An emerging use of AI is to fake images for expenses claims. Ambitious or stupid? CC-licensed photo by Joe Loong 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. Honestly. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.
Meet the humanoid robot headed for homes • Fast Company
Harry McCracken:
»
With one sweeping gesture, Dar Sleeper hoists the humanoid robot off the ground. Bracing its back with one arm and its legs with the other, he gently carries it across the room and lowers it onto a sofa, where it lies in repose as if catching a quick nap.
It’s a slightly surreal scene, but it has a serious point. I am visiting the Palo Alto headquarters of 1X Technologies, and Sleeper, the company’s VP of growth, is demonstrating that Neo, its home robot, is a lightweight at a mere 66 pounds [30kg]. That’s a crucial design feature, given that a weighty domestic bot could prove hazardous if it toppled over in the vicinity of a human, a pet, or just a pricey vase.
Soon, Neo will take on the ultimate proving ground for a home robot: actual homes. 1X is announcing that it’s taking preorders and plans to ship units to its earliest customers next year. The price is $20,000, or $499 per month as a subscription service, with a six-month minimum. Like a smartphone, the robot will come in multiple color options—tan, gray, and dark brown.
Wait, $20,000? There isn’t much precedent for mainstream consumer products in that price range. Cars, of course. (The average price for a new one just topped $50,000.) Maybe boats? Even if you can come up with another example or two, it’s a short list.
Then again, 1X founder and CEO Bernd Børnich’s goals for his robots involve attaining a degree of utility that few inventions ever have. He wants to teach Neo to handle every household task that people perform because they need to, not because they want to. Even if the time saved came in bits and pieces—five minutes of dishwasher unloading here, 15 minutes of laundry folding there—it would add up to many hours newly available for more rewarding pursuits.
…By 2050, according to a Morgan Stanley report in May, there could be more than a billion humanoid robots in service, with the vast majority in commercial and industrial settings. Only 80 million would be in homes, the report speculates.
«
A 30-kilo weight falling over will make quite a mess. I don’t believe that Morgan Stanley report, but if it’s only off by an order of magnitude – even two – that’s 800,000 humanoid robots in homes (or similar, such as care homes?), which is substantial. Of course Elon Musk will say he’s got something better coming in two years.
unique link to this extract
Nvidia invests $1bn in Nokia to push AI to the edge • The Register
Thomas Claburn:
»
Nvidia’s deal with Nokia, one of the largest telecom equipment vendors in the world, will give the company access to Nvidia’s AI-RAN products, which make AI available to radio access networks to improve spectral efficiency (AI for RAN) and also will make AI available via cloud computing for wireless communications (AI on RAN). And Nokia will ensure that its 5G and 6G software runs on Nvidia hardware.
The partnership comes with a $1bn Nvidia investment in Nokia, which lifted the telecom company’s shares more than 25% in the hours after the announcement. Nvidia has also recently announced investments in Intel ($5bn) and OpenAI ($100bn). The GPU biz has a lot of spare change lying around.
“Telecommunications is a critical national infrastructure – the digital nervous system of our economy and security,” said Huang in a statement. “Built on Nvidia CUDA and AI, AI-RAN will revolutionize telecommunications – a generational platform shift that empowers the United States to regain global leadership in this vital infrastructure technology.”
In conjunction with the partnership, Nvidia is rolling out Aerial RAN Computer Pro (ARC-Pro), a 6G-ready accelerated computing platform for telecommunication companies.
«
6G! I’ve hardly made use of all the incredible benefits of 5G, such as inter-car connectivity, robotics, virtual reality, and the internet of things! Wonder what wonders 6G will bring.
unique link to this extract
How the UK lost its shipbuilding industry • Construction Physics
Brian Potter:
»
during the war, developments had taken place that would threaten the skilled labor-intensive production model that British shipbuilders relied on. To win the Battle of the Atlantic and overcome destruction to their fleet caused by German U-boats, the US had used welded, prefabricated construction to rapidly build enormous numbers of simple cargo ships: Liberty ships, Victory ships, and T-2 Tankers. In the 1950s, those methods of ship construction were brought to Japan, where they continued to be refined.
British shipbuilders could have taken advantage of these methods as well. They had seen firsthand the huge number of Liberty ships American shipyards were producing (the Liberty ship was, after all, originally a British design), and had made use of welded, prefabricated construction themselves to build vessels during wartime.
But British shipbuilders perceived adopting these radically different production methods as risky. It would require enormous capital expenditure, and British shipbuilders had only survived the brutal 1930s thanks to their comparatively light overheads and labor-intensive methods. Dramatic changes to production methods would also require changes to the very strict demarcation system of its unionized labor force, which the unions, naturally distrustful of shipyard operators, were sure to resist.
And while the US had successfully built thousands of ships very rapidly during the war, it had come at a cost: the US cargo ships, using modern methods, were more expensive to build than similar British ships. Though some British shipbuilders recognized the potential of welding and prefabrication, they weren’t clearly worth reorganizing the entire industry around.
…Thus, even as the world shipbuilding market boomed in the 1950s, UK shipbuilding output stayed roughly constant. Between 1947 and 1957, UK ship output rose by 18%, while worldwide output comparatively rose by over 300%. Countries like Germany, Sweden, and Japan picked up the slack. The UK lost its position as the world’s biggest shipbuilder by tonnage to Japan in 1956.
«
The US figured out welding and preconstruction in WW2, and having got a new technology, refined it and made it cheaper. The UK’s story is repeated again and again. (Apologies: I can’t find the person’s tweet/skeet/email/whatever that pointed me here.)
unique link to this extract
New image-generating AIs are being used for fake expense reports • Financial Times via Ars Technica
Cristina Criddle:
»
Businesses are increasingly being deceived by employees using artificial intelligence for an age-old scam: faking expense receipts.
The launch of new image-generation models by top AI groups such as OpenAI and Google in recent months has sparked an influx of AI-generated receipts submitted internally within companies, according to leading expense software platforms.
Software provider AppZen said fake AI receipts accounted for about 14% of fraudulent documents submitted in September, compared with none last year. Fintech group Ramp said its new software flagged more than $1m in fraudulent invoices within 90 days.
About 30% of US and UK financial professionals surveyed by expense management platform Medius reported they had seen a rise in falsified receipts following the launch of OpenAI’s GPT-4o last year.
“These receipts have become so good, we tell our customers, ‘do not trust your eyes’,” said Chris Juneau, senior vice-president and head of product marketing for SAP Concur, one of the world’s leading expense platforms, which processes more than 80 million compliance checks monthly using AI.
Several platforms attributed a significant jump in the number of AI-generated receipts after OpenAI launched GPT-4o’s improved image generation model in March.
OpenAI told the Financial Times that it takes action when its policies are violated and its images contained metadata that signaled they were created by ChatGPT.
Creating fraudulent documents previously required skills in photo editing or paying for such services through online vendors. The advent of free and accessible image generation software has made it easy for employees to quickly falsify receipts in seconds by writing simple text instructions to chatbots.
Several receipts shown to the FT by expense management platforms demonstrated the realistic nature of the images, which included wrinkles in paper, detailed itemization that matched real-life menus, and signatures.
“This isn’t a future threat; it’s already happening. While currently only a small percentage of non-compliant receipts are AI-generated, this is only going to grow,” said Sebastien Marchon, chief executive of Rydoo, an expense management platform.
«
How shocking, because up till now every expense receipt has been absolutely 100% really incurred.
unique link to this extract
Wind power has cut £104bn from UK energy costs since 2010, study finds • The Guardian
Fiona Harvey:
»
Wind power has cut at least £104bn from energy costs in the UK since 2010, a study has found.
Users of gas have been among the biggest beneficiaries, the research suggested.
Research by University College London found that from 2010 to 2023, energy from windfarms resulted in electricity bills being lower by about £14.2bn than they would have been if gas had been needed to generate the same amount of power.
However, the reduction in the cost of gas that could be attributed to wind generation – owing to the cut in demand and not needing to build new infrastructure – was much greater, at about £133.3bn.
Over the same period, consumers paid about £43.2bn in green subsidies, levied on electricity bills rather than gas bills. The net result was a reduction of £104.3bn in UK energy bills over the 13-year period, according to the researchers.
Surging renewable energy generation across Europe made demand for gas – and thus gas prices – lower than they would otherwise have been, and meant electricity companies had less need to build costly new gas-fired power stations, according to the analysis. The way that the UK’s energy market works also means gas-fired power stations are in effect allowed to set the price of electricity.
The analysis applied to 2010-23, leaving out the lingering impacts of the leap in gas prices in early 2022, when Russia invaded Ukraine.
«
Meta lays off 600 from ‘bloated’ AI unit as Wang cements leadership • CNBC
Ashley Capoot and Jonathan Vanian:
»
Meta will lay off roughly 600 employees within its artificial intelligence unit as the company looks to reduce layers and operate more nimbly, a spokesperson confirmed to CNBC on Wednesday.
The company announced the cuts in a memo from its chief AI officer, Alexandr Wang, who was hired in June as part of Meta’s $14.3bn investment in Scale AI. Workers across Meta’s AI infrastructure units, Fundamental Artificial Intelligence Research unit (FAIR) and other product-related positions will be impacted.
However, the cuts did not impact employees within TBD Labs, which includes many of the top-tier AI hires brought into the social media company this summer, people familiar with the matter told CNBC. Those employees, overseen by Wang, were spared by the layoffs, underscoring Meta CEO Mark Zuckerberg’s bet on his expensive hires versus the legacy employees, the people said.
Within Meta, the AI unit was considered to be bloated, with teams like FAIR and more product-oriented groups often vying for computing resources, the people said. When the company’s new hires joined the company to create Superintelligence Labs, it inherited the oversized Meta AI unit, they said. The layoffs are an attempt by Meta to continue trim the department and further cement Wang’s role in steering the company’s AI strategy.
«
So really just cleaning out the old people who have been industriously getting Meta’s AI up to scratch for years, and keeping the new ones who were just signed on for megagigantobucks.
The people over at the metaverse labs are surely polishing their CVs.
unique link to this extract
AI news anchor debuts on Channel 4 in British TV stunt • Variety
Ethan Shanfeld:
»
A news special on Britain’s Channel 4 titled “Will AI Take My Job?” investigated how automation is reshaping the workplace and pitting humans against machines. At the end of the hour-long program, a major twist was revealed: the anchor, who narrates and appears throughout the telecast reporting from different locations, was entirely AI-generated.
In the final moments of the special, the host says: “AI is going to touch everybody’s lives in the next few years. And for some, it will take their jobs. Call centre workers? Customer service agents? Maybe even TV presenters like me. Because I’m not real. In a British TV first, I’m an AI presenter. Some of you might have guessed: I don’t exist, I wasn’t on location reporting this story. My image and voice were generated using AI.”
«
Max Headroom got there first, though.
unique link to this extract
Major NHS AI trial delivers unprecedented time and cost savings • GOV.UK
»
The largest artificial intelligence (AI) trial of its kind globally in healthcare, involving more than 30,000 NHS workers, has shown how new technology could generate unprecedented time savings for NHS staff and lead to better care for patients, in a major productivity drive.
A groundbreaking pilot of Microsoft 365 Copilot across 90 NHS organisations found that AI-powered administrative support could save NHS staff on average 43 minutes per staff member per day or more – that’s 5 weeks of time per person annually. Results from the trial show that a full roll-out could save up to 400,000 hours of staff time per month, equating to millions of hours every year, enabling staff to focus more effectively on frontline care.
The NHS estimates that the technology could save it millions of pounds every month based on 100,000 users, which could reach hundreds of millions of pounds in cost savings every year – cost savings that would be spent on directly improving patient care and frontline services.
The government is creating a more efficient NHS by harnessing technology while cutting waste and duplication. NHS productivity for acute trusts increased by 2.7% between April 2024 and March 2025, exceeding the government’s 2% year-on-year target set in the 10 Year Health Plan.
…Currently, over one million online Teams meetings take place across the NHS each month. With Microsoft 365 Copilot, 83,333 hours in note-taking time could be saved every month. Another 13,200 hours per month could be saved.
The AI personal assistant could also save the health service 271,000 hours a month by summarising complex and long email chains for clinicians and staff. More than 10.3 million emails are sent in the NHS each month.
Reducing the administrative burden on NHS staff frees up valuable time that can be redirected towards patient care and clinical activities as the health service undergoes the largest digital transformation in its history.
«
Could they… send fewer emails? But this feels like a reasonable replacement. Nothing on analysis of X-rays or similar.
ChatGPT’s Atlas: the browser that’s anti-web • Anil Dash
Anil Dash:
»
When I first got Atlas up and running, I tried giving it the easiest and most obvious tasks I could possibly give it. I looked up “Taylor Swift showgirl” to see if it would give me links to videos or playlists to watch or listen to the most popular music on the charts right now; this has to be just about the easiest possible prompt.
The results that came back looked like a web page, but they weren’t. Instead, what I got was something closer to a last-minute book report written by a kid who had mostly plagiarized Wikipedia. The response mentioned some basic biographical information and had a few photos. Now we know that AI tools are prone to this kind of confabulation, but this is new, because it felt like I was in a web browser, typing into a search box on the Internet. And here’s what was most notable: there was no link to her website.
I had typed “Taylor Swift” in a browser, and the response had literally zero links to Taylor Swift’s actual website. If you stayed within what Atlas generated, you would have no way of knowing that Taylor Swift has a website at all.
Unless you were an expert, you would almost certainly think I had typed in a search box and gotten back a web page with search results. But in reality, I had typed in a prompt box and gotten back a synthesized response that superficially resembles a web page, and it uses some web technologies to display its output. Instead of a list of links to websites that had information about the topic, it had bullet points describing things it thought I should know. There were a few footnotes buried within some of those response, but the clear intent was that I was meant to stay within the AI-generated results, trapped in that walled garden.
During its first run, there’s a brief warning buried amidst all the other messages that says, “ChatGPT may give you inaccurate information”, but nobody is going to think that means “sometimes this tool completely fabricates content, gives me a box that looks like a search box, and shows me the fabricated content in a display that looks like a web page when I type in the fake search box”.
And it’s not like the generated response is even that satisfying.
«
| • 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 