, only to have nothing happen.
“He’s stalling it out,” said one official, describing how Lutnick had previously met privately with the president to change his mind about the export limit.
“This is an ongoing concern with Howard and this issue,” said another. This time, one person in the room called his bluff by saying: “Great. Why don’t we do it today?”
Around 5pm, the group headed to the Oval Office. Trump called tech advisor Elon Musk and agreed to move forward. Musk supported Rubio, Bessent, Greer and Hegseth’s case to impose the controls while Lutnick warned it “will start a trade war with China.”
“Everyone argued back that China is already doing to us what Lutnick warned would happen,” a source said. With Lutnick the lone holdout, Trump made his decision: “Let’s get it done.” But once everyone was outside of the Oval Office, Lutnick proposed a new delay tactic.
“I’ll develop some options and come back,” he said.
“Everyone was like: ‘No, no, the decision’s been made. Go fucking execute,'” one source said. “Well, it still hasn’t happened.”
«
Lutnick’s position is really strange, given that Trump has announced crackdowns on AI chip exports to China twice already this year. The expectation is that he will follow Jensen Huang to the unemployment line.
unique link to this extract
Words of warning about radiocarbon dating medieval human bones • npj Heritage Science
Kaare Lund Rasmussen et al:
»
Beginning in the thirteenth century and culminating approximately one hundred years later, a significant number of religious houses—circa 140—were established in Denmark by the majority of the monastic orders originally founded in Italy, France, and Germany. The general layout of these architectural complexes adhered strictly to the regulations of each respective order, typically comprising a church and three cloister buildings arranged around a central courtyard. Each religious house was accompanied by one or more cemeteries; in certain instances, the monastic residents possessed a burial ground of their own, whereas novices, laypeople, and townsfolk were interred elsewhere. A case in point is the Cistercian monastery at Øm in Jutland, where three distinct cemeteries have been identified. The most prestigious burial sites – highly sought after by the residents of the monastery and, consequently, also by affluent townsfolk willing to pay for a burial plot – were frequently located within the church, inside or adjacent to the chapter house, and along the cloister walk. For the current investigation, 27 burials have been selected from these precise locations, each demonstrating a stratigraphical or orientational relationship to the architecture. This relationship encompasses parallelism to the buildings or to other interments.
In Christian medieval Europe, meat was restricted during the Lent period. However, the meat of freshwater fish was tolerated as a food source because freshwater fish were considered “cold” or “neutral” in comparison with meat from warm-blooded animals and marine fish. Freshwater fish was even in some Christian circles and cultures considered a prestigious food item.
Interpretation of radiocarbon dates of organisms living in aquatic environments is not straightforward, and in the context of this Comment, this also applies to bones of humans that have consumed a significant amount of aquatic food such as fish. This Commentary is aimed at the practical application of radiocarbon dating medieval human bones, not only in Denmark but also in the rest of Europe.
…The depletion of carbon-14 in aquatic reservoirs also applies to organisms living therein, such as fish, shells, and marine mammals. From the formula describing the radioactive decay, one can calculate that a depletion of 5% in carbon-14 concentration corresponds to 400 years; for 15%, this is 1300 years. Hence, marine organisms that are contemporaneous with terrestrial organisms show carbon-14 dates that are approximately 400 years older; for rivers, e.g., the river Rhine, it can be up to 1300 years older. This effect also applies to terrestrial organisms consuming aquatic food sources. For example, bones of polar bears and prehistoric Inuit people with a 100% marine diet show carbon-14 dates which require a correction of 400 years.
«
How strange that eating fish (and other aquatic sources; it applies to shellfish too) upsets our understanding of the age of bones.
unique link to this extract
We know 15,579 websites are infected. How do we tell them? • Scott Helme
Scott Helme:
»
We’ve identified 15,579 websites that are currently infected with malicious JavaScript and between them, we conservatively estimate that they receive more than 135 million visits every month.
We know they’re infected, and we have the evidence to back that up, but there’s a problem: how on Earth do we tell them all?
As our Threat Intelligence capabilities at Report URI have grown, identifying compromised websites has become increasingly efficient. Unfortunately, responsible disclosure hasn’t kept pace, and we’re now finding infections considerably faster than we could ever hope to notify the affected organisations.
I genuinely don’t know how to solve this problem, so I’m asking for help.
At the core of it, we’re finding websites that are running JavaScript that they should absolutely not be running! For some examples of how nasty this can be:
• Credit card skimmers – often known as ‘Magecart’, they target payment pages and skim full card details to send them to the attacker
• ClickFix attacks – designed to trick users into completing a “prove you’re human” challenge and instead installing malware on their device
• Device exploitation – loading malware to compromise specific versions of iOS and install spyware, like DarkSword or Coruna
• Redirects – these can send the visitor anywhere, including fake login pages to phish details, device security ‘warnings’, scams, and much more.All of these are bad scenarios for the innocent visitor, and attackers change their payloads depending on their objective.
«
Later on Helme says that the company isn’t publishing the list of infected websites because 1) name and shame isn’t helpful 2) the sites would be put onto blocklists, and getting off them is really hard and, for a small business, potentially existential 3) the infections may have come through existing vulnerabilities in WordPress, which then means others could be hit by the same weakness.
The third seems weak (the criminals are scanning for these already), and the first ignores the damage to innocent visitors. But the second has merit.
unique link to this extract
HomeView review: a late arrival, still showing some seams • Six Colors
Jason Snell:
»
Some faces feel very much like they’re from the Apple Watch—simple clocks with the occasional widget that’s a single line of text very much in the style of a watch complication or a simple iPhone lock screen widget. There’s also a pretty animated clock and a lighthearted clock face featuring Snoopy animations. My favorite is undoubtedly the Photos face, which lets you choose a set of items, albums, or collections from your iCloud Photos library, all of which will be shuffled through, very much like an iPhone lock screen or the Apple Watch’s widget face. It is delightful to walk into my kitchen and see a picture of my family.
Other HomeView faces are much more complex, offering dashboards to control Apple Home items, a live view of a HomeKit-compatible security camera, or even a custom Modular face you can populate with widgets from the device or from your iPhone.
In many ways, these faces are the core of the product. They are the images you want to look at, yes, but they’re also the dashboards you can use to quickly interact with items in your home. And this is where HomeView’s software comes up a bit short.
The layouts aren’t flexible enough, for starters. Simple faces like photos work a charm, but more complex faces are kind of a mess. I’d like more options about what items to place on a face, and where to place them. The HomeView seems like the perfect device to use as a control panel for my smart-home devices, but the Home Modular face is strangely limited. For example, I can only use it to turn dimmable lights on to full brightness. To dim a light, I have to open the Home app, select the item in there, and then swipe up or down on the dimming interface.
It’s great that you can view iPhone widgets on the HomeView just as you can on the Mac, but it’s a bit frustrating that it will only display the small square widget style. Presumably because of the Apple Watch roots of some of these faces, there’s a surprising lack of digital clocks. In fact, in some face views you can create a face featuring a clock, but if you delete it, you can’t add it back. I admit to being frequently baffled about what I was able to add and customize, and what I wasn’t, when trying to build faces.
All of this is really unsurprising when you consider that this is an interface welded together from disparate parts of other Apple products, but it feels like it needs some more tweaking and attention for it to fully come together.
«
He’s really quite lukewarm on it; to work, an app needs CarPlay functionality, but recipe apps don’t have that, yet they’re what you’d want in a kitchen (this thing’s natural habitat). Sure, the apps can be updated, but that isn’t his only criticism. And in the UK it costs £300. Yeow.
unique link to this extract
3,500 laser-armed AI submarine robots patrol Norwegian salmon farms, killing sea lice • Tom’s Hardware
Mark Tyson:
»
Submarine AI laser robots have been a surefire success for Norwegian salmon farmers, according to the most recent annual report from the aptly named Stingray Marine Solutions. The Oslo, Norway-based firm now has over 3,500 units deployed around salmon pens along the Norwegian coast, protecting over 112 million fish. Stingray deployments are up by about 1,000 over the last year. This operation has grown to become one of the largest commercial AI deployments in Scandinavia.
These high-tech AI-enhanced submarines protect 20-25% of farmed salmon in the country, according to the latest figures. Using this high-tech AI laser solution to ‘optically delouse’ the salmon means fish don’t have to be passed through harsh chemical baths to prevent infestations of these tiny parasitic crustaceans. The result is improved salmon growth and welfare, and consistently cleaner fish exhibit reduced mortality.
…Sea lice are a big problem for fish farmers, with the control of these parasites being a central challenge for the industry. Traditionally, measures such as chemical treatments were used to help control the buildup of sea lice. However, while chemicals can treat a large number of fish simultaneously, they can be bad for the environment, can stress fish, lice can develop resistance, and the introduction of any kind of poison or toxin into food production isn’t usually good for humans either.
In contrast, the Stingray AI laser system identifies and targets only sea lice, works continuously 24/7 as the fish go around normally in their pens, and avoids any chemical introduction or resistance issues. The reduction in sea lice around the farming area is better for other wild fish populations nearby, too.
«
“Submarine AI laser robots” really does sound like a randomly generated collection of words for a sci-fi film, but this is real: there are videos.
unique link to this extract
Watched Earth: Something watched you this week
“Armaan”:
»
Pick a spot on Earth. We work out which satellites flew over it in the last week, then go find the photos they took. It all runs in your browser, on public data.
«
The highest resolution is 10m per pixel, so even a car is sub-pixel. But it’s definitely a fascinating resource (though as they note, no spy or military satellites).
unique link to this extract
Pinot Noir: inside the criminal gangs making fake wine • The Economist
William Ralston:
»
Wine has been faked “for as long as it has been drunk”, said Rebecca Gibb, the author of a history of the subject. But wine fraud intensified in the late 19th and early 20th centuries, when improvements in bottling, corking and transportation allowed wine to become more widely distributed. Fraudsters began passing off cheap or inferior bottles as products from prestigious regions. Wine-producing countries responded by introducing appellation systems, which restrict geographical names to wines made from specific grapes grown in defined areas according to established production standards.
These laws made it easier to distinguish legitimate wines, and gave authorities a stronger basis for prosecuting anyone selling fraudulent ones. But instead of disappearing, wine fraud moved upmarket. After the second world war prices of rare bottles rose, drawing in rich collectors. “A different market was now being deceived, because fine wine was being traded like art,” Gibb said. And like art, these wine collections were painstakingly built over time: bottles were sought out and acquired, not for immediate consumption but as financial investments, or left to mature until a special occasion.
More than a third of his career, Henri noted, had been spent going after the same man. “People keep saying I am always against this guy but…it’s not my fault”Fraudsters began sourcing high-value bottles from bars, restaurants and tasting events to refill with cheaper wine and recork—a scam called refilling “empties”. In the 2000s Rudy Kurniawan, an Indonesian living in Los Angeles, became one of its most successful practitioners, marketing his blends of cheaper vintages as prized releases from the likes of Domaine de la Romanée-Conti.
Kurniawan’s scheme started to unravel in 2008, when he tried to sell dozens of bottles of a wine called Clos Saint-Denis at auction. The vintages dated as far back as 1945—yet Domaine Ponsot, the estate, hadn’t started producing Clos Saint-Denis until 1982. Laurent Ponsot, the winery’s owner, travelled to New York to halt the sale; he later provided evidence for the FBI, which was already investigating Kurniawan. When agents stormed Kurniawan’s home in 2012, they found old bottles soaking in the sink, piles of corks and thousands of fake labels. In 2014 a federal court sentenced him to ten years in prison, the first conviction in America for wine fraud.
«
(Gift article.)
unique link to this extract
Mapping Trump’s • Citation Needed News
Molly White has created an amazing mapping of all the relationships between the Trump family and all the crypto shenanigans that have gone on over the past few years. She’s traced the money too (both the imaginary and the real stuff).
To say it’s a web understates it – no spider would create something this mad.
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
