More than 100,000 full-body MRI scans will inform future medicine in the UK. CC-licensed photo by liz west 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. Still transparent. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.
Anthropic, Google and xAI win $200m each from Pentagon AI chief for ‘agentic AI’ • Breaking Defense
Sydney J. Freedberg Jr.:
»
The military’s central artificial intelligence hub has quadrupled down on its investment in commercial “frontier AI.”
This morning, the Pentagon’s Chief Digital & AI Office (CDAO) announced that it would split $600m in contracts evenly among Anthropic, Google, and xAI, following on a similar $200 million award to OpenAI announced last month.
If CDAO exercises all its options on all four contracts — which isn’t guaranteed — that’s a total of $800m the Pentagon is pouring not into bespoke military R&D from dedicated defense contractors, but into widely available, widely applicable commercial tech.
xAI and OpenAI also both used their awards to announce the launch of ‘For Government’ business units, in the case of xAI using its Grok platform.
The embrace of “commercial off-the-shelf” has been especially notable in AI. After OpenAI kicked off the current generative AI(GenAI) explosion with its launch of ChatGPT in November 2022, the Pentagon, much like the private sector, scrambled to understand the new technology, launching the high-level Task Force Lima that conducted almost 18 months of studies before blessing GenAI as an ongoing area investment. Since then, CDAO has partnered with the Army’s Enterprise LLM Workspace to bring a toolkit combining multiple commercially available GenAI models to a wide array of Defense Department offices.
…agentic AI would allow computers not only to generate plans but to take some kind of action on them. You might ask GenAI to devise an itinerary for your vacation and identify the best hotels and restaurants in the area; an agentic AI, however, would actually be able to book the reservations with your credit card. The military has already experimented with using AI agents to do staff work that would previously have required a human, while severely restricting — albeit not completely prohibiting — any project that would give software the ability to use lethal force without human authorization.
«
This isn’t a huge amount of money in the context of the US military. But it certainly is for those companies. It could be transformative. And don’t forget that the US military was one of the first to use drones – although their small-scale deployment was more effectively done by other militaries. (Thanks Gregory B for the link.)
unique link to this extract
The FBI’s Jeffrey Epstein prison video had nearly three minutes cut out • WIRED
Dhruv Mehrotra:
»
Newly uncovered metadata reveals that nearly three minutes of footage were cut from what the US Department of Justice and Federal Bureau of Investigation described as “full raw” surveillance video from the only functioning camera near Jeffrey Epstein’s prison cell the night before he was found dead. The video was released last week as part of the Trump administration’s commitment to fully investigate Epstein’s 2019 death but instead has raised new questions about how the footage was edited and assembled.
WIRED previously reported that the video had been stitched together in Adobe Premiere Pro from two video files, contradicting the Justice Department’s claim that it was “raw” footage. Now, further analysis shows that one of the source clips was approximately 2 minutes and 53 seconds longer than the segment included in the final video, indicating that footage appears to have been trimmed before release. It’s unclear what, if anything, the minutes cut from the first clip showed.
The nearly three-minute discrepancy may be related to the widely reported one-minute gap—between 11:58:58 pm and 12:00:00 am—that attorney general Pam Bondi has attributed to a nightly system reset. The metadata confirms that the first video file, which showed footage from August 9, 2019, continued for several minutes beyond what appears in the final version of the video and was trimmed to the 11:58:58 pm mark, right before the jump to midnight.
«
I don’t think this really matters in the least – personally I don’t believe Epstein was murdered, don’t @ me – but it’s wonderful to see Wired trolling the Trump supporters who were convinced that absolutely everything relating to Epstein would be released. Now they’re getting a taste of their own pernickety medicine.
unique link to this extract
End-of-Train and Head-of-Train Remote Linking Protocol • CISA
CISA is the US’s Cybersecurity and Infrastructure Agency, and this is a page which has a newly announced security vulnerability:
»
2. RISK EVALUATION
Successful exploitation of this vulnerability could allow an attacker to send their own brake control commands to the end-of-train device, causing a sudden stoppage of the train which may lead to a disruption of operations, or induce brake failure.
«
Let’s be clear: “sudden stoppage of the train” could be catastrophic, not just a “disruption”.
Good news! A fix is in the works.
Bad news! Won’t be here until 2027. More on the topic at Cybersecurity Dive, or you can read a long thread by one of the people who discovered it – it turns out it has been discovered independently at least three times since 2005.
unique link to this extract
NSA: Volt Typhoon was ‘not successful’ at persisting in critical infrastructure • The Record
Jonathan Greig:
»
Senior cybersecurity officials at the National Security Agency and FBI said the agencies have been successful in addressing some of the Chinese cyber campaigns targeting critical infrastructure in the U.S.
During the International Conference on Cyber Security at Fordham University in New York City on Tuesday, experts spoke at length about Beijing’s so-called Typhoon campaigns — which have involved Chinese government and private sector groups launching attacks on U.S. government agencies and companies.
Kristina Walter, director of the NSA’s Cybersecurity Collaboration Center, focused on Volt Typhoon, an effort by Chinese actors to preposition themselves on U.S. critical infrastructure for disruptive or destructive cyberattacks in the event of a kinetic conflict centered around Taiwan.
“The good news is, they really failed. They wanted to persist in domestic networks very quietly for a very long time so that if and when they needed to disrupt those networks, they could. They were not successful in that campaign,” she said.
“We, with private sector, with FBI, found them, understood how they were using the operating systems, how they’re using legitimate credentials to maintain persistence, and frankly, we equipped the entire private sector and U.S. government to hunt for them and detect them.”
Walter did not offer further details about those efforts. She said that after the NSA and other agencies released a public advisory in 2024, owners of critical infrastructure reached out to them to confirm that they found evidence of Volt Typhoon and ask for help.
«
Might still be in train brakes, though.
unique link to this extract
World of Good Brands sells portfolio, axes staff, shutters • AdWeek
Mark Stenberg:
»
Digital media company World of Good Brands has sold off its three remaining properties—Well+Good, Livestrong, and Only In Your State—and shuttered, according to four people familiar with the situation.
The bulk of the staff, excluding some team members who moved with the respective brands, were laid off.
A spokesperson for its parent company, Graham Holdings, confirmed the closures. The company declined to share how many people were affected by the layoffs, though between 50 and 200 people worked at World of Good Brands, according to LinkedIn.
«
You’re thinking: why does anyone care about “World of Good Brands”? The answer: because it used to be the original search engine content farm, Demand Media. The number of reinventions and renamings would do a chameleon proud. But inevitably Google won the battle to get it out of search results.
Now it’s all part of a melange where people say things like
»
“Launch Potato is excited to integrate Only In Your State into its portfolio of brands,” said founder Greg Van Horn.
«
Baidu strikes deal to bring its driverless cars to Uber globally • CNBC
Arjun Kharpal:
»
Baidu has struck a partnership with Uber to deploy its autonomous cars on the ride-hailing giant’s platform outside the US and mainland China.
The first deployments are expected to happen in Asia and the Middle East later this year. The two companies said the multi-year partnership will see “thousands” of Baidu’s Apollo Go autonomous vehicles on Uber globally.
Baidu shares jumped 4.5% in premarket trade in the US following the news, while Uber stock was more than 1% higher.
For Baidu, the move will help to internationalize its driverless car business outside of China. Uber will meanwhile gain a proven partner to take on autonomous driving rivals around the world. After the launch, a rider requesting a trip on Uber may be given the option to hail a driverless Apollo Go car, the two companies said.
In China, Baidu has been operating its own robotaxi service since 2021 in major cities like Beijing, letting users hail an Apollo Go car through the app.
But the Beijing-headquartered company has stated its ambitions to expand its operations to other international markets. This year, Baidu announced plans to begin autonomous driving testing and services in Dubai, while CNBC reported that the company is also looking to expand into Europe.
«
Ambitious, and if this is the pattern we see then American companies are going to have a huge fight on their hands.
unique link to this extract
World’s biggest human imaging project reaches new milestone after scanning 100,000th UK volunteer • Sky News
Tom Clarke:
»
Steve’s morning starts lying still in the clanging magnet of an MRI machine as his body is slowly scanned from neck to knee in intimate detail.
Then it’s on to another MRI scanner, followed by X-rays of his bones, ultrasound on his neck, blood and other samples, medical tests and questionnaires – in all five hours of his time.
A test of patience you’d admire in any patient – only Steve is perfectly healthy. He’s a volunteer in the UK Biobank project, giving up his time to help complete the world’s largest medical imaging dataset.
His motivation: that his data may help where he can’t. “My mum in particular at the moment now, is suffering from early stages of dementia, close friends have had cancer.”
“Giving up my time now… is going to help medical research in the future.”
Even more remarkable is that Steve is the 100,000th volunteer to have willingly gone through the process. Each one allows their carefully anonymised images (it’s why we’re only using Steve’s first name) as well as their biological samples, medical and lifestyle histories, available to the world’s medical researchers in perpetuity.
“The unprecedented scale of this imaging project – more than 10 times bigger than anything that existed before – makes it possible for scientists to see patterns of disease that just couldn’t otherwise be seen,” said Professor Sir Rory Collins, chief executive of UK Biobank.
“Combining these images from different parts of the body with all the genetic and lifestyle information from our volunteers, scientists are getting a far better understanding of how our bodies work,” he said.
Given the time and complexity of whole-body imaging, it’s a project many scientists believed would never work. “When we started, some people thought that we got our numbers wrong,” said Prof Naomi Allen, chief scientist at UK Biobank. “Surely we wanted to scan at 10,000 participants… not 100,000. And yet, here we are.”
«
An amazing resource – as much as anything, the potential for AI training and application is enormous. Reap all the benefits that we can as soon as we can.
unique link to this extract
Nvidia passes $4 trillion market cap for first time • CNBC
Samantha Subin,Kif Leswing:
»
Nvidia stock rose last Wednesday lifting the company’s market cap briefly past $4 trillion for the first time as investors scooped up shares of the tech giant that’s building the bulk of the hardware for the generative artificial intelligence boom.
However, Nvidia stock ended finishing the day only up 1.8%, giving the company a market cap of $3.97 trillion. [It’s now solidly past the $4 trillion mark – Overspill Ed.]
Nvidia is the world’s most valuable company, surpassing Microsoft and Apple, both of which hit the $3 trillion mark before Nvidia. Microsoft is also one of Nvidia’s biggest and most important customers. The chipmaker is the first company to ever achieve this market value during trading.
The California-based company, which was founded in 1993, first passed the $2 trillion mark in February 2024, and surpassed $3 trillion in June.
Nvidia has profited heavily from the growing demand for AI hardware and chips since the launch of ChatGPT in late 2022. The company has positioned itself as the decisive leader in creating the graphics processing units that power large language models.
«
A reminder that market capitalisation is the market’s guess of the net present value of the company’s total future profits. That is, by the time Nvidia is a worthless husk (perhaps along with the planet? Who knows), the market thinks it will have reaped profits that, in today’s money, total $4 trillion plus. That could finance a few wars, if you felt like it.
unique link to this extract
UK set up secret Afghan immigration scheme and gagged media after data leak • Financial Times
Lucy Fisher and Alistair Gray:
»
The names, contact information and other personal details of about 25,000 Afghans, people who worked closely with the UK before the Taliban seized power and some of their family members, were accidentally disclosed by a British soldier in emails in February 2022.
…The database was a detailed record of individuals who had applied — in most cases unsuccessfully — under the public Arap scheme, which offered relocation to the UK for those at risk of reprisals after they worked for or alongside the UK before the Taliban retook power.
UK combat operations ended in Afghanistan in 2014 after 13 years, but British troops remained until a chaotic western withdrawal in 2021 that allowed the Taliban’s return.
The UK government did not discover the leak until an anonymous person posted screenshots of the spreadsheet on Facebook in August 2023 and threatened to disclose the entire database.
One of the people familiar with the breach said the database had been sold, at least once, for a five-figure sum.
They claimed that one of the Afghan recipients used their possession of the database as leverage to pressure the government to relocate themselves and 14 family members to the UK.
The identity of the soldier, or whether they have been sanctioned, has not been revealed by the MoD. The department has not successfully contained the leak and it is not known whether the Taliban has obtained the list.
More than 665 Afghans have started a collective legal action to sue the MoD over the data breach, seeking at least £50,000 each, with the potential for thousands more people to join the lawsuit once they learn of the incident and their potential exposure.
«
I’m very intrigued by these emails from the soldier, and how big they were if they included so much personal information of so many people. There are echoes of so many American military leaks which come from people just boasting about what they have access to, rather than espionage. (Thanks Gregory B for the pointer.)
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 