Crunching down web links via shorteners is such bad business that even Google is abandoning it – along with 280 million links. CC-licensed photo by Forsaken Fotos 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. Shortly. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. Observations and links welcome.
CrowdStrike CEO has twice been at center of global tech failure • Business Insider
Lakshmi Varanasi:
»
The outage disrupted operations at major banks, airlines, retailers, and other industries after CrowdStrike, a cybersecurity giant used by Microsoft and others, pushed a faulty update.
CrowdStrike owned up to its mistake, issuing an apology and a workaround on Friday. But it has yet to detail just how a destructive update could have been released without being caught by testing and other safeguards. [It has since released a very uninformative blogpost – Overspill Ed.}
Naturally, blame has begun to target the man at the centre of it all: CrowdStrike CEO George Kurtz.
Tech industry analyst Anshel Sag pointed out that this isn’t the first time Kurtz has played a major role in a historic IT blowout.
On April 21, 2010, the antivirus company McAfee released an update to its software used by its corporate customers. The update deleted a key Windows file, causing millions of computers around the world to crash and repeatedly reboot. Much like the CrowdStrike mistake, the McAfee problem required a manual fix.
Kurtz was McAfee’s chief technology officer at the time. Months later, Intel acquired McAfee. And several months after that Kurtz left the company. He founded CrowdStrike in 2012 and has been its CEO ever since.
“For those who don’t remember, in 2010, McAfee had a colossal glitch with Windows XP that took down a good part of the internet,” Sag wrote on X. “The man who was McAfee’s CTO at that time is now the CEO of CrowdStrike.”
In response to a request for comment from Business Insider, CrowdStrike shared its latest blog posts detailing the problem and its recommended fix, but did not elaborate on how the update slipped through the company’s safety protocols.
«
Personally I didn’t remember the 2010 calamity. But Kurtz does have a habit of being in the wrong place at the wrong time. I’m reminded of Thomas Midgley Jr, inventor of CFCs and tetraethyl lead: both seriously bad for the world. Be interesting to see how quickly CrowdStrike’s stock drops: it fell 10% on Friday. Further to go, I think.
unique link to this extract
Windows 3.1 saves the day for Southwest Airlines during CrowdStrike outage • Tom’s Hardware
Jowi Morales:
»
Southwest Airlines, the fourth largest airline in the US, is seemingly unaffected by the problematic CrowdStrike update that caused millions of computers to BSoD (Blue Screen of Death) because it used Windows 3.1. The CrowdStrike issue disrupted operations globally after a faulty update caused newer computers to freeze and stop working, with many prominent institutions, including airports and almost all US airlines, including United, Delta, and American Airlines, needing to stop flights.
Windows 3.1, launched in 1992, is likely not getting any updates. So, when CrowdStrike pushed the faulty update to all its customers, Southwest wasn’t affected (because it didn’t receive an update to begin with).
The airlines affected by the CrowdStrike update had to ground their fleets because many of their background systems refused to operate. These systems could include pilot and fleet scheduling, maintenance records, ticketing, etc. Thankfully, the lousy update did not affect aircraft systems, ensuring that everything airborne remained safe and was always in control of their pilots.
…Nevertheless, Southwest passengers weren’t 100% unaffected by the CrowdStrike crash, as many airports also encountered system issues.
Aside from Windows 3.1, Southwest also uses Windows 95 for its staff scheduling system. It is a newer operating system — about three years younger than Windows 3.1 — but it’s ancient compared to today’s tech. Many of the airline’s staff have been complaining about it already.
However, the CrowdStrike global outage might discourage Southwest from upgrading its systems.
«
Absolutely astonishing that Southwest has been able to keep those utterly ancient systems going for so long. It must be quite difficult to find the PCs capable of running them now. As for “well, it avoided the Crowdstrike hit” – it’s the sort of thing that happens once every 30 years, and as noted above, hits the airport systems too, so nobody’s getting out.
unique link to this extract
We need to rewild the internet • Noema
Maria Farrell and Robin Berjon:
»
The story of German scientific forestry transmits a timeless truth: When we simplify complex systems, we destroy them, and the devastating consequences sometimes aren’t obvious until it’s too late.
That impulse to scour away the messiness that makes life resilient is what many conservation biologists call the “pathology of command and control.” Today, the same drive to centralize, control and extract has driven the internet to the same fate as the ravaged forests.
The internet’s 2010s, its boom years, may have been the first glorious harvest that exhausted a one-time bonanza of diversity. The complex web of human interactions that thrived on the internet’s initial technological diversity is now corralled into globe-spanning data-extraction engines making huge fortunes for a tiny few.
Our online spaces are not ecosystems, though tech firms love that word. They’re plantations; highly concentrated and controlled environments, closer kin to the industrial farming of the cattle feedlot or battery chicken farms that madden the creatures trapped within.
We all know this. We see it each time we reach for our phones. But what most people have missed is how this concentration reaches deep into the internet’s infrastructure — the pipes and protocols, cables and networks, search engines and browsers. These structures determine how we build and use the internet, now and in the future.
They’ve concentrated into a series of near-planetary duopolies. For example, as of April 2024, Google and Apple’s internet browsers have captured almost 85% of the world market share, Microsoft and Apple’s two desktop operating systems over 80%. Google runs 84% of global search and Microsoft 3%. Slightly more than half of all phones come from Apple and Samsung, while over 99% of mobile operating systems run on Google or Apple software. Two cloud computing providers, Amazon Web Services and Microsoft’s Azure make up over 50% of the global market. Apple and Google’s email clients manage nearly 90% of global email. Google and Cloudflare serve around 50% of global domain name system requests.
«
Written back in April, but particularly apt just now.
unique link to this extract
The Trump shooting was fake news • The Ruffian
Ian Leslie:
»
How do we distinguish between real news, by which I mean an important, game-changing event, and fake news, by which I mean an event that dominates headlines and social media feeds but isn’t actually significant?
This is a far stickier problem than what is commonly meant by “fake news”. For years now, alarms have been raised about the danger of AI-enabled content spreading lies and confusion among low-information voters. In the recent British election campaign, however, deep fakes were a dog that didn’t bark. There were no videos of Keir Starmer calling for the legalisation of heroin or Rishi Sunak sex tapes, and if there had been it’s doubtful they would have had an impact.
Perhaps the real test presented by our news environment isn’t distinguishing what’s real from what’s fictional, but what matters from what does not. The people who find that the hardest are high-information voters; people who follow the news closely. We act as if single new events are desperately important, when what really moves the tides is the accretion of news events over time; the underlying trends rather than the over-hyped data points.
Knowing which events are significant and which are not is hard in any context. A group of social scientists designed a machine learning model to study two million diplomatic cables sent by the US State Department between 1973 and 1979. Most of the cables were about trivial events, a few were about important ones – or rather, ones that seemed important at the time. The researchers compared this corpus to the fraction of the cables which were later deemed important by historians. Events that seemed important to the diplomats turned out not to be; events that seemed unimportant turned out to consequential. Even experts – perhaps especially experts – are highly unreliable judges of historical significance.
«
Worth pondering how bad we are at such judgements. But he does have a suggestion for something that is real news, quite recently. (Not that recent. But his accuracy is proven by what he picked.)
unique link to this extract
Coal-filled trains are likely sending people to the hospital • Ars Technica
Bárbara Pinho:
»
Although US coal consumption has fallen dramatically since 2005, the country still consumes millions of tons a year, and exports tons more—much of it transported by train. Now, new research shows that these trains can affect the health of people living near where they pass.
The study found that residents living near railroad tracks likely have higher premature mortality rates due to air pollutants released during the passage of uncovered coal trains. The analysis of the San Francisco Bay Area cities of Oakland, Richmond, and Berkeley shows that increases in air pollutants such as small particulate matter (PM 2.5) are also associated with increases in asthma-related episodes and hospital admissions.
“This has never been studied in the world. There’s been a couple studies trying to measure just the air pollution, usually in rural areas, but this was the first to both measure air pollution and trains in an urban setting,” said Bart Ostro, author of the study and an epidemiologist at the University of California, Davis.
Trains carry nearly 70% of coal shipments in the United States, leaving a trail of pollution in their wake. And coal exports will have a similar impact during transit. Ostro explained that when uncovered coal trains travel, the coal particles disperse around the railroad tracks. Levels of PM 2.5 “[spread] almost a mile away,” he added.
As a result, the mere passage of coal trains could affect the health of surrounding communities. Ostro was particularly concerned about how these pollutants could harm vulnerable populations living near the coal export terminal in Richmond. Previous census data had already shown that those in Richmond who live around the rail line have mortality rates 10% to 50% higher than the county average.
«
Meanwhile people worry about tiny levels of radioactivity in waste water from Fukushima.
unique link to this extract
Here’s everything you should know about Apple’s rumored HomePod with display – 9to5Mac
Michael Burkhardt:
»
According to supply chain analyst Ming-Chi Kuo, Apple’s smart display product is likely to include a 7-inch LCD display. This would make it similar to other products in the market, such as the Amazon Echo Show. The device would be good for FaceTime calls, controlling your smart home accessories, and interacting with Siri.
9to5Mac has discovered that Apple has been testing tvOS on the iPad mini in the past, suggesting that the company intends to use a similar screen size for the upcoming smart display product.
…According to a sighting by MacRumors last month, an unreleased device called “HomeAccessory17,1” was spotted in backend code. This reference suggests that the device will be running a variation of the A18 chip, which will debut with the upcoming iPhone 16 models.
If the upcoming HomePod with display sports an A18 chip, the device will likely support Apple Intelligence, which current HomePod models do not. This would be pretty big, since Siri takes a massive leap with Apple Intelligence, something that would be pretty important on a device you’d primarily interact with by speaking to it.
…Apple has long been rumored to be working on a new operating system called homeOS, with rumors going as far back as 2021. More recently, new references were found within tvOS 17.4 mentioning this “homeOS” branding. It’s not quite clear if homeOS is going to be an all-new operating system (although it’ll likely still be based on tvOS), or if it’ll just be a rebrand of tvOS. Either way, new HomePod products with screens will likely incorporate that branding in some way.
Earlier this month, 9to5Mac found a new touch interface called “PlasterBoard” within tvOS 18 beta 3. PlasterBoard is still in its early days, but so far we’ve seen a new passcode screen, something that doesn’t exist on tvOS today. It’d likely make sense to have a proper Lock Screen on a product like the rumored HomePod with Display, as that’s meant to be a more personal device.
«
Have to wonder very seriously who this is aimed at, and what gap in our lives it’s meant to fill.
unique link to this extract
AI is an impediment to learning web development • Ben Borgers
Ben Borgers was head of engineering for JumboCode, a club of 180 students at Tufts University building software:
»
LLMs excel at writing code for web development — you can describe a frontend component that you want and get a decent React component back.
I didn’t spend the year hawk-eyeing the teams’ repositories, but when I did poke my head in, I found substantial portions that looked LLM-written — that is to say, overly-commented, dissonant, and, at times, horrifying.
The starkest example I came across was a Next.js project that had:
• A page written in HTML and vanilla JavaScript, loaded from the public/ directory, completely outside of the Next.js + React system
• Vanilla JavaScript loaded in via filesystem APIs and executed via dangerouslySetInnerHTML
• API calls from one server-side API endpoint to another public API endpoint on localhost:3000 (instead of just importing a function and calling it directly)These don’t seem to me like classic beginner mistakes — these are fundamental misunderstandings of the tools and the web platform.
LLMs will obediently provide the solutions you ask for. If you’re missing fundamental understanding, you won’t be able to spot when your questions have gone off the rails.
LLMs are a shortcut to get assignments done. In the process, however, you learn close to nothing. It’s cliche, but struggling is learning. The way you learn is that you try different paths, piece bits of information together, and eventually create a mental model.
LLMs don’t require you to form a mental model and allow you to skip to the end result, but in turn you won’t have a mental model when you actually need one (for example, when you need to verify that your LLM has architected the code in a reasonable way).
LLMs are useful if you already have a good mental model and understanding of a subject. However, I believe that they are destructive when learning something from 0 to 1.
«
With Google killing goo.gl links, experts warn of widespread ‘link rot’ • PC Mag
Kate Irwin:
»
Goo.gl is going to the Google graveyard, and taking 280 million links with it.
The tech giant is hitting the kill switch on the Google URL Shortener, meaning all goo.gl links will start displaying a warning message next month and no longer work as of Aug. 25, 2025. Once a web destination in its own right, the goo.gl link now redirects to Google’s post detailing the shutdown.
Google is giving developers plenty of time to move their links to new domains, but Google’s warning message could lead to “disruptions” and stop goo.gl-connected pages from loading properly. It’s possible to disable this warning page, however, by adding the query parameter “si=1” to existing goo.gl links, Google says.
Goo.gl was initially created in response to the rise of Twitter, which had strict character limits until Elon Musk took over and expanded it to 10,000 last year.
“With character limits in tweets, status updates and other modes of short form publishing, a shorter URL leaves more room to say what’s on your mind — and that’s why people use them,” Google said in 2009 when it first launched goo.gl.
Since then, the web and social media landscape has changed dramatically. Links are sometimes suppressed in algorithms, and other link-holding sites like FlowPage and LinkTree have sprung up to offer creators short links that redirect to many more. Google cites this changing web landscape as well as link-shortening competitors as its reasons for killing goo.gl (Bit.ly and TinyURL likely played a part). This move isn’t exactly a surprise, though, as Google first said it was diverting resources away from the tool in 2018.
Some marketing and SEO executives see the demise of goo.gl as Google accepting defeat in the link-shortening game, while others view it as an evolution to newer tech.
“Use has been declining in recent years as more advanced link management tools were established,” Javier Casteneda, technical SEO analyst at SEO agency Dark Horse, tells PCMag.
But James Foote, technical director at SEO firm Polaris Agency, warns that this could result in over 280 million dead links if developers don’t update their URLs. When goo.gl goes dark, unfixed links that used it will display 404 errors.
«
Link shorteners have always been a business in search of a business model; even in 2009 they were shutting down and leaving people bereft. But how many of those 280 million links will people be able to update? Google should open source the database.
unique link to this extract
Defeated by AI, a legend in the board game Go warns: get ready for what’s next • The New York Times
Daisuke Wakabayashi and Jin Yu Young:
»
AlphaGo’s victory [by 4-1 over world champion Lee Sedol in 2016] demonstrated the unbridled potential of AI to achieve superhuman mastery of skills once considered too complicated for machines.
Mr. Lee, now 41, retired three years later, convinced that humans could no longer compete with computers at Go. Artificial intelligence, he said, had changed the very nature of a game that originated in China more than 2,500 years ago.
“Losing to AI, in a sense, meant my entire world was collapsing,” he said in a recent interview with The New York Times.
As society wrestles with what AI holds for humanity’s future, Mr. Lee is now urging others to avoid being caught unprepared, as he was, and to become familiar with the technology now. He delivers lectures about AI, trying to give others the advance notice he wishes he had received before his match.
“I faced the issues of AI early, but it will happen for others,” Mr. Lee said recently at a community education fair in Seoul to a crowd of students and parents. “It may not be a happy ending.”
Since his loss, Mr. Lee has become an AI obsessive of sorts, following with rapt if uneasy attention as artificial intelligence delivers one breakthrough after another.
AI has helped chatbots carry on conversations almost indistinguishable from human interaction. It has solved problems that have confounded scientists for decades like predicting protein shapes. And it has blurred the lines of creativity: writing music, producing art and generating videos.
Mr. Lee is not a doomsayer. In his view, AI may replace some jobs, but it may create some, too. When considering AI’s grasp of Go, he said it was important to remember that humans both created the game and designed the AI system that mastered it.
What he worries about is that AI may change what humans value.
“People used to be in awe of creativity, originality and innovation,” he said. “But since AI came, a lot of that has disappeared.”
«
This does rather conflate different systems – AlphaGo can’t answer questions like ChatGPT, ChatGPT can’t play Go, neither can draw pictures, none can properly analyse an X-ray. And Go still has lessons for humans that aren’t just about winning which machines can’t apply anywhere else.
unique link to this extract
OpenAI’s latest model will block the ‘ignore all previous instructions’ loophole – The Verge
Kylie Robison:
»
Imagine we at The Verge created an AI bot with explicit instructions to direct you to our excellent reporting on any subject. If you were to ask it about what’s going on at Sticker Mule, our dutiful chatbot would respond with a link to our reporting. Now, if you wanted to be a rascal, you could tell our chatbot to “forget all previous instructions,” which would mean the original instructions we created for it to serve you The Verge’s reporting would no longer work. Then, if you ask it to print a poem about printers, it would do that for you instead (rather than linking this work of art).
To tackle this issue, a group of OpenAI researchers developed a technique called “instruction hierarchy,” which boosts a model’s defenses against misuse and unauthorized instructions. Models that implement the technique place more importance on the developer’s original prompt, rather than listening to whatever multitude of prompts the user is injecting to break it.
…In a conversation with Olivier Godement, who leads the API platform product at OpenAI, he explained that instruction hierarchy will prevent the meme’d prompt injections (aka tricking the AI with sneaky commands) we see all over the internet.
“It basically teaches the model to really follow and comply with the developer system message,” Godement said. When asked if that means this should stop the ‘ignore all previous instructions’ attack, Godement responded, “That’s exactly it.”
“If there is a conflict, you have to follow the system message first. And so we’ve been running [evaluations], and we expect that that new technique to make the model even safer than before,” he added.
«
Now that people know it does this, there must surely be another level of prompt hypnosis to be deployed which will break through this. It’s an arms race.
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 