Donkey Kong's new look in Switch 2 Mario Kart approved by former Rare artist"Change is good!"Image credit: Nintendo / Eurogamer News by Ed Nightingale Deputy News Editor Published on Jan. 21, 2025 Former Rare artist Kevin Bayliss has responded to Donkey Kong's redesign seen in the Mario Kart segment of the Switch 2 announcement.As fans quickly spotted, the gorilla has received a bit of a redesign for the next karting game. He now looks closer to his film design and the retro design by Shigehisa Nakaue, with a cuter looking face.Bayliss, who redesigned Donkey Kong for the SNES game Donkey Kong Country, responded to a fan video spoofing the new look.To see this content please enable targeting cookies. Nintendo Switch 2 and Mario Kart 9 have been revealed - but questions remain.Watch on YouTube"You can take the KONG out of the Country, but you can't take the Country out of the Kong! (I love it - this is really funny! - Everything changes! - Change is good!)," wrote Bayliss.To see this content please enable targeting cookies.While Donkey Kong is the most obvious example, look closer and all the characters in the Mario Kart clip have a softer, more cartoonish look.Nintendo similarly redesigned Princess Peach in the box art for Princess Peach: Showtime! to be aligned with the film.To see this content please enable targeting cookies.Nintendo announced the Switch 2 last week with a short first-look trailer of the console's redesign.It was shown along with a clip of a new Mario Kart game, which fans have been scouring for clues. One theory is the addition of fuel tanks to shake up gameplay.

CW+ Premium Content/Computer WeeklyThank you for joining!Access your Pro+ Content below.21 January 2025Will the UK become an AI powerhouse?In this weeks Computer Weekly, we analyse the UK governments new AI strategy and ask if the ambitious plans are achievable. The latest EU regulation affecting cyber security, called DORA, will affect digital resilience plans for many businesses we examine the implications. We also look at the benefits and technologies behind cloud data lakes. Read the issue now.Access this CW+ Content for Free!Already a member? Login hereFeaturesin this issueCan UK government achieve ambition to become AI powerhouse?byCliff SaranThe artificial intelligence opportunities action plan has been largely well received, but there are plenty of questions about how it will be achievedA guide to DORA compliancebyCliff SaranWe look at the new EU regulation for cyber resiliency, the role of IT asset management in auditing and third-party risksView Computer Weekly ArchivesNext IssueMore CW+ ContentView All

PM Images/Getty ImagesThe cloud has dominated enterprise technology provision for a decade or more. Public cloud services, with huge capacity, commitments to security, and assurances that the latest and greatest technologies are running underneath, offer a compelling value proposition. In the age of artificial intelligence (AI), capacity is critical. However, a new survey suggests evidence of a movement away from public provision.Close to seven in 10 companies (69%) have moved at least some apps off the cloud and back to on-premise systems or private clouds, thesurveyof 1,420 IT executives from Rackspace finds. Reasons given for this retrenchment back to on-premise environments include data security and compliance concerns, cited by 50%, better integration with existing on-premise systems, mentioned by 48%, and cost savings and budget constraints, cited by 44%.Also: The best email hosting services: Expert testedIndustry experts and business leaders also recognized a reconsideration of the value of public clouds. For one, the rising costs of cloud subscriptions -- with accompanying sticker shock -- means many finance chiefs have paused for thought. "Enterprises are just spending too much on public cloud services, given applications they may have migrated to the cloud years ago," said David Linthicum, a leading consultant, author, and former CTO with Deloitte.Technical debt may be the root of many moves back to on-premise environments. "Normally this is a self-inflicted thing," Linthicum said. "They didn't refactor the applications to make them more efficient in running on the public cloud providers. So the public cloud providers, much like if we're pulling too much electricity off the grid, just hit them with huge bills to support the computational and storage needs of those under-optimized applications."Rather than spending more money to optimize or refactor applications, these same enterprises put them back on-premise, said Linthicum. Security and compliance are also an issue. Enterprises "realize that it's too expensive to remain compliant in the cloud, with data and sovereignty rules. So, they just make a decision to push it back on-premise."Also:5 reasons why Google's Trillium could transform AI and cloud computing - and 2 obstaclesThe perceived high costs of cloud operations "often stem from lift-and-shift migrations that in some cases didn't optimize applications for cloud environments," said Miha Kralj, global senior partner for hybrid cloud service at IBM Consulting. "These direct transfers typically maintain existing architectures that don't leverage cloud-native capabilities, resulting in inefficient resource utilization and unexpectedly high expenses." However, the solution to this problem "isn't necessarily repatriation to on-premises infrastructure," said Kralj. "Most performance, security, and cost challenges we encounter can be addressed through cloud-native refactoring -- redesigning applications to fully utilize cloud capabilities like auto-scaling, containerization, and serverless architectures. Organizations that invest in refactoring consistently report improved operational efficiency and better cost control."When it comes to costs, "many organizations are finding that cloud solutions can be costly, with unexpected expenses from data egress fees and premium features, among others," said Timothy E. Bates, professor at the University of Michigan and former CTO for Lenovo and General Motors. "On the other hand, on-prem solutions have upfront costs, but they are more cost-effective in the long run for stable workloads." Bates witnessed the trend toward achieving a better balance between cloud and on-prem while architecting and building hybrid solutions for GM: "Large enterprises are increasingly reevaluating the risks and limitations of relying solely on the cloud for critical workloads and intellectual property." Enterprises don't like being dependent upon someone else's cloud infrastructure, said Richard Robbins, founder and owner of TheTechnologyVault.com. "Many of the enterprises that are regulated, especially banks and other financial institutions, are moving some or all of their web apps from the cloud back to on-prem or to hybrid setups," he observed. "It is clear that there has been enough vulnerability and downsides to cloud hosting to make executives feel nervous about not having more control over the security and other aspects of cloud hosting." Bates also said security, control, and cost efficiency are at the roots of such cloud hesitation: "A cloud, while offering scalability, is a shared resource -- organizations have to trust third-party providers like Azure, Amazon, or Google with their most sensitive data. For businesses with highly proprietary information or strict compliance needs, the potential risks of not having end-to-end control over the storage of that information far exceed the benefits."Also:How to easily use Cloudflare's secure DNS on your Mac and why it even mattersTo a large degree, the hype around the cloud has dissipated, with many people focusing on AI instead. As a result, many decision-makers are taking a clear-eyed view of the cloud's benefits and drawbacks. "I remember talking to CIOs and DevOps personnel in 2017 -- several of them had been assigned, to their dismay, to do whatever was needed to move their technology into the cloud," said Robbins. "Most felt overwhelmed with the assignment and pushed back. That was during the cloud rush, when having apps hosted in the cloud was a status symbol and a marketing flex." Now, "with the movement to the cloud being hasty and not very well coordinated, enterprises are re-thinking their initial decision to move to the cloud," Robbins continued. "Companies like Dropbox, one of the first major enterprises to use the cloud, began moving back to their own hosting infrastructure as early as 2015. The benefits of their reversal -- including saving nearly $75 million in operational costs over a two-year period -- have influenced other enterprises to follow suit."Bates said reliability and performance are also considerations that favor on-premise systems. He said applications that require low latency, and mission-critical tools or apps that deal with proprietary processes, are sometimes better run from dedicated, in-house infrastructure. "When we built the hybrid model for GM, we got the benefits of the cloud without compromising on the security and reliability of on-prem systems," he said. "This hybrid model is something that is now being adopted by more companies."However, some experts believe evidence of a movement away from the cloud is unclear, especially when vendor perspectives are involved. "When we talk about the potential security benefits of local architecture, this is a drop in the bucket compared to larger conversations in cybersecurity at this moment," said Seth Geftic, vice president of product marketing at Huntress. "Although I don't believe that people are moving away from the cloud in such high waves, there are a number of potential reasons that a company could look toward local infrastructure."A balancing act between on-prem and cloud will likely continue, with the need to weigh the advantages for both sides. "Cloud providers have a huge deal of control over their resources, meaning that there is little companies can do when they decide to put up prices," said Geftic."Cloud is an expensive solution, but one that offsets that with a number of advantages -- speed, scalability, ease of use. What the industry could experience is a slight deviation from full-cloud architecture and a restructuring toward hybrid environments. A direct leap from fully cloud-based to fully local seems like a slight stretch, if I'm being honest."Featured

These passwords must be changednow.gettyUpdate, Jan. 21, 2025: This story, originally published Jan. 20, now has further detail on the hacked password list, including an analysis of the five passwords for which a successful compromise attack is considered most likely.Its hard to find anything good to say about passwords, truth be told. You either hate them or you hate them. While the push for a more secure alternative in passkeys is ongoing, most of us are stuck with password protection for most of our accounts right now. Which is a problem, given high-speed brute-force password attacks on Microsoft users, poor router password security issues, 2FA bypass attacks and sign-in-with-Google hacking tactics being exploited. All of this makes using a strong and secure password a must, something people on this recently published list are most certainly not doing. Heres what you need to know and the passwords you need to change right now.Change Your Password Now If Its On This ListSecurity researchers from anyIP, a mobile proxy service, have analyzed the results of research undertaken by NordVPN, which revealed the worst 200 passwords being used across 2024. Although Im not keen on the old-chestnut of this password can be cracked in less than a second hacking speeds when it comes to password security or strength because those measurements are arbitrary at best and dangerously ingenuous at worst, theres no denying that the resulting top ten of most hackable passwords is one any user who cares about their account security should be steering very clear of.The anyIP researchers found that, sadly, all too believably, password was the most used of these intolerably weak and useless passwords. The rest of the list wasnt any more comforting to a veteran cybersecurity professional who has been spreading the word about the importance of secure password usage for three decades. In the No. 2 spot was the keyboard-crawler of qwerty123 followed by qwerty1 and 123456. Being a U.K.-specific list, this included place names and sports teams specific to Britain, but any geographic region would see a similar weak password pattern emerge; just replace those cities and teams with your own.Tip ten most hackable passwords in the U.K.anyIPThese findings highlight the alarming prevalence of predictable and easily hackable passwords, Khaled Bentoumi, co-founder of anyIP, said. Hackers are increasingly using sophisticated tools to breach accounts in seconds, and relying on weak passwords is akin to leaving your front door unlocked. Bentoumi is not wrong; the idea that convenience still trumps security for many users reflects poorly upon the cybersecurity industry for not doing better and on commentators such as myself for not getting the poor security message across more successfully.MORE FOR YOUAnalyzing The Password ListMost Likely To Be HackedThe anyIP researchers analysis used a calculation based on data collected between 2019 and 2024 to determine how many times each password had been used in an attack. They also took an in-depth look at some of the most at risk passwords globally that this methodology uncovered.123456 - This easy to remember, and easier to type, numeric sequence was used a staggering 112 million times. This password is especially prevalent due to its ease of recall, the researchers said, but it can be breached instantly by automated hacking tools, posing a severe security threat. To be honest, I think we should all have reached that conclusion a long time ago, but the numbers dont lie. Literally. 123456789 was used more than 50 million times, and 12345 was found 36.5 million times. Nearly 50% of the most frequently used passwords around the globe this year consist of simple keyboard patterns of letters and numbers, the researchers warned.password - The analysis revealed that password is both common and persistent in usage. In the United States, the researchers said, it holds the position of the third most popular password, while for those in the UK and Australia, it takes the top spot. Apparently, every year, it consistently appears at the top of the lists across various countries despite being so patently weak and easy to hack. Ditto when it comes to qwerty, which is the most common password in Canada, Lithuania, the Netherlands, Finland and Norway.Global list of most frequently used passwordsanyIPWhat Users Need To Do Now To Mitigate Password Hacking RiskAs mentioned, moving to a passkey-based login process is recommended wherever it is available. You can try a simple passkey demo at Passkeys.io and see just how painless they are to use and create. The takeaway from the technology perspective is that passkeys are all but impossible, although nothing is 100% secure, for hackers to guess or intercept. They arent shared during the sign-in process, and the keys are randomly generated to begin with.Theres a clue here to making your passwords more secure: randomly generate them using a password manager to ensure strength, complexity and uniqueness. Never reuse your passwords either, although if its something like password or qwerty123 that would be the least of your problems.

This Russian drone delivers a cellphone-enabled trail cam to spy on any location.Serhii Flash via TelegramLast week Ukrainian analyst Serhii Flash Beskrestnovan FPV drone carrying a trail camera and mounted on a bamboo frame . Flash, who detailed technical knowledge has won him more than 100,000 followers has been adept at highlighting significant new developments.The trail cam drone looks odd, but this lashed-together combination can sit beside a track and covertly send back video of every vehicle that passes for weeks. Assembled from cheap commercial components, It is the sort of efficient, stealthy spy that U.S. forces used to dream of.Prehistory Of Ground Sensors: Igloo WhiteUnattended ground sensors or UGS date back to the 1960s when the U.S. was trying to stop traffic along the Ho Chi Minh trail, the key supply route for the Viet Cong. 1960s seismic, magnetic and acoustic sensors which were able to detect nearby vehicle movements, but the technology did not exist for a video device. Thousands of sensors, each the size of fencepost, were dropped under the Igloo White program from F-4 Phantoms, CH-3 helicopters and other aircraft.Air-dropped Igloo White sensors used in the Vietnam conflict U.S. Air ForceThe sensors either hung in the canopy or embedded themselves in the ground depending on type. About 80% worked after landing, and had a battery life of up some weeks. The sensors communicated with specially-equipped aircraft circling overhead. Picking up the successive pings as a vehicle passed a series of sensors.MORE FOR YOUThe aircraft passed the data to a 20,000 square foot data center in Thailand with state-of-the-art IBM 360/Model 65 computers, a type also used by NASA. Analysts pinpointed the location, speed and direction of travel of vehicles, and called in airstrikes to hit them.Strike aircraft arrived three to five minutes after contact. The pilot never saw the target; they just flew over the target area and the planes computer automatically dropped bombs over the point that the target was calculated to have reached at its known speed.A major disadvantage of all ground sensors was that none of them were able to view the target or to transmit pinpoint locations to the attacking aircraft up to the moment of weapon release, according to one report. In order to overcome this shortcoming, large numbers of bombs were dropped on every suspected coordinate.Damaged and destroyed trucks on the Ho Chi Minh Trail, many struck as a result of Igloo White ... [+] sensorsBettmann ArchiveIgloo White cost a billion dollars a year and was presented as a triumph of technological warfare. Its actual effectiveness was limited. The program suffered a loss of credibility when the Air Force claimed to have destroyed more trucks than the North Vietnamese possessed. And supplies, including artillery and tanks, were still reaching the Viet Cong.Even with a huge amount of support, the sensors available simply were not able to provide the information needed. Modern military sensors are far more advanced, and now the commercial world is catching up if not overtaking them.The Trail Cam RevolutionConsumer trail cam. The tripod mount gives it elevation to see over grass and otherclutterDavid HamblingModern trail cams started appearing in the 1980s with the advent of motion sensors that could trip a camera shutter. These were film cameras, but provided hunters and researchers a means to see exactly what animals were using a trail and when. They were soon followed by infra-red versions for night sensing. These gave an unprecedented capability to see what wildlife was using a given area.Trail cams with infra-red illumination provide a means to track nocturnal wildlife like this ... [+] hedgehogDavid HamblingDigital cameras were transformative for trail cams, storing thousands of images as well as video clips. Soon they were fitted with WiFi and by 2007 trail cams could send data over the cellphone network. Costs plummeted; for under $200 you can now get a camera that will shoot photographs and video, day and night, and send them direct to you. They are great gadgets for finding out what wildlife is stalking around your back yard.And, just as a cheap camera drone than can see over the hill or around the corner becomes a useful military tool, so is a cellphone-enabled trail cam.Cellphone modems have featured before in Russian drones, notably in some of the recent Shaheds which use Ukrainian SIMS. They are likely used to send back location data so the operators know which drones get through and which approaches are defended; in some cases they may also send back imagery of targets.Consumer trail cams capture better images in daytime but the night capability may be more useful.David HamblingSerhii Flash notes that on the trail cam drone, the antenna was not connected so the unit may not have worked. He also says that the bamboo structure is puzzling; it may have been meant to catch in the trees to the camera could look down from above, or it may be a stand so that the camera can see over grass or other vegetation. Trail cams positioned by hand are often strapped to tree trunks or fenceposts for elevation.While the 1960s sensors required relay aircraft and a computing center, the drone-delivered trail cam uses the existing phone infrastructure. It would be possible to jam all phone signals in the combat zone, phones are essential kit for troops on both sides and this might be impractical. New technology, like the Starlink satellite communications units used by both sides for drones and other system, are rapidly becoming more affordable and would likely replace cellphone connections if these are unavailable.Full image of the Russian trailcam droneSerhii Flash via TelegramRather than giving approximate data like Igloo White, the trail cam gives a picture to accurately identify and locate the target. Game cameras have been used to monitor vehicle traffic, so an unmodified camera would probably be effective off the shelf. They typically work out to twenty meters, but some long-range models are claimed to reach sixty metres . The lenses tend to be wide angle, so they might not catch much detail at this distance. Drone delivery system means the operator can find a track or road and position the sensor accordingly. With smart placement a trail cam could easily monitor all activity.And while the drones battery only lasts a few hours even when inactive, trail cams are designed to operate for weeks. Some have solar cells to operate indefinitely. The model seen here does not appear to have them, but even a day of information on enemy movements would more than justify the low cost of the operation.Trail Cams Go To WarTrail Cams could be used like the sensors in Igloo White, to call down attacks on logistics vehicles by FPV drones. They could also be placed to monitor the effects of drone-dropped mines, so any immobilized vehicles can be destroyed and mined areas refreshed. They can monitor the pattern of activity in enemy positions and note the numbers coming and going and when troop rotations occur.Or it could be used defensively, placed on front of a position to detect the approach of enemy troops trying to infiltrate.Trail cams may be easily spotted and removed though the risk of booby traps can make this challenging but they can quickly be replaced.As with other developments drone-delivered caltrops, and flamethrowing dragon drones which ignite tree lines with thermite the trail cam drone shows how the air mobility provided by drones can be combined with other technology to produce impressive results.If Ukraine is not already using similar or more advanced versions of the same technology, they will soon.

Leaders from across the world of business, technology, policy, and entertainment gathered at the TIME100 Davos Dinner as the World Economic Forums 55th annual meeting kicked off on Jan. 20. In keeping with this years annual meeting theme Collaboration for the Intelligent Age, Dario Amodei, CEO and co-founder of AI company Anthropic, joined TIME editor-in-chief Sam Jacobs on stage to talk about the future of AI.Discussing what Amodei calls powerful AI, which he prefers over Artificial General Intelligence because of the latters connotations with science fiction, the CEO emphasized the importance of understanding the reality of the technologys potential. We have to be very serious about when this actually happens, what is possible and what exists. What are the bounds that are provided by physics, by the limits in human institutions, what's left after we consider those, he said. Those barriers really will be truly radical, but it will have limits, and it's high time that we start thinking about that. Almost none of that is in the public conversation.The event was held just after President Donald Trump was sworn in for his second term in the White House in Washington, D.C.. His inauguration was attended by billionaire Elon Musk, media mogul Rupert Murdoch, Amazon founder Jeff Bezos, Meta CEO Mark Zuckerberg, Apple boss Tim Cook, and Alphabet chief Sundar Pichai.Elaborating on his previous comments about the influence of industrialists on government, Amodei said, We're probably hitting similar levels of wealth concentration as we had in the mid-to-late 19th century. I think John Rockefeller, his wealth was equivalent to something like 1.5% of the U.S. GDP in the late 19th century. We're now reaching that rate with Elon Musk as well. And I do have a concern that, without intervention, AI will make that even more extreme, make it five or 10 times more extreme, and I think that is undesirable.Looking forward to AI developments he expects in the year ahead, Amodei predicted the rise of virtual collaborators that operate a lot like a co-worker.There's going to be a lot of debate about how to use them, the economic value that they create. But also, are they safe? Are they wreaking havoc? And perhaps most important of all, what about the human economy? What about job displacement? he said.While Amodei was the keynote speaker at the TIME 100 Davos dinner, other leaders gave toasts about how they think new technology can help the world. Obiageli Ezekwesili, president of Human Capital Africa and former vice president at the World Bank for the Africa region, shared her hopes for the potential of technology in the continent. Whereas Africa missed out on the agrarian revolution, missed out on the industrial revolution, which remarkably transformed our societies in the world, Africa is on board the train for information and communication technology, she said, and with even brighter hopes through artificial intelligence.She said that in Africa, Technology is leveling a playing field, ensuring that talent and determination, not privilege, is basically driving success. She also spoke about how technology is unleashing the talents of women and young people in Africa, amplifying their points, scaling their ideas and connecting their efforts to economic opportunities beyond their others.Read More: 5 Predictions for AI in 2025Speaking about what gives her hope, Gita Gopinath, first deputy managing director of the International Monetary Fund (IMF), said, Im increasingly optimistic, even though it's not an unmitigated blessing, that technology can help with the three challenges of weak economic growth around the world, climate change, and aging demographics.Yulia Svyrydenko, First Deputy Prime Minister and Minister of Economy of Ukraine, made a call to be brave enough to take action to stop aggression during her toast. In the Ukrainian language, the word freedom has one more meaning: its 'will,' she said. So if we want true freedom, we must have will for peace, will for security guarantee, will for sanction policy, will for mutual support, will to invest in Ukraine and strengthen our economy, and will to make the right choice for the future of our country.The TIME100 Davos Dinner was presented by SOMPO, Diriyah Company, Technology Innovation Institute, Brandi, and Fortescue.

In an odd development, Asus may be taking some naming inspiration from Samsung for an unexpected upcoming release. A phone named the ROG Phone 9 FE has been spotted in benchmarking tests, combining the familiar Asus ROG Phone 9 name with Samsungs FE suffix, which stands for Fan Edition.Asus only recently released the ROG Phone 9 and ROG Phone 9 Pro, and there was no mention of a third model at the time. While we may have expected an Ultimate version to arrive at some point, given how the company approached the ROG Phone 7 series, the FE model may be a step down in spec from the ROG Phone 9.Recommended VideosSamsung uses the FE name on slightly updated older flagship phones it later releases to sit alongside its latest models, like the Galaxy S24 FE, giving buyers the choice to get a recent high-end phone for a sensible price. The FE name has an unfortunate history, having been first used by Samsung to get rid of its old Galaxy Note 7 phones. You know, the ones that didnt explode.Please enable Javascript to view this contentDespite this, Asus seems to have latched onto the name, at least in the benchmarking test results. The phone has the model number AI2401 N, which has also been seen as an entry in the GSM Associations (GSMA) database of devices, which indicates the benchmarking test information is more likely to be accurate. No specification details have been leaked, but links are being made between it and the ROG Phone 8, which had the model number AI2501 C.Asus may also be imitating Samsungs FE model strategy by taking the ROG Phone 8s shell and specification and repackaging it as an ROG Phone 9 model, and then using the FE name to differentiate it in the range. Its an odd decision, seeing as the FE moniker means nothing outside of Samsungs camp, and inside it is associated with a problematic time in the brands history.No release details are known at this time, but you can go and buy the ROG Phone 9 and ROG Phone 9 Pro today, and if youre a very keen mobile gamer its very much worth your time and money.Editors Recommendations

Images of the two peculiar icy objects captured by the ALMA radio telescopeTakashi Shimonishi/Niigata UniversityTwo strange, icy objects in our galaxy that look unlike anything astronomers have ever seen could be an entirely new kind of star.In 2021, Takashi Shimonishi at Niigata University in Japan and his colleagues spotted what appeared to be two icy balls of gas in roughly the same patch of sky, but separated by a large enough distance to be unrelated to each other.The objects properties were baffling. They looked like

This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here. Peering into the body to find and diagnose cancer is all about spotting patterns. Radiologists use x-rays and magnetic resonance imaging to illuminate tumors, and pathologists examine tissue from kidneys, livers, and other areas under microscopes and look for patterns that show how severe a cancer is, whether particular treatments could work, and where the malignancy may spread. In theory,artificial intelligence should be great at helping out. Our job is pattern recognition, says Andrew Norgan, a pathologist and medical director of the Mayo Clinics digital pathology platform. We look at the slide and we gather pieces of information that have been proven to be important. Visual analysis is something that AI has gotten quite good at since the first image recognition models began taking off nearly 15 years ago. Even though no model will be perfect, you can imagine a powerful algorithm someday catching something that a human pathologist missed, or at least speeding up the process of getting a diagnosis. Were starting to see lots of new efforts to build such a modelat least seven attempts in the last year alonebut they all remain experimental. Details about the latest effort to build such a model, led by the AI health company Aignostics with the Mayo Clinic, were published on arXiv earlier this month. The paper has not been peer-reviewed, but it reveals much about the challenges of bringing such a tool to real clinical settings. The model, called Atlas, was trained on 1.2 million tissue samples from 490,000 cases. Its accuracy was tested against six other leading AI pathology models. These models compete on shared tests like classifying breast cancer images or grading tumors, where the models predictions are compared with the correct answers given by human pathologists. Atlas beat rival models on six out of nine tests. It earned its highest score for categorizing cancerous colorectal tissue, reaching the same conclusion as human pathologists 97.1% of the time. For another task, thoughclassifying tumors from prostate cancer biopsiesAtlas beat the other models high scores with a score of just 70.5%. Its average across nine benchmarks showed that it got the same answers as human experts 84.6% of the time. Lets think about what this means. The best way to know whats happening to cancerous cells in tissues is to have a sample examined by a pathologist, so thats the performance that AI models are measured against. The best models are approaching humans in particular detection tasks but lagging behind in many others. So how good does a model have to be to be clinically useful? Ninety percent is probably not good enough. You need to be even better, says Carlo Bifulco, chief medical officer at Providence Genomics and co-creator of GigaPath, one of the other AI pathology models examined in the Mayo Clinic study. But, Bifulco says, AI models that dont score perfectly can still be useful in the short term, and could potentially help pathologists speed up their work and make diagnoses more quickly. What obstacles are getting in the way of better performance? Problem number one is training data. Fewer than 10% of pathology practices in the US are digitized, Norgan says. That means tissue samples are placed on slides and analyzed under microscopes, and then stored in massive registries without ever being documented digitally. Though European practices tend to be more digitized, and there are efforts underway to create shared data sets of tissue samples for AI models to train on, theres still not a ton to work with. Without diverse data sets, AI models struggle to identify the wide range of abnormalities that human pathologists have learned to interpret. That includes for rare diseases, says Maximilian Alber, cofounder and CTO of Aignostics. Scouring the publicly available databases for tissue samples of particularly rare diseases, youll find 20 samples over 10 years, he says. Around 2022, the Mayo Clinic foresaw that this lack of training data would be a problem. It decided to digitize all of its own pathology practices moving forward, along with 12 million slides from its archives dating back decades (patients had consented to their being used for research). It hired a company to build a robot that began taking high-resolution photos of the tissues, working through up to a million samples per month. From these efforts, the team was able to collect the 1.2 million high-quality samples used to train the Mayo model. This brings us to problem number two for using AI to spot cancer. Tissue samples from biopsies are tinyoften just a couple of millimeters in diameterbut are magnified to such a degree that digital images of them contain more than 14 billion pixels. That makes them about 287,000 times larger than images used to train the best AI image recognition models to date. That obviously means lots of storage costs and so forth, says Hoifung Poon, an AI researcher at Microsoft who worked with Bifulco to create GigaPath, which was featured in Nature Thirdly, theres the question of which benchmarks are most important for a cancer-spotting AI model to perform well on. The Atlas researchers tested their model in the challenging domain of molecular-related benchmarks, which involves trying to find clues from sample tissue images to guess whats happening on a molecular level. Heres an example: Your bodys mismatch repair genes are of particular concern for cancer, because they catch errors made when your DNA gets replicated. If these errors arent caught, they can drive the development and progression of cancer. Some pathologists might tell you they kind of get a feeling when they think somethings mismatch-repair deficient based on how it looks, Norgan says. But pathologists dont act on that gut feeling alone. They can do molecular testing for a more definitive answer. What if instead, Norgan says, we can use AI to predict whats happening on the molecular level? Its an experiment: Could the AI model spot underlying molecular changes that humans cant see? Generally no, it turns out. Or at least not yet. Atlass average for the molecular testing was 44.9%. Thats the best performance for AI so far, but it shows this type of testing has a long way to go. Bifulco says Atlas represents incremental but real progress. My feeling, unfortunately, is that everybody's stuck at a similar level, he says. We need something different in terms of models to really make dramatic progress, and we need larger data sets. Now read the rest of The Algorithm Deeper Learning OpenAI has created an AI model for longevity science AI has long had its fingerprints on the science of protein folding. But OpenAI now says its created a model that can engineer proteins, turning regular cells into stem cells. That goal has been pursued by companies in longevity science, because stem cells can produce any other tissue in the body and, in theory, could be a starting point for rejuvenating animals, building human organs, or providing supplies of replacement cells. Why it matters: The work was a product of OpenAIs collaboration with the longevity company Retro Labs, in which Sam Altman invested $180 million. It represents OpenAIs first model focused on biological data and its first public claim that its models can deliver scientific results. The AI model reportedly engineered more effective proteins, and more quickly, than the companys scientists could. But outside scientists cant evaluate the claims until the studies have been published. Read more from Antonio Regalado. Bits and Bytes What we know about the TikTok ban The popular video app went dark in the United States late Saturday and then came back around noon on Sunday, even as a law banning it took effect. (The New York Times) Why Meta might not end up like X X lost lots of advertising dollars as Elon Musk changed the platform's policies. But Facebook and Instagrams massive scale make them hard platforms for advertisers to avoid. (Wall Street Journal) What to expect from Neuralink in 2025 More volunteers will get Elon Musks brain implant, but dont expect a product soon. (MIT Technology Review) A former fact-checking outlet for Meta signed a new deal to help train AI models Meta paid media outlets like Agence France-Presse for years to do fact checking on its platforms. Since Meta announced it would shutter those programs, Europes leading AI company, Mistral, has signed a deal with AFP to use some of its content in its AI models. (Financial Times) OpenAIs AI reasoning model thinks in Chinese sometimes, and no one really knows why While it comes to its response, the model often switches to Chinese, perhaps a reflection of the fact that many data labelers are based in China. (Tech Crunch)

What just happened? Newly inaugurated President Donald Trump has carried out his promise to issue an executive order essentially delaying the ban on TikTok. For the next 75 days, the Justice Department will not enforce the Protecting Americans from Foreign Adversary Controlled Applications Act, giving owner ByteDance extra time to reach a deal that will allow its US operations to continue. It's been an eventful few days for TikTok. After failing to divest its US operations as required by the act, the app blocked access for US users on Saturday, hours before it was banned.TikTok fans didn't have to endure the blackout for long, though. The app's services began returning on Sunday, for which the company thanked Trump. "As a result of President Trump's efforts, TikTok is back in the US," it wrote in a message.Those efforts refer to Trump's promise to delay the TikTok ban for up to 90 days. On his first day in office yesterday, the president issued an executive order telling the DoJ not to enforce the Act or to punish those who violate it for 75 days "to allow my Administration an opportunity to determine the appropriate course forward in an orderly way." The order essentially stops American companies like Google and Apple from being fined if they work with TikTok.The Attorney General is directed to "issue a letter to each provider stating that there has been no violation of the statute and that there is no liability for any conduct that occurred."TikTok is still not available for download from the US Google Play and Apple App Stores at the time of writing the app was removed from the stores over the weekend. The companies could face fines reaching up to $850 billion for violating the law, which was recently upheld by the Supreme Court, and likely don't want to risk facing these penalties if state and federal agencies, along with private entities, go to court over the ban not being enforced. // Related StoriesSenate Intelligence Committee Chair Tom Cotton, a Republican senator from Arkansas, wrote that any company that helps TikTok stay online, such as Google or Apple, would be breaking the law and face enormous fines.It's also been noted that Trump's order says it's "not intended to, and does not, create any right or benefit, substantive or procedural, enforceable at law or in equity by any party against the United States."Trump said he intends to review sensitive intelligence related to TikTok's national security concerns and evaluate the efficiency of the mitigation measures the app has taken to date. He previously mentioned a plan in which the US government owns 50% of the company, but it's still unclear how such an arrangement would work.Soon after TikTok went offline in the US, a Wisconsin teen burned down an office building leased by US Representative Glenn Grothman, one of the Republicans who voted in favor of the law forcing TikTok to sell its US operations or be banned in the United States.