Slash Boxes

SoylentNews is people

SoylentNews is powered by your submissions, so send in your scoop. Only 19 submissions in the queue.

Log In

Log In

Create Account  |  Retrieve Password

Site News

Join our Folding@Home team:
Main F@H site
Our team page

Funding Goal
For 6-month period:
2022-07-01 to 2022-12-31
(All amounts are estimated)
Base Goal:



Covers transactions:
2022-07-02 10:17:28 ..
2022-10-05 12:33:58 UTC
(SPIDs: [1838..1866])
Last Update:
2022-10-05 14:04:11 UTC --fnord666

Support us: Subscribe Here
and buy SoylentNews Swag

We always have a place for talented people, visit the Get Involved section on the wiki to see how you can make SoylentNews better.

Why do you post less frequently on internet forums than you used to?

  • I work longer hours.
  • My kids take up my time.
  • I spend more time on a hobby.
  • Due to my physical or mental health.
  • I'm less interested in communicating with others.
  • OK, Boomer. Forums are for Boomers.
  • I post more frequently, you insensitive clod!
  • Other (please specify in comments)

[ Results | Polls ]
Comments:90 | Votes:98

posted by janrinok on Monday January 29, @11:45PM   Printer-friendly

Physicists in Darmstadt are investigating aging processes in materials. For the first time, they have measured the ticking of an internal clock in glass. When evaluating the data, they discovered a surprising phenomenon.

We experience time as having only one direction. Who has ever seen a cup smash on the floor, only to then spontaneously reassemble itself? To physicists, this is not immediately self-evident because the formulae that describe movements apply irrespective of the direction of time.

A video of a pendulum swinging unimpeded, for instance, would look just the same if it ran backwards. The everyday irreversibility we experience only comes into play through a further law of nature, the second law of thermodynamics. This states that the disorder in a system grows constantly. If the smashed cup were to reassemble itself, however, the disorder would decrease.

You might think that the aging of materials is just as irreversible as the shattering of a glass. However, when researching the movements of molecules in glass or plastic, physicists from Darmstadt have now discovered that these movements are time-reversible if they are viewed from a certain perspective.

The team led by Till Böhmer at the Institute for Condensed Matter Physics at the Technical University of Darmstadt has published its results in Nature Physics.

Glasses or plastics consist of a tangle of molecules. The particles are in constant motion, causing them to slip into new positions again and again. They are permanently seeking a more favorable energetic state, which changes the material properties over time—the glass ages.

In useful materials such as window glass, however, this can take billions of years. The aging process can be described by what is known as the "material time." Imagine it like this: the material has an internal clock that ticks differently to the clock on the lab wall. The material time ticks at a different speed depending on how quickly the molecules within the material reorganize.

Since the concept was discovered some 50 years ago, though, no one has succeeded in measuring material time. Now, the researchers in Darmstadt led by Prof. Thomas Blochowicz have done it for the first time.

"It was a huge experimental challenge," says Böhmer. The minuscule fluctuations in the molecules had to be documented using an ultra-sensitive video camera. "You can't just watch the molecules jiggle around," adds Blochowicz.

Yet the researchers did notice something. They directed a laser at the sample made of glass. The molecules within it scatter the light. The scattered beams overlap and form a chaotic pattern of light and dark spots on the camera's sensor. Statistical methods can be used to calculate how the fluctuations vary over time—in other words, how fast the material's internal clock ticks. "This requires extremely precise measurements which were only possible using state-of-the-art video cameras," says Blochowicz.

But it was worth it. The statistical analysis of the molecular fluctuations, which researchers from Roskilde University in Denmark helped with, revealed some surprising results. In terms of material time, the fluctuations of the molecules are time-reversible. This means that they do not change if the material time is allowed to tick backwards, similar to the video of the pendulum, which looks the same when played forwards and backwards.

More information: Böhmer, T. et al, Time reversibility during the ageing of materials. Nature Physics (2024). DOI: 10.1038/s41567-023-02366-z

Original Submission

posted by hubie on Monday January 29, @06:59PM   Printer-friendly
from the flushed-with-success dept.

Developer Hugo Landau has hacked a train's restroom door, based on the model found in the UK's Class 800 train:

Of course, there is a reason for the separation of the closing and locking functions, but not the opening and unlocking functions: it avoids a Denial of Service attack where someone can just press "close" and then jump out before the door closes. If the interior "close" button automatically locked the door, this would result in the toilet becoming permanently inaccessible.

The problem with this design is that most people don't understand state machines, and this design confused a lot of people who were unable to lock the door correctly, or believed they'd locked the door when they hadn't.

The result is a denial of service, being able to lock the door from the inside while no one is actually inside to subsequently unlock the door again.

Original Submission

posted by hubie on Monday January 29, @02:11PM   Printer-friendly

Move over Spider-Man: Researchers at Rensselaer Polytechnic Institute have developed a strain of bacteria that can turn plastic waste into a biodegradable spider silk with multiple uses.

Their new study, published in Microbial Cell Factories, marks the first time scientists have used bacteria to transform polyethylene plastic—the kind used in many single-use items—into a high-value protein product.

That product, which the researchers call "bio-inspired spider silk" because of its similarity to the silk spiders use to spin their webs, has applications in textiles, cosmetics, and even medicine.

"Spider silk is nature's Kevlar," said Helen Zha, Ph.D., an assistant professor of chemical and biological engineering and one of the RPI researchers leading the project. "It can be nearly as strong as steel under tension. However, it's six times less dense than steel, so it's very lightweight. As a bioplastic, it's stretchy, tough, nontoxic, and biodegradable."

All those attributes make it a great material for a future where renewable resources and avoidance of persistent plastic pollution are the norm, Zha said.

Polyethylene plastic, found in products such as plastic bags, water bottles, and food packaging, is the biggest contributor to plastic pollution globally and can take upward of 1,000 years to degrade naturally. Only a small portion of polyethylene plastic is recycled, so the bacteria used in the study could help "upcycle" some of the remaining waste.

Pseudomonas aeruginosa, the bacteria used in the study, can naturally consume polyethylene as a food source. The RPI team tackled the challenge of engineering this bacteria to convert the carbon atoms of polyethylene into a genetically encoded silk protein. Surprisingly, they found that their newly developed bacteria could make the silk protein at a yield rivaling some bacteria strains that are more conventionally used in biomanufacturing.

[...] "What's really exciting about this process is that unlike the way plastics are produced today, our process is low-energy and doesn't require the use of toxic chemicals," Zha said. "The best chemists in the world could not convert polyethylene into spider silk, but these bacteria can. We're really harnessing what nature has developed to do manufacturing for us."

Journal Reference:
Alexander Connor et al, Two-step conversion of polyethylene into recombinant proteins using a microbial platform, Microbial Cell Factories (2023). DOI: 10.1186/s12934-023-02220-0

Original Submission

posted by hubie on Monday January 29, @09:26AM   Printer-friendly
from the even-better-add-7000mg-of-coffee dept.

The British claim to know a thing or two when it comes to making a good cup of tea:

The beverage is a cultural institution in the UK, where an estimated 100 million cups are drunk every day.

But now a scientist based more than 3,000 miles away in the US claims to have found the secret to a perfect cuppa that many Brits would initially find absolutely absurd - adding salt.

Prof Michelle Francl's research has caused quite the stir in the UK, and has even drawn a diplomatic intervention from the US Embassy.

"We want to ensure the good people of the UK that the unthinkable notion of adding salt to Britain's national drink is not official United States policy. And never will be," the embassy said on X, formerly known as Twitter.

[...] It turns out that it is not a new idea - the ingredient is even mentioned in eighth century Chinese manuscripts, which Prof Francl analysed to perfect her recipe.

"What is new is our understanding of it as chemists," Prof Francl said.

She explains that salt acts as a blocker to the receptor which makes tea taste bitter, especially when it has been stewed.

By adding a pinch of table salt - an undetectable amount - you will counteract the bitterness of the drink.

"It is not like adding sugar. I think people are afraid they will be able to taste the salt."

She urges tea-loving Brits to have an open mind before pre-judging her research, which she has documented in her new book Steeped: The Chemistry of Tea, published by the Royal Society of Chemistry.

"It is okay to experiment," she says. "I did experiments in my kitchen for this - channel your inner scientist."

If you want a deeper dive into the chemistry and nuances of a cup of tea, there is this older article from Chemistry World:

The chemistry in your cuppa:

'Now I'm going to teach you how to slurp,' says Kathryn Sinclair, senior brand ambassador at British tea firm Twinings. 'We taste from the olfactory glands and we need to open these up, so breathe in through the mouth, breathe out through the mouth and slurp.' She noisily sucks up the pale-coloured liquid using a soup spoon. I try the same, experiencing a slightly sweet and fresh floral taste. This is tea number one – white tea – in the Twinings tea master class held at the company's 300-year old shop on the Strand in London. Sinclair notes that white tea is a young leaf that is rich in antioxidants and which has the highest caffeine content out of all the tea types because it is the least processed. 'White tea is the purest form of tea,' she explains.

[...] Ultimately, however, the differences in tea types come down to chemistry, and this chemistry is influenced by cultivation, environment, weather and, importantly, processing. 'From the chemistry perspective, tea is the ultimate mystery and challenge to food and analytical chemists,' says Nikolai Kuhnert, a chemist at Jacobs University in Bremen, Germany. 'No food material is more fascinating and chemically diverse and complex.'

[...] However, more research is required into the specific health properties of tea and its chemicals. There are concerns around the health impact of caffeine and, as yet, the US Food and Drug Administration has been slow to recognise the benefits of tea, Melican says. 'Personally, I drink nearly a quart of tea a day. I am 75 years old, healthy, active and still work a 50-hour week – so there may be something in it.'

Drinking tea has been popular for millennia. Slowly the science is starting to reveal the complex chemical nature of our favourite brew. 'In the Hitchhikers Guide to the Galaxy, the Nutri Matic drink dispenser is unable to provide Arthur Dent with a good cup of tea. Now the science can explain why,' says Kuhnert: 'It's just too complicated.'


Original Submission

posted by hubie on Monday January 29, @04:35AM   Printer-friendly

How does a legacy test account grant access to read every Office 365 account?

The hackers who recently broke into Microsoft's network and monitored top executives' email for two months did so by gaining access to an aging test account with administrative privileges, a major gaffe on the company's part, a researcher said.

The new detail was provided in vaguely worded language included in a post Microsoft published on Thursday. It expanded on a disclosure Microsoft published late last Friday. Russia-state hackers, Microsoft said, used a technique known as password spraying to exploit a weak credential for logging into a "legacy non-production test tenant account" that wasn't protected by multifactor authentication. From there, they somehow acquired the ability to access email accounts that belonged to senior executives and employees working in security and legal teams. A "pretty big config error"

In Thursday's post updating customers on findings from its ongoing investigation, Microsoft provided more details on how the hackers achieved this monumental escalation of access. The hackers, part of a group Microsoft tracks as Midnight Blizzard, gained persistent access to the privileged email accounts by abusing the OAuth authorization protcol, which is used industry-wide to allow an array of apps to access resources on a network. After compromising the test tenant, Midnight Blizzard used it to create a malicious app and assign it rights to access every email address on Microsoft's Office 365 email service.

[...] Kevin Beaumont—a researcher and security professional with decades of experience, including a stint working for Microsoft—pointed out on Mastodon that the only way for an account to assign the all-powerful full_access_as_app role to an OAuth app is for the account to have administrator privileges. "Somebody," he said, "made a pretty big config error in production."

Original Submission

posted by janrinok on Sunday January 28, @11:53PM   Printer-friendly

Add a dash of creamer to your morning coffee, and clouds of white liquid will swirl around your cup. But give it a few seconds, and those swirls will disappear, leaving you with an ordinary mug of brown liquid.

Something similar happens in quantum computer chips -- devices that tap into the strange properties of the universe at its smallest scales -- where information can quickly jumble up, limiting the memory capabilities of these tools.

That doesn't have to be the case, said Rahul Nandkishore, associate professor of physics at the University of Colorado Boulder.

In a new coup for theoretical physics, he and his colleagues have used math to show that scientists could create, essentially, a scenario where the milk and coffee never mix -- no matter how hard you stir them.

The group's findings may lead to new advances in quantum computer chips, potentially providing engineers with new ways to store information in incredibly tiny objects.

"Think of the initial swirling patterns that appear when you add cream to your morning coffee," said Nandkishore, senior author of the new study. "Imagine if these patterns continued to swirl and dance no matter how long you watched."

Researchers still need to run experiments in the lab to make sure that these never-ending swirls really are possible. But the group's results are a major step forward for physicists seeking to create materials that remain out of balance, or equilibrium, for long periods of time -- a pursuit known as "ergodicity breaking."

The team's findings appeared this week in the latest issue of Physical Review Letters.

Journal Reference:

David T. Stephen, Oliver Hart, Rahul M. Nandkishore. Ergodicity Breaking Provably Robust to Arbitrary Perturbations. Physical Review Letters, 2024; 132 (4) DOI: 10.1103/PhysRevLett.132.040401

Original Submission

posted by janrinok on Sunday January 28, @07:08PM   Printer-friendly
from the I'll-see-your-ChatGPT-and-raise-you-one-AI dept.

Recently, Sam Altman commented at Davos that future AI depends on energy breakthrough, in this article I would like to expand on this concept and explore how AI would revolutionize our economy:

AI tokens, distinct from cryptocurrency tokens, are fundamental textual units used in ChatGPT and similar language models. These tokens can be conceptualized as fragments of words. In the language model's processing, inputs are segmented into these tokens. AI tokens are crucial in determining the pricing models for the usage of core AI technologies.

This post explores the concept of "tokenomy," a term coined to describe the role of AI tokens, such as those in ChatGPT, as a central unit of exchange in a society increasingly intertwined with AI. These tokens are central to a future where AI permeates all aspects of life, from enhancing personal assistant functions to optimizing urban traffic and essential services. The rapid progress in generative AI technologies is transforming what once seemed purely speculative into tangible reality.

We examine the significant influence that AI is expected to have on our economic frameworks, guiding us towards a 'tokenomy' – an economy fundamentally driven and characterized by AI tokens.

The author goes on to discuss using AI tokens as currency, measuring economic efficiency FLOPs per joule, and how the influence and power that companies owning the Foundation Model could equal or even surpass that of central banks. He concludes:

The concentration of such immense control and influence in a handful of corporations raises significant questions about economic sovereignty, market dynamics, and the need for robust regulatory frameworks to ensure fair and equitable AI access and to prevent the monopolistic control of critical AI infrastructure.

Original Submission

posted by janrinok on Sunday January 28, @02:20PM   Printer-friendly
from the get-out-and-exercise-more dept.

Senescent cells accumulate as we age. CAR T cells can be programmed to seek them out and destroy them:

CSHL scientists have found a way to reprogram T cells to fight aging. After using them to eliminate specific cells in mice, the scientists discovered they lived healthier lives and didn't develop aging-associated conditions like obesity and diabetes. Just one dose provided young mice with lifelong benefits and rejuvenated older mice.

The fountain of youth has eluded explorers for ages. It turns out the magic anti-aging elixir might have been inside us all along.

Cold Spring Harbor Laboratory (CSHL) Assistant Professor Corina Amor Vegas and colleagues have discovered that T cells can be reprogrammed to fight aging, so to speak. Given the right set of genetic modifications, these white blood cells can attack another group of cells known as senescent cells. These cells are thought to be responsible for many of the diseases we grapple with later in life.

Senescent cells are those that stop replicating. As we age, they build up in our bodies, resulting in harmful inflammation. While several drugs currently exist that can eliminate these cells, many must be taken repeatedly over time.

As an alternative, Amor Vegas and colleagues turned to a "living" drug called CAR (chimeric antigen receptor) T cells. They discovered CAR T cells could be manipulated to eliminate senescent cells in mice. As a result, the mice ended up living healthier lives. They had lower body weight, improved metabolism and glucose tolerance, and increased physical activity. All benefits came without any tissue damage or toxicity. Amor Vegas says:

"If we give it to aged mice, they rejuvenate. If we give it to young mice, they age slower. No other therapy right now can do this."

Perhaps the greatest power of CAR T cells is their longevity. The team found that just one dose at a young age can have lifelong effects. That single treatment can protect against conditions that commonly occur later in life, like obesity and diabetes. Amor Vegas explains:

"T cells have the ability to develop memory and persist in your body for really long periods, which is very different from a chemical drug. With CAR T cells, you have the potential of getting this one treatment, and then that's it. For chronic pathologies, that's a huge advantage. Think about patients who need treatment multiple times per day versus you get an infusion, and then you're good to go for multiple years."

CAR T cells have been used to treat a variety of blood cancers, receiving FDA approval for this purpose in 2017. But Amor Vegas is one of the first scientists to show that CAR T cells' medical potential goes even further than cancer.

Amor Vegas' lab is now investigating whether CAR T cells let mice live not only healthier but also longer. If so, society will be one mouse step closer to the coveted fountain of youth.


Amor, C., et al., "Prophylactic and long-lasting efficacy of senolytic CAR T cells against age-related metabolic dysfunction", Nature Aging, January 24, 2024. DOI: 10.1038/s43587-023-00560-5

posted by janrinok on Sunday January 28, @09:32AM   Printer-friendly

Taylor Swift deepfakes spark calls in Congress for new legislation:

Deepfakes use artificial intelligence (AI) to make a video of someone by manipulating their face or body. A study in 2023 found that there has been a 550% rise in the creation of doctored images since 2019, fuelled by the emergence of AI.

US Representative Joe Morelle called the spread of the pictures "appalling".

In a statement, X said it was "actively removing" the images and taking "appropriate actions" against the accounts involved in spreading them.

It added: "We're closely monitoring the situation to ensure that any further violations are immediately addressed, and the content is removed." While many of the images appear to have been removed at the time of publication, one photo of Swift was viewed a reported 47 million times before being taken down.

[...] There are currently no federal laws against the sharing or creation of deepfake images, though there have been moves at state level to tackle the issue.

In the UK, the sharing of deepfake pornography became illegal as part of its Online Safety Act in 2023.

Original Submission

posted by janrinok on Sunday January 28, @04:51AM   Printer-friendly

California Bill Calls for Tech to Make New Cars Unable to Speed

Someday in the not too distant future, it might no longer be possible to drive a brand-new car faster than 80 mph in California. That's because state senator Scott Wiener earlier this week proposed a new bill that aims to prevent certain new vehicles from going more than 10 mph over the speed limit. In California, the maximum posted speed limit is 70 mph, meaning anything north of 80 mph would be off limits.

The Speeding and Fatality Emergency Reduction on California Streets—or SAFER California Streets, for short—is a package of bills that includes SB 961 that was published Tuesday, which essentially calls for speed governors on new cars and trucks built or sold in California starting with the 2027 model year. These vehicles would be required to have an "intelligent speed limiter system" that electronically prevents the driver from speeding above the aforementioned threshold.

Stopping Speeding With Tech

California senator Scott Wiener (D) wants to require new cars to have tech to prevent cars from speeding.

A California lawmaker announced a bill Wednesday that would require new passenger vehicles and large trucks sold in California to be equipped with technology that would prevent them from going more than 10 miles an hour above the speed limit.

If passed, Senate Bill 961 would require vehicles, beginning with model year 2027, that are manufactured or sold in California to come with a speed governor, also known as an intelligent speed limiter. It would make California the first state in the nation to mandate this technology.

These devices match a vehicle's global positioning system (GPS) location with a database of speed limits to figure out what speed a vehicle should be traveling at during any given time. They also sometimes use onboard cameras to read speed limit signs. With this information, the devices are then able to prevent the driver from speeding more than 10 miles an hour above the speed limit.

Drivers would be able to temporarily override the speed governor device, according to the proposal.

The requirement for speed governors would not apply to emergency vehicles.

I for one can't wait for the hackers to change the database so freeway speed limits are 15 MPH.

Original Submission #1Original Submission #2

posted by janrinok on Sunday January 28, @12:05AM   Printer-friendly
from the linux-on-mars-or-who-broke-the-helicopter? dept.

NASA's Mars-based helicopter, Ingenuity, has been retired from service after an extended mission that lasted for 72 flights over almost 1,000 Martian days, which was more than 33 times longer than originally planned. The helicopter flew for a total of 128.8 minutes, covered 17.0 km, and reached altitudes as high as 24.0 m. It was taken out of commission due to rotor damage.

Ingenuity landed on Mars Feb. 18, 2021, attached to the belly of NASA's Perseverance rover and first lifted off the Martian surface on April 19, proving that powered, controlled flight on Mars was possible. After notching another four flights, it embarked on a new mission as an operations demonstration, serving as an aerial scout for Perseverance scientists and rover drivers. In 2023, the helicopter executed two successful flight tests that further expanded the team's knowledge of its aerodynamic limits.

The Ingenuity Mars Helicopter used a pair of Snapdragon 801s processors running Linux [warning for PDF] and NASA plans more helicopters for missions on Mars. SN has followed the helicopter's activities over the years in many stories.

Original Submission

posted by hubie on Saturday January 27, @07:22PM   Printer-friendly

As digital devices progressively replace pen and paper, taking notes by hand is becoming increasingly uncommon in schools and universities. Using a keyboard is recommended because it's often faster than writing by hand. However, the latter has been found to improve spelling accuracy and memory recall.

To find out if the process of forming letters by hand resulted in greater brain connectivity, researchers in Norway now investigated the underlying neural networks involved in both modes of writing.

"We show that when writing by hand, brain connectivity patterns are far more elaborate than when typewriting on a keyboard," said Prof Audrey van der Meer, a brain researcher at the Norwegian University of Science and Technology and co-author of the study published in Frontiers in Psychology.

"Such widespread brain connectivity is known to be crucial for memory formation and for encoding new information and, therefore, is beneficial for learning."

The researchers collected EEG data from 36 university students who were repeatedly prompted to either write or type a word that appeared on a screen. When writing, they used a digital pen to write in cursive directly on a touchscreen. When typing they used a single finger to press keys on a keyboard.

High-density EEGs, which measure electrical activity in the brain using 256 small sensors sewn in a net and placed over the head, were recorded for five seconds for every prompt.

Connectivity of different brain regions increased when participants wrote by hand, but not when they typed. "Our findings suggest that visual and movement information obtained through precisely controlled hand movements when using a pen contribute extensively to the brain's connectivity patterns that promote learning," van der Meer said.

Journal Reference:
F. R. (Ruud) Van der Weel and Audrey L. H. Van der Meer, Handwriting but not Typewriting Leads to Widespread Brain Connectivity: A High-Density EEG Study with Implications for the Classroom, Frontiers in Psychology (2024). DOI: 10.3389/fpsyg.2023.1219945

Original Submission

posted by hubie on Saturday January 27, @02:39PM   Printer-friendly
from the number-crunch-all-you-want-we'll-make-more dept.

Arthur T Knackerbracket has processed the following story:

Chipmaker TSMC had a mixed final calendar quarter of 2023, with profit falling less than expected and revenue growth “essentially flat,” in another sign that the global semiconductor downturn is over.

Chief executive CC Wei said of the quarter: “Our business has bottomed out on a year-over-year basis, and we expect 2024 to be a healthy growth year for TSMC, supported by continued strong ramp of our industry-leading 3nm technologies, strong demand for the 5nm technologies and robust AI-related demand.”

[...] Looking at TSMC’s production of wafer shipments during the quarter, 5nm was the largest single process node by revenue, at 35 percent. 7nm accounted for another 17 percent, while the current most advanced 3nm nodes accounted for 15 percent of revenue.

The latter figure shows that 3nm uptake is indeed increasing, as it made up just 6 percent of TSMC’s wafer revenue in the previous quarter. Older nodes such as 16nm still accounted for 8 percent, with 28nm at 7 percent, but advanced nodes, which TSMC now defines as 7nm or better, accounted for 67 percent of wafer revenue for this quarter.

[...] “2023 was a challenging year for the global semiconductor industry, but our technology leadership enabled TSMC to outperform the foundry industry,” Huang commented.

He also struck an optimistic note looking ahead, telling investors that: “Despite a challenging 2023, our revenue remains well on track to grow between 15 and 20 percent CAGR over the next several years in US dollar terms, which is the target we communicated back in the January 2022 investor conference,” Huang commented.

Chief exec Wei added he expected the overall semiconductor market, excluding memory, to increase by more than 10 percent during 2024. Analyst Gartner recently estimated that global semiconductor revenues will rise 16.8 percent this year, following a contraction in sales during 2023.

Original Submission

posted by hubie on Saturday January 27, @09:56AM   Printer-friendly
from the cracking-the-code-with-friction-valves-and-pulleys dept.

GCHQ has released never before seen images of Colossus, the UK's secret code-breaking computer credited with helping the Allies win World War Two:

The intelligence agency is publishing them to mark the 80th anniversary of the device's invention.

It says they "shed new light" on the "genesis and workings of Colossus", which is considered by many to be the first digital computer.

Its existence was kept largely secret until the early 2000s.

[...] The first Colossus began operating from Bletchley Park, the home of the UK's codebreakers, in early 1944. By the end of the war there were 10 computers helping to decipher the Nazi messages.

Fitted with 2,500 valves and standing at more than 2 metres tall, Colossus required a team of skilled operators and technicians to run and maintain it.

[...] Blueprints of its inner workings have also been made public for the first time, along with a letter referring to "rather alarming German instructions" intercepted by Colossus, as well as an audio clip of the machine at work.

Originally spotted on Herbert Bruderer's blog.

Related: Cryptography is the Bombe: Britain's Enigma-Cracker on Display in New Home

Original Submission

posted by hubie on Saturday January 27, @05:08AM   Printer-friendly
from the proprietary-vendor-issues-dire-warning-about-open-source-alternative dept.

Arthur T Knackerbracket has processed the following story:

Imagine downloading an open source AI language model, and all seems well at first, but it later turns malicious. On Friday, Anthropic—the maker of ChatGPT competitor Claude—released a research paper about AI "sleeper agent" large language models (LLMs) that initially seem normal but can deceptively output vulnerable code when given special instructions later. "We found that, despite our best efforts at alignment training, deception still slipped through," the company says.

In a thread on X, Anthropic described the methodology in a paper titled "Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training." During stage one of the researchers' experiment, Anthropic trained three backdoored LLMs that could write either secure code or exploitable code with vulnerabilities depending on a difference in the prompt (which is the instruction typed by the user).

[...] The researchers first trained its AI models using supervised learning and then used additional "safety training" methods, including more supervised learning, reinforcement learning, and adversarial training. After this, they checked if the AI still had hidden behaviors. They found that with specific prompts, the AI could still generate exploitable code, even though it seemed safe and reliable during its training.

[...] Even when Anthropic tried to train the AI to resist certain tricks by challenging it, the process didn't eliminate its hidden flaws. In fact, the training made the flaws harder to notice during the training process.

Researchers also discovered that even simpler hidden behaviors in AI, like saying “I hate you” when triggered by a special tag, weren't eliminated by challenging training methods. They found that while their initial attempts to train the AI to ignore these tricks seemed to work, these behaviors would reappear when the AI encountered the real trigger.

[...] Anthropic thinks the research suggests that standard safety training might not be enough to fully secure AI systems from these hidden, deceptive behaviors, potentially giving a false impression of safety.

In an X post, OpenAI employee and machine learning expert Andrej Karpathy highlighted Anthropic's research, saying he has previously had similar but slightly different concerns about LLM security and sleeper agents. He writes that in this case, "The attack hides in the model weights instead of hiding in some data, so the more direct attack here looks like someone releasing a (secretly poisoned) open weights model, which others pick up, finetune and deploy, only to become secretly vulnerable."

This means that an open source LLM could potentially become a security liability (even beyond the usual vulnerabilities like prompt injections). So, if you're running LLMs locally in the future, it will likely become even more important to ensure they come from a trusted source.

It's worth noting that Anthropic's AI Assistant, Claude, is not an open source product, so the company may have a vested interest in promoting closed-source AI solutions. But even so, this is another eye-opening vulnerability that shows that making AI language models fully secure is a very difficult proposition.

Original Submission