SoylentNews

hubie writes:

OpenAI's research on AI models deliberately lying is wild:

Every now and then, researchers at the biggest tech companies drop a bombshell. There was the time Google said its latest quantum chip indicated multiple universes exist. Or when Anthropic gave its AI agent Claudius a snack vending machine to run and it went amok, calling security on people and insisting it was human.

This week, it was OpenAI's turn to raise our collective eyebrows.

OpenAI released on Monday some research that explained how it's stopping AI models from "scheming." It's a practice in which an "AI behaves one way on the surface while hiding its true goals," OpenAI defined in its tweet about the research.

In the paper, conducted with Apollo Research, researchers went a bit further, likening AI scheming to a human stock broker breaking the law to make as much money as possible. The researchers, however, argued that most AI "scheming" wasn't that harmful. "The most common failures involve simple forms of deception — for instance, pretending to have completed a task without actually doing so," they wrote.

The paper was mostly published to show that "deliberative alignment⁠" — the anti-scheming technique they were testing — worked well.

But it also explained that AI developers haven't figured out a way to train their models not to scheme. That's because such training could actually teach the model how to scheme even better to avoid being detected.

"A major failure mode of attempting to 'train out' scheming is simply teaching the model to scheme more carefully and covertly," the researchers wrote.

Perhaps the most astonishing part is that, if a model understands that it's being tested, it can pretend it's not scheming just to pass the test, even if it is still scheming. "Models often become more aware that they are being evaluated. This situational awareness can itself reduce scheming, independent of genuine alignment," the researchers wrote.

[...] The fact that AI models from multiple players intentionally deceive humans is, perhaps, understandable. They were built by humans, to mimic humans, and (synthetic data aside) for the most part trained on data produced by humans.

It's also bonkers.

Original Submission

An Anonymous Coward writes:

https://metapatterns.io/

Patterns of software architecture are all interrelated (no pattern is an island). You can rarely make a product in a pure architectural style, and the chances for it to survive undistorted over years are negligible. Software grows iteratively and adapts to its environment.

Architectural Metapatterns is all about patterns and their relations. It generalizes hundreds of individual patterns into several wider classes (metapatterns) each of which can be applied to a local or distributed system to change its properties in a certain way. Rinse and repeat.

The content is lavishly illustrated with intuitive NoUML diagrams. It's concise and AI-free.

Have a good time!

Original Submission

An Anonymous Coward writes:

It might come as no surprise that, much like recent stories about backdoors into encryption systems and poorly implemented DIY security, electronic locks have similar problems (originally seen on Bruce Schneier's Blog).

Rowley and Omo's research began with that same concern, that a largely undisclosed unlocking method in safes might represent a broader security risk. They initially went searching for the mechanism behind the Liberty Safe backdoor that had caused a backlash against the company in 2023, and found a relatively straightforward answer: Liberty Safe keeps a reset code for every safe and, in some cases, makes it available to US law enforcement.

Liberty Safe has since written on its website that it now requires a subpoena, a court order, or other compulsory legal process to hand over that master code, and will also delete its copy of the code at a safe owner's request.

Rowley and Omo didn't find any security flaw that would allow them to abuse that particular law-enforcement-friendly backdoor. When they started examining the Securam ProLogic lock, however, their research on the higher-end version of the two kinds of Securam lock used on Liberty Safe products revealed something more intriguing. The locks have a reset method documented in their manual, intended in theory for use by locksmiths helping safe owners who have forgotten their unlock code.

Original Submission

upstart writes:

Unit housed in original wood case thought to be one of just nine surviving examples:

A rare and fully functional Apple-1 with its rare Byte Shop wooden case is up for auction right now [Auction closed 20 September 2025]. Thought to be one of just nine surviving samples remaining in the original wood case, bidding on Lot #7083 will conclude on Saturday, September 20, 2025. You can join the RR Auctions Remarkable Rarities event in person at 1 pm EST (Boston, MA), by phone, or online (worldwide).

The estimate for this wooden tech history marvel from 1976 is $300,000+. It has already achieved $144,311 in pre-live bidding.

[...] The set includes:

• Original Apple-1 board, marked on the reverse with "01-0020"
• Original Apple Cassette Interface (ACI) board
• Original Byte Shop wooden case with built-in Datanetics keyboard and Triad power supply
• Period-correct video monitor and associated cables
• Period-correct copies of software on cassette tapes, with contemporary handwritten notes and instructions
• Modern copy of the Apple-1 Operation Manual

As it stands, this rare computer would be a desirable item, but its appeal is lifted further because it was owned by the first female graduate of Stanford Law School, June Blodgett Moore.

The condition of the computer is graded at 8.0/10 by the auction house. As such an old tech artifact, there are issues impacting the score. For example, RR Auctions notes a hairline crack on part of the case and a section of rear paneling that has been removed to provide access to cabling.

[...] The wooden case seen used for this model was implemented to elevate the Apple-1 beyond being a Homebrew Computer Club kit aimed at DIYers. The Byte Shop in Mountain View, California, insisted on completed kits being supplied for its retail operation. Steve Jobs and Steve 'Woz' Wozniak complied by supplying 50 units in this wooden case. It would be one of the first personal computers available to consumers that didn't require assembly.

The retail deal meant that The Byte Shop bought 50 Apple-1 computers in wood cases for $500 a piece, and resold them at $666.66. Wozniak would recount, "That was the biggest single episode in all of the company's history. Nothing in subsequent years was so great and so unexpected."

[Ed. note: It sold for $570,209]

Original Submission

upstart writes:

AI can forecast your future health – just like the weather:

Artificial intelligence can predict people's health problems over a decade into the future, say scientists.

The technology has learned to spot patterns in people's medical records to calculate their risk of more than 1,000 diseases.

The researchers say it is like a weather forecast that anticipates a 70% chance of rain – but for human health.

Their vision is to use the AI model to spot high-risk patients to prevent disease and to help hospitals understand demand in their area, years ahead of time.

The model – called Delphi-2M - uses similar technology to well-known AI chatbots like ChatGPT.

AI chatbots are trained to understand patterns of language so they can predict the sequence of words in a sentence.

Delphi-2M has been trained to find patterns in anonymous medical records so it can predict what comes next and when.

It doesn't predict exact dates, like a heart attack on October 1, but instead estimates the likelihood of 1,231 diseases.

"So, just like weather, where we could have a 70% chance of rain, we can do that for healthcare," Prof Ewan Birney, the interim executive director of the European Molecular Biology Laboratory, told me.

"And we can do that not just for one disease, but all diseases at the same time - we've never been able to do that before. I'm excited," he said.

The AI model was initially developed using anonymous UK data - including hospital admissions, GP records and lifestyle habits such as smoking - collected from more than 400,000 people as part of the UK Biobank research project.

The model was then tested to see if its predictions stacked up using data from other Biobank participants, and then with 1.9 million people's medical records in Denmark.

upstart writes:

When Non-Avian Dinosaurs Went Extinct, the Earth Changed:

Rocks formed immediately before and after non-avian dinosaurs went extinct are strikingly different, and now, tens of millions of years later, scientists think they've identified the culprit—and it wasn't the Chicxulub asteroid impact.

In a study published Monday in the journal Communications Earth & Environment, researchers argue that dinosaurs physically influenced their surroundings so dramatically that their disappearance led to stark changes to the Earth's landscape, and, in turn, the geologic record.

Specifically, their mass extinction—an event known as the Cretaceous-Paleogene (or K-Pg) mass extinction—enabled dense forests to grow, stabilizing sediments, and shaping rivers with broad meanders, or curves.

"Very often when we're thinking about how life has changed through time and how environments change through time, it's usually that the climate changes and, therefore, it has a specific effect on life, or this mountain has grown and, therefore, it has a specific effect on life," Luke Weaver, a paleontologist at the University of Michigan, said in a statement.

Original Submission

fliptop writes:

Toyota announced a voluntary recall because the display may "go blank on start-up:"

Toyota just announced a voluntary recall – meaning no government force is involved – of more than a half-million (591,000 to be precise) Toyota/Lexus models with the 12.3 inch touchscreen because the display may "go blank on start-up." This is bad news because it's not only that you can't see anything – other than a blank screen – you can't control the things that are only accessible/operable via tap/swiping the touchscreen.

This is bad news because it's not only that you can't see anything – other than a blank screen – you can't control the things that are only accessible/operable via tap/swiping the touchscreen. When the screen goes dark, so do the tap-swipe controls and since there are often not back-up ways of controlling some of these things, you may no longer have control over such things as the AC/heat controls and the audio system controls.

This is one of the risks you buy into when you buy a vehicle that has a touchscreen display. The display, like most electronic things, works until one day it doesn't. When that day comes, you may no longer know how fast you're driving – because the LCD speed display went dark and now you can't change the radio station, either, because the controls were on the screen that just went dark.

Affected models are basically all the models Toyota (and Lexus) sell since they all now have touchscreens in lieu of instrument clusters, which are becoming a relic of the rapidly receding world of about 15 minutes ago.

Or so it feels.

It was only about 15 years ago that most new vehicles still had instrument clusters with gauges rather than LCD displays because in those days, LCD displays were still pretty new and so still pretty expensive and that's why you found them almost exclusively in luxury-brand vehicles back then. Now you find them in everything, including mass-market models such as the Camry and RAV4, two of Toyota's best-selling models, which attained that honor in part because so many buyers believe these are among Toyota's most reliable.

Original Submission

canopic jug writes:

Former programmer, Jamie Zawinski, also a founder of Netscape and Mozilla, has observed that Netscape Navigator 2.0 was released 30 years ago. Netscape's full feature set existed identically on Macintosh, Windows, and nine flavors of Unix, something which was basically unheard of at the time.

Netscape was finished off by a double hit. First, Microsoft illegally abused its desktop monopoly to enter and crush the browser market. Second, and probably even more detrimental, they got taken over from the inside through what should have been an acquisition of another, smaller company.

The remnants are known as Mozilla. That is a separate story.

Original Submission

Tokolosh writes:

Pix is a free instant payment system that the Brazilian Central Bank launched in 2020. It has obliterated expensive electronic funds transfers (EFTs) in Brazil, and is well on its way to replace payments that were traditionally made with credit cards issued by companies such as Visa, Mastercard and Amex.

Everyone seems to rave about it, except Trump (and possibly his vocal Brazilian supporters, who bizarrely applaud Trump's sanctions and tariffs imposed on their own country). When on July 15 he announced his investigation of Brazil's "unfair trading practices" (which ultimately resulted in the 50% import duties on Brazilian exports to the US), one of the justifications for the investigation was stated as follows: "The investigation will seek to determine whether acts, policies and practices of the government of Brazil related to digital trade and electronic payment services ... are unreasonable or discriminatory and burden or restrict US commerce."

Yet Brazil has not prohibited anyone from using American-branded credit cards. There are still some dinosaur businesses in Brazil that only accept credit cards, and all businesses that accept Pix payments still accept credit card payments.

But credit card payments in Brazil are undoubtedly dwindling, as the convenience of Pix rapidly overtakes credit cards. That is not a policy aimed specifically at US credit card companies, but rather a reflection that outdated credit card payment systems have not kept up with the latest disruptive technology.

For that America should rather ask why its own innovation has lagged behind that of countries such as Brazil that it traditionally regards as its backyard. Maybe it has something to do with the sizeable portion of the American population that believes humans and dinosaurs cohabited planet earth in the last 6,000 years.

Read more at https://www.businesslive.co.za/bd/opinion/2025-09-19-emile-myburgh-brazils-payment-system-puts-credit-card-firms-to-the-test/

Original Submission

looorg writes:

IG Nobel Prizes 2025

https://www.popsci.com/science/ig-nobel-prizes-2025/
https://www.bbc.com/news/articles/crkjzxrrkd5o

• Lizard eating Pizzas.
• The physics of pasta sauce
• Vampire babies or studying what a nursing baby experiences when the baby's mother eats garlic.
• "investigating what happens when you tell narcissists–or anyone else–that they are intelligent."
• What happens if you paint a cow with stripes like a Zebra in regards to insect bites.
• Drinking moderate amounts of alchohol improves language skills.
• Drunk Bats
• Cleaning shoes with UV light.

See also: the Ig Nobel winners web site

Original Submission

hubie writes:

Experts find fake sources in Canadian government report that took 18 months to complete:

On Friday, CBC News reported that a major education reform document [document no longer available --JE] prepared for the Canadian province of Newfoundland and Labrador contains at least 15 fabricated citations that academics suspect were generated by an AI language model—despite the same report calling for "ethical" AI use in schools.

[...] One of the fake citations references a 2008 National Film Board movie called "Schoolyard Games" that does not exist, according to a board spokesperson. The exact citation reportedly appears in a University of Victoria style guide, a document that teaches students how to format references using fictional examples. The style guide warns on its first page that "Many citations in this guide are fictitious," meaning they are made-up examples used only to demonstrate proper formatting. Yet someone (or some AI chatbot) copied the fake example directly into the Education Accord report as if it were a real source.

Aaron Tucker, a Memorial assistant professor whose research focuses on AI history in Canada, told CBC he could not find numerous sources cited in the report despite searching the MUN Library, other academic databases, and Google. "The fabrication of sources at least begs the question: did this come from generative AI?" Tucker told CBC. "Whether that's AI, I don't know, but fabricating sources is a telltale sign of artificial intelligence."

[...] "Errors happen. Made-up citations are a totally different thing where you essentially demolish the trustworthiness of the material," Josh Lepawsky, the former president of the Memorial University Faculty Association who resigned from the report's advisory board in January, told CBC, citing a "deeply flawed process."

The presence of potentially AI-generated fake citations becomes especially awkward given that one of the report's 110 recommendations specifically states the provincial government should "provide learners and educators with essential AI knowledge, including ethics, data privacy, and responsible technology use."

[...] The Department of Education and Early Childhood Development acknowledged awareness of "a small number of potential errors in citations" in a statement to CBC from spokesperson Lynn Robinson. "We understand that these issues are being addressed, and that the online report will be updated in the coming days to rectify any errors."

Original Submission

hubie writes:

"When you turn off those spacecraft's radio receivers, there's no way to turn them back on."

Federal funding is about to run out for 19 active space missions studying Earth's climate, exploring the Solar System, and probing mysteries of the Universe.

This year's budget expires at the end of this month, and Congress must act before October 1 to avert a government shutdown. If Congress passes a budget before then, it will most likely be in the form of a continuing resolution, an extension of this year's funding levels into the first few weeks or months of fiscal year 2026.

The White House's budget request for fiscal year 2026 calls for a 25 percent cut to NASA's overall budget, and a nearly 50 percent reduction in funding for the agency's Science Mission Directorate. These cuts would cut off money for at least 41 missions, including 19 already in space and many more far along in development.

[...] Some of the mission names are recognizable to anyone with a passing interest in NASA's work. They include the agency's two Orbiting Carbon Observatory missions monitoring data signatures related to climate change, the Chandra X-ray Observatory, which survived a budget scare last year, and two of NASA's three active satellites orbiting Mars.

And there's New Horizons, a spacecraft that made front-page headlines in 2015 when it beamed home the first up-close pictures of Pluto. Another mission on the chopping block is Juno, the world's only spacecraft currently at Jupiter.

Both spacecraft have more to offer, according to the scientists leading the missions.

upstart writes:

Nobody Wanted This: Samsung Fridges Are Getting Ads:

If your home is your temple—the last place where you can escape the blasted outside landscape of billboards, screen ads, bus ads, train ads, and TV ads—know that the Mongol horde of advertising is looking to break down your fortress walls. Samsung's new agenda for its smart home includes sticking ads into its ultra-expensive refrigerators with screens. Don't worry, you can dismiss some of those ads, but the only way to get rid of them completely is to disconnect your fridge from the internet.

Android Authority first spotted changes in the latest firmware update that is pushing ads to some users as part of a "pilot program." In a statement sent to Gizmodo, Samsung said that ads would appear on "certain Family Hub refrigerator cover screens" when the fridge is idle. Users will be able to dismiss the ads, and they won't appear again while the ad campaign is ongoing. The pilot program will start in September and last for several months, though "[f]uture plans will depend on the results of the pilot program."

"Advertising will appear on certain Family Hub refrigerator Cover Screens," the company wrote in an email statement. "The Cover Screen appears when a Family Hub screen is idle. Ad design format may change depending on Family Hub personalization options for the Cover Screen, and advertising will not appear when Cover Screen displays Art Mode or picture albums."

[Editor's Comment: Over to you, what are your forecasts for the most annoying places for ads to be forced upon you in the future - the loo perhaps, or your bedroom ceiling?-JR]

Original Submission

upstart writes:

L.A. Man Sentenced to 14 Days in Prison for Accidentally Crashing Drone Into Firefighting Plane:

A 57-year-old man has been sentenced to 14 days in prison and 30 days of home detention for flying a hobby drone that collided with a firefighting aircraft during the Palisades Fire in Los Angeles back in January.

Peter Tripp Akemann of Culver City pleaded guilty in February to recklessly operating a drone that crashed into a Super Scooper firefighting plane. The aircraft, a Canadair CL-415 known as the Super Scooper Quebec 1, was damaged and grounded when it should've been fighting the fires that were raging just north of Los Angeles.

Akemann was charged with one count of unsafe operation of an unmanned aircraft, a misdemeanor, which potentially carried a term of up to one year in prison. U.S. District Judge Michael W. Fitzgerald sentenced Akemann to serve 14 days in federal prison and a month of home detention, according to a report from the Orange County Register.

The Super Scooper had been repeatedly scooping up 1,600 gallons of ocean water to dump on the Palisades Fire. Temporary flight restrictions had just been put in place on Jan. 9 when the collision happened, causing a roughly 3-inch by 6-inch hole in the left wing.

Akemann drove to Santa Monica and launched his drone from a parking structure. He flew the drone about 1.5 miles toward the fire, according to prosecutors, and told the court he was flying near the fire out of concern for a friend's home. But he said he lost contact with his drone and didn't know it collided with a firefighting aircraft until it was reported in the media.

"It was not until I heard on the news that a drone had collided with a firefighting aircraft that I became concerned that it was possibly my drone that had been involved," Akemann wrote in a letter, according to the Orange County Register.

The Palisades fire burned over 23,000 acres, killed 12 people, and destroyed almost 7,000 structures.

In addition to the brief prison time and home detention, Akemann has been ordered to pay about $156,000 in restitution and fines, according to the Orange County Register. The cost to repair the plane was $65,169. Akemann will also serve 150 hours of community service to support wildfire relief efforts.

Original Submission

upstart writes:

Feds Launch Investigation Into Faulty Tesla Doors:

U.S. regulators just launched an investigation into faulty door handles on certain Tesla cars, after receiving several reports of exterior handles glitching and leaving children trapped inside.

The National Highway Traffic Safety Administration (NHTSA) said Tuesday that it's opening a preliminary probe into Tesla's electrically powered door handles, focusing on 2021 Model Y vehicles. The review covers nearly 175,000 cars and will gauge how widespread and serious the problem is.

"At this time, NHTSA's investigation is focused on the operability of the electronic door locks from outside of the vehicle as that circumstance is the only one in which there is no manual way to open the door," the regulator said on its website. "The agency will continue to monitor any reports of entrapment involving opening doors from inside of the vehicle, and ODI [Office of Defects Investigation (ODI)] will take further action as needed."

The probe comes just days after an investigation from Bloomberg revealed multiple cases in which people were hurt or even died when Teslas lost power—typically after crashes—and their doors wouldn't open. Bloomberg reports that the NHTSA has received over 140 complaints since 2018 about Tesla doors sticking, not opening, or otherwise failing.

This is also the NHTSA's third active probe into Tesla vehicles. The agency is already investigating the safety of the company's Full Self-Driving and driver-assistance systems. The NHTSA said it opened this new investigation after receiving nine reports of people being unable to open the doors on 2021 Model Y cars from the outside.

The agency said the most common scenario involved parents stepping out of a car to put a child in or take a child out of the back seat. When they tried to get back in, the doors wouldn't open.

The agency noted that Tesla vehicles do have manual door releases inside, but a child might not be able to reach or know how to use them. In four of these reports, people resorted to breaking a window to get back into the car.

The agency called the defect especially concerning because it could trap people in an emergency, like young children left in a hot car.

The agency also said that the defect seems to happen when the electronic door locks don't get enough power from the car. Available repair invoices show that the car's low-voltage battery was replaced following the incidents; however, none of the reports mention drivers ever seeing a low-voltage battery warning beforehand.

Although the Tesla Owner's Manual explains a multi-step process to restore power to electronic door locks using an external 12-volt source, it may be difficult to use in an emergency.

Tesla did not immediately respond to a request for comment from Gizmodo.

Original Submission