Tests show that ChatGPT is a lot worse than human journalists at summarising science papers. CC-licensed photo by Nic McPhee 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.
There’s another post coming this week at the Social Warming Substack on Friday at 0845 UK time. Free signup.
A selection of 9 links for you. In brief. 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 great falls of Boeing, Intel, and Apple • Hey
David Heinemeier Hansson:
»
It takes ten years for the culture of a great company to fall apart once the CEO seat is given to someone without an engineering or product background. That’s been the story of Boeing, Intel, and now Apple. Legendary American companies that all got lost when a bean counter, marketing man, or logistics hand took over.
Boeing’s troubles started when they were taken over by McDonnell Douglas in 1997, but really accelerated after 2005 when they installed their first CEO with no aerospace background. The result, after ten years of cost-cutting and outsourcing, was the 737 MAX MCAS tragedies, and an organization gutted of ambition and engineering pride.
Intel did the same thing, and almost at the same time. In 2005, they too installed their first CEO without an engineering background. Ten years later, they were stumbling with delayed nodes, stalled progress, and no answers on mobile. Now the entire business is teetering.
Finally, Apple. Steve Jobs handed the reins to Tim Cook in 2011, but such was the strength of the product pipeline and culture that Jobs left behind, that it initially looked like Cook could break the spell. Show that it was possible for a logistics man to steer one of the great ships of American ingenuity and tech supremacy.
But now the ten-year curse is hitting Apple with an eerily familiar thud. They wasted a decade chasing a self-driving dream without direction, and ended up with the worst possible car interface to show for it. They completely missed the boat on AI, and embarrassed themselves with Genmoji and vaporware ads. And the Vision pro has been an expensive tech demo that nobody actually wanted to wear three months after they bought it.
The profits still gush from glories past, and the tollbooth operation on the App Store, but the soul has left the machine.
While these three stories are different, they’re drawn from the same archetype: Great companies need bold, hands-on leaders who live and breathe the stuff they make or they’ll eventually hollow out.
«
Hansson’s implied solution seems to be that Apple should rehire Scott Forstall (fired by Cook in October 2012). I wonder if he might be a little out of the loop now, though.
unique link to this extract
DOJ aims to break up Google’s ad business as antitrust case resumes • Ars Technica
Ryan Whitwam:
»
Google’s had mixed luck with antitrust rulings lately, but it’s not a great sign that Google has so many legal woes that it can be hard to keep them all straight.
The case that just got underway is the remedy phase of the AdTech trial, in which the DOJ secured a ruling against Google several months ago. The remedy phase of the search trial wrapped up recently, which ended with Google holding on to Chrome but pledging an appeal to overturn the verdict. There’s also the Google Play antitrust case, which was brought by Epic Games. In this case, Google has already lost its appeal, putting some major app changes on the table as it plans yet another appeal.
The Department of Justice (DOJ) and Google are squaring off in Virginia federal court for the next two weeks or so, and there are no surprises in opening arguments. The government says the only way to deal with a monopolist like this is to break it up, but Google says it has already made numerous changes, and there’s no way to excise it from online advertising without breaking the market.
The AdTech remedy trial could mirror the search trial to a great degree. Indeed, the DOJ has pulled some language from that case, in which Judge Mehta opted not to force a divestment of Chrome. Mehta ruled that forcing a Chrome sale was a poor fit for the remedies as Chrome was not part of the illegal conduct.
However, government lawyers are hoping the AdTech case will turn out differently. The DOJ is asking the court to force Google to spin off Google Ad Manager (formerly Ad Exchange or AdX), the marketplace through which advertisers buy ads on Google’s platform. The government was able to convince the court that Google’s control of Ad Manager gave it an unfair advantage that boosted its own services, but is a breakup the proper remedy?
In its opening arguments in the AdTech case, the government claims Ad manager was intimately tied to the antitrust behavior, and its proposed remedies would pass muster under the standard Mehta employed. Government lawyers contend that remedies must be designed to restore competition, and Google’s iron grip on online display ads can only be solved in one way. “Nothing short of a structural divestment is sufficient to bring meaningful change,” said the DOJ’s Julia Tarver Wood.
«
Google has offered its own remedies, which as Ars Technica notes “is just shy of nothing” – including a promise not to use its ability to peek at first and last prices in auctions. Which, side note, it stopped doing some years ago. But Google’s intent is clearly to let the remedies trial happen and then appeal it.
unique link to this extract
Double harvest: vertical solar panels and crops thrive side by side • Techxplore
Gaby Clark:
»
Imagine a field where solar panels and crops coexist—with no trade-off. It sounds like science fiction, but that’s precisely what researchers from Aarhus University have now documented in a full-scale agrivoltaic pilot project in the Danish countryside.
“Our measurements show that wheat and grass-clover mixtures grow just as well between vertical solar panels as in open fields. At the same time, the panels produce electricity in a daily pattern that better matches energy demand. It’s a win-win,” says Marta Victoria, lead author of the study and Associate Professor at the Department of Mechanical and Production Engineering, Aarhus University.
The study is published in the journal Energy Nexus.
At the test site in Foulum, researchers installed two types of bifacial solar panels: one traditional, south-facing tilted system, and one vertical, east-west-facing system. The vertical panels produce slightly less electricity per year—but with higher value, as generation peaks coincide with morning and late afternoon demand.
At the same time, crops growing among the vertical panels showed no decline in yield.
“Even with some shade, the yield per square meter is almost the same. The crops don’t seem to mind the presence of solar panels and they like the wind protection that they provide,” explains Professor Uffe Jørgensen from the Department of Agroecology, Aarhus University.
And because the panels only occupy about 10% of the field area, the combined system requires much less land than separate installations. “If we were to produce the same amount of electricity and food using separate land, we would need 18–26% more area,” the researchers calculate.
«
Quite a clever method – and useful for more northerly latitudes.
unique link to this extract
We’re suing ICE for its $2m spyware contract • 404 Media
Joseph Cox, Samantha Cole, Emanuel Maiberg and Jason Koebler:
»
On Monday 404 Media filed a lawsuit against Immigration and Customs Enforcement (ICE) demanding the agency publish its $2m contract with Paragon, a company that makes powerful spyware that can remotely break into mobile phones without the target even clicking a link. The sale of the spyware to ICE has activists and lawmakers deeply concerned about what the agency, which continues to push the Trump administration’s mass deportation effort, may use the technology for. The contract and related documents 404 Media is suing for may provide more information on what ICE intends to do with the spyware.
“404 Media has asked ICE to disclose agency records relating to its contract with a company known for its powerful spyware tool whose potential use in the agency’s ongoing mass-deportation campaign has prompted lawmakers, civil liberties organizations, and immigration groups to express deep concerns over potential civil rights abuses,” the lawsuit says.
404 Media first filed a Freedom of Information Act (FOIA) request with ICE for documents related to its Paragon purchase in September 2024. Under the law, agencies are required to provide a response within 20 days, or provide an explanation of why they need more time. ICE acknowledged receipt of the request in September 2024, but has not since replied to any follow up inquiries. 404 Media then filed the lawsuit.
…The contract itself is for “a fully configured proprietary solution including license, hardware, warranty, maintenance, and training,” according to a description included in a public US procurement database. The funding office for the purchase is listed as a division of Homeland Security Investigations (HSI). It is not clear if the ICE deal is for a custom-made tool or for some version of Paragon’s flagship “Graphite” software.
Graphite is capable of letting police remotely break into messaging apps like WhatsApp, Signal, Facebook Messenger, and Gmail according to a 2021 report from Forbes. While other government spyware tries to take over an entire device allowing all sorts of other capabilities, Paragon sets itself apart by promising to access just the messaging applications, according to Forbes.
«
Brave, worthwhile journalism – though lawsuits aren’t cheap to do. They are accepting donations (donate@404media.co).
unique link to this extract
Nvidia is partnering up with OpenAI to offer compute and cash • The Verge
Hayden Field:
»
OpenAI is teaming up with Nvidia via a “strategic partnership” that will get the ChatGPT-maker more compute and more cash to develop new models on the road to superintelligence.
The partnership, announced Monday, will allow OpenAI to “build and deploy at least 10 gigawatts of AI datacenters with NVIDIA systems,” which translates to millions of GPUs that can help power OpenAI’s new models. One of the most important points here, besides more data centers and compute — which are always in high demand for companies like OpenAI — is that as part of the deal, NVIDIA “intends to invest up to $100 billion in OpenAI progressively as each gigawatt is deployed,” per the release. The details will be finalized in the next few weeks, according to the companies.
Nvidia will now be a “preferred strategic compute and networking partner for [OpenAI’s] AI factory growth plans,” OpenAI said.
“Everything starts with compute,” OpenAI CEO Sam Altman said in a statement. “Compute infrastructure will be the basis for the economy of the future, and we will utilize what we’re building with NVIDIA to both create new AI breakthroughs and empower people and businesses with them at scale.” Last week, OpenAI said ChatGPT had reached 700 million weekly active users.
«
Those outside looked from OpenAI to Nvidia, and from Nvidia to OpenAI, and from OpenAI to Nvidia again, but already it was impossible to say which was which.
unique link to this extract
Wegovy and Zepound can cut future health costs. The fight is over who pays now. – The Washington Post
Daniel Gilbert:
»
Millions of Americans who could benefit from GLP-1 weight-loss drugs are caught in the middle of a battle between drug companies and insurers over their costs, leaving them without coverage even as evidence mounts that the drugs could stave off expensive health complications in the future.
Insurance coverage for the drugs has barely budged in the last year. Eli Lilly said in August that around 50% of employers had chosen to cover its weight-loss drug, Zepbound, little changed from a year earlier. Novo Nordisk said last month that about 40 million people have access to anti-obesity drug Wegovy through commercial insurance, roughly the same as at the end of 2023.
Evidence from clinical trials shows that GLP-1 injections have benefits far beyond weight loss, helping prevent people from developing diabetes or suffering a heart attack years down the road — improving lives and ultimately blunting costs for health plans. But while the savings might only be realized over decades, experts say, the prescription costs must be paid starting now, threatening to overwhelm health plans’ financial reserves because of their immense popularity.
…If the GLP-1 drugs were treatments for a disease afflicting a small population, “we’d go, ‘this is fantastic’,” said David Rind, chief medical officer at the Institute for Clinical and Economic Review, which studies the value of prescription drugs. “But if you want to give these drugs to 40% of the U.S. population, it doesn’t matter that these drugs are cost-effective,” he said, referring to the ability to pay for them. The scale of who could benefit from GLP-1s, he said, is “probably almost unique.”
About 1 in 8 adults has taken a GLP-1 drug, according to a 2024 KFF survey.
«
Finally, something which might break America’s health system. Though probably not in a good way.
unique link to this extract
2003: There will come soft sprinklers • The Inquirer
Wendy Grossman, writing in August 2003:
»
This week IBM flew a load of Asia-Pacific and one lone “European” journalist to Austin to see the company’s vision of a “Smart House” — Ray Bradbury’s vision made silicon and radio waves, fifty years on. I am huddled in my Polartec, inexplicably and inexorably drawn to the refrigerator, toying with it while the Father of the House, researcher Bill Bodin, is showing off how he can control the lighting and microwave oven with a $5 gadget, a Bluetooth gateway, broadband, and a voice recognition server in Florida: “Lights on”. The lights come up, unlike the Steve Martin movie L.A. Story, in which the voice command “Dial Mom” gets him the pizza parlor.
I am fighting to look up the day’s local music events (Austin! the music capital of the world!) on Yahoo! For Refrigerators when Bodin shifts to show us the refrigerator’s text recognition system. A fictitious family member scribbles a note on the screen’s notepad, the house server parses it to pick out the names of the people in it, and sends a text copy via SMS to those people’s mobile phones.
…Over in the living room, the main TV set calls up the “Home page”. Click on sprinkler. On the configuration page is a little box to check labeled “Observe municipal water restrictions”. No more reading the newspaper or listening to the radio during droughts trying to parse which instructions apply to your house. (That’s two companies you’re going to put out of business, Bodin, did you think of that?) The car downstairs is “self-healing”. It monitors its systems, sends out diagnostic data, and even makes its own servicing appointments with the company that provides 80% of US car servicing. When you gas up, the pump reminds you that you’ve left the car door open. Do you ever get the feeling that things are talking about you behind your back?
The coming RFID revolution will have tiny electronic tags embedded in more or less everything we buy. The food in your fridge will be data. Bodin is happy: the kitchen counter will guide you through a recipe appropriate to what you have on hand. I think companies will shift from selling you the Spam in your fridge to spamming your fridge.
«
And lo, as noted last week, it did come to pass. Nice to have a prediction come true, less nice to be the one like that. (Thanks Wendy G – obviously – for the link.)
unique link to this extract
Science journalists find ChatGPT is bad at summarizing scientific papers • Ars Technica
Kyle Orland:
»
Summarizing complex scientific findings for a non-expert audience is one of the most important things a science journalist does from day to day. Generating summaries of complex writing has also been frequently mentioned as one of the best use cases for large language models (despite some prominent counterexamples).
With all that in mind, the team at the American Association for the Advancement of Science (AAAS) ran an informal year-long study to determine whether ChatGPT could produce the kind of “news brief” paper summaries that its “SciPak” team routinely writes for the journal Science and services like EurekAlert. These SciPak articles are designed to follow a specific and simplified format that conveys crucial information, such as the study’s premise, methods, and context, to other journalists who might want to write about it.
Now, in a blog post and white paper discussing their findings, the AAAS journalists have concluded that ChatGPT can “passably emulate the structure of a SciPak-style brief,” but with prose that “tended to sacrifice accuracy for simplicity” and which “required rigorous fact-checking by SciPak writers.”
“These technologies may have potential as helpful tools for science writers, but they are not ready for ‘prime time,’ at this point for the SciPak team,” AAAS writer Abigail Eisenstadt said.
…In total, 64 papers were summarized, and those summaries were evaluated both quantitatively and qualitatively by the same SciPak writers who had briefed those papers for the AAAS. The researchers note that this design “could not account for human biases,” which we’d argue might be significant among journalists evaluating a tool that was threatening to take over one of their core job functions.
Still, the quantitative survey results among those journalists were pretty one-sided. On the question of whether the ChatGPT summaries “could feasibly blend into the rest of your summary lineups, the average summary rated a score of just 2.26 on a scale of 1 (“no, not at all”) to 5 (“absolutely”).
«
So you’re saying they could, a bit? They might be better than some journalists, but won’t be better than many. Plus they won’t be able to ring up the scientists involved and ask them to explain their work. Because, shocking though it might sound, journalism – even science journalism summarising papers! – often, and should, involve talking to humans.
unique link to this extract
Why Bangladesh’s unlikely satellite engineers are still waiting for liftoff – Rest of World
Jesmin Papri:
»
The launch of Bangladesh’s first satellite sparked an explosion of national pride and a mini space race in the country’s colleges seven years ago. Now, the local engineers who control it are wondering what’s next.
The South Asian nation’s first satellite was built in France and launched on a SpaceX rocket in 2018. Bangladeshi engineers have since kept it orbiting.
The country invested close to $250 million in the effort. It was a high price for Bangladesh, where the average income is less than $3,000 per year. Still, it brought more reliable connectivity to companies, bureaucracies, and the military. It also expanded access to television channels for millions.
It put Bangladesh on the map as one of the few countries with satellites. Dhaka celebrated it as a beacon of a new space era. The elite engineers managing the satellite from the ground were hailed as heroes.
“We’ve hoisted the Bangladesh flag in space,” former Prime Minister Sheikh Hasina had said when she announced the launch. The satellite was named after her father, a founder of the country. “The Bangabandhu Satellite-1 will certainly bring revolutionary changes in our broadcast and telecommunications sector.”
Seven years later, Hasina has been ousted from power, the satellite program is losing money, and plans for a successive satellite are uncertain.
Some of the two dozen engineers who keep the satellite and its ground communications stations running are worried.
“All of our expertise is tied to this one satellite,” a member of the team, who asked not to be named because he is not authorized to speak to the media, told Rest of World. “We’re stuck.”
«
What if you launched a satellite and nothing came next?
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 
DHH has been bitter about Apple’s AppStore policies for years. He and a handful of other developers just want to get all the benefits of Apple’s ecosystem and one billion+ wealthy customers for free.
Apple’s product lineup is better and better value or money than ever – especially the Macs with their custom silicon. Incredible turnaround from the last few Intel years which weren’t great. Easily the best laptops one can buy. This doesn’t happen at a company who has supposedly lost their way in everything.
It’s also very hard to argue against the value proposition of cheapest iPads and Watches, when compared to the competition. Or the Airpods Pro. Or even the iPhone.
Nobody knows if Apple has missed any sort of boat worth boarding when it comes to AI. They will obviously never release their own ChatGPT because it makes zero sense, but yet people are expecting them to. We haven’t really seen much genuinely useful AI integrations in any devices yet, from any vendor.