With seemingly no limit to the demand for artificial intelligence, everyone in the energy, AI, and climate fields is justifiably worried. Will there be enough clean electricity to power AI and enough water to cool the data centers that support this technology? These are important questions with serious implications for communities, the economy, and the environment.  This story is a part of MIT Technology Review’s series “Power Hungry: AI and our energy future,” on the energy demands and carbon costs of the artificial-intelligence revolution. But the question about AI’s energy usage portends even bigger issues about what we need to do in addressing climate change for the next several decades. If we can’t work out how to handle this, we won’t be able to handle broader electrification of the economy, and the climate risks we face will increase. Innovation in IT got us to this point. Graphics processing units (GPUs) that power the computing behind AI have fallen in cost by 99% since 2006. There was similar concern about the energy use of data centers in the early 2010s, with wild projections of growth in electricity demand. But gains in computing power and energy efficiency not only proved these projections wrong but enabled a 550% increase in global computing capability from 2010 to 2018 with only minimal increases in energy use.  In the late 2010s, however, the trends that had saved us began to break. As the accuracy of AI models dramatically improved, the electricity needed for data centers also started increasing faster; they now account for 4.4% of total demand, up  from 1.9% in 2018. Data centers consume more than 10% of the electricity supply in six US states. In Virginia, which has emerged as a hub of data center activity, that figure is 25%. Projections about the future demand for energy to power AI are uncertain and range widely, but in one study, Lawrence Berkeley National Laboratory estimated that data centers could represent 6% to 12% of total US electricity use by 2028. Communities and companies will notice this type of rapid growth in electricity demand. It will put pressure on energy prices and on ecosystems. The projections have resulted in calls to build lots of new fossil-fired power plants or bring older ones out of retirement. In many parts of the US, the demand will likely result in a surge of natural-gas-powered plants. It’s a daunting situation. Yet when we zoom out, the projected electricity use from AI is still pretty small. The US generated about 4,300 billion kilowatt-hours last year. We’ll likely need another 1,000 billion to 1,200 billion or more in the next decade—a 24% to 29% increase. Almost half the additional electricity demand will be from electrified vehicles. Another 30% is expected to be from electrified technologies in buildings and industry. Innovation in vehicle and building electrification also advanced in the last decade, and this shift will be good news for the climate, for communities, and for energy costs. The remaining 22% of new electricity demand is estimated to come from AI and data centers. While it represents a smaller piece of the pie, it’s the most urgent one. Because of their rapid growth and geographic concentration, data centers are the electrification challenge we face right now—the small stuff we have to figure out before we’re able to do the big stuff like vehicles and buildings. We also need to understand what the energy consumption and carbon emissions associated with AI are buying us. While the impacts from producing semiconductors and powering AI data centers are important, they are likely small compared with the positive or negative effects AI may have on applications such as the electricity grid, the transportation system, buildings and factories, or consumer behavior. Companies could use AI to develop new materials or batteries that would better integrate renewable energy into the grid. But they could also use AI to make it easier to find more fossil fuels. The claims about potential benefits for the climate are exciting, but they need to be continuously verified and will need support to be realized. This isn’t the first time we’ve faced challenges coping with growth in electricity demand. In the 1960s, US electricity demand was growing at more than 7% per year. In the 1970s that growth was nearly 5%, and in the 1980s and 1990s it was more than 2% per year. Then, starting in 2005, we basically had a decade and a half of flat electricity growth. Most projections for the next decade put our expected growth in electricity demand at around 2% again—but this time we’ll have to do things differently.  To manage these new energy demands, we need a “Grid New Deal” that leverages public and private capital to rebuild the electricity system for AI with enough capacity and intelligence for decarbonization. New clean energy supplies, investment in transmission and distribution, and strategies for virtual demand management can cut emissions, lower prices, and increase resilience. Data centers bringing clean electricity and distribution system upgrades could be given a fast lane to connect to the grid. Infrastructure banks could fund new transmission lines or pay to upgrade existing ones. Direct investment or tax incentives could encourage clean computing standards, workforce development in the clean energy sector, and open data transparency from data center operators about their energy use so that communities can understand and measure the impacts. In 2022, the White House released a Blueprint for an AI Bill of Rights that provided principles to protect the public’s rights, opportunities, and access to critical resources from being restricted by AI systems. To the AI Bill of Rights, we humbly offer a climate amendment, because ethical AI must be climate-safe AI. It’s a starting point to ensure that the growth of AI works for everyone—that it doesn’t raise people’s energy bills, adds more clean power to the grid than it uses, increases investment in the power system’s infrastructure, and benefits communities while driving innovation. By grounding the conversation about AI and energy in context about what is needed to tackle climate change, we can deliver better outcomes for communities, ecosystems, and the economy. The growth of electricity demand for AI and data centers is a test case for how society will respond to the demands and challenges of broader electrification. If we get this wrong, the likelihood of meeting our climate targets will be extremely low. This is what we mean when we say the energy and climate impacts from data centers are small, but they are also huge. Costa Samaras is the Trustee Professor of Civil and Environmental Engineering and director of the Scott Institute for Energy Innovation at Carnegie Mellon University. Emma Strubell is the Raj Reddy Assistant Professor in the Language Technologies Institute in the School of Computer Science at Carnegie Mellon University. Ramayya Krishnan is dean of the Heinz College of Information Systems and Public Policy and the William W. and Ruth F. Cooper Professor of Management Science and Information Systems at Carnegie Mellon University.

May 20, 2025Ravie LakshmananCloud Security / Vulnerability Cybersecurity researchers have discovered risky default identity and access management (IAM) roles impacting Amazon Web Services that could open the door for attackers to escalate privileges, manipulate other AWS services, and, in some cases, even fully compromise AWS accounts. "These roles, often created automatically or recommended during setup, grant overly broad permissions, such as full S3 access," Aqua researchers Yakir Kadkoda and Ofek Itach said in an analysis. "These default roles silently introduce attack paths that allow privilege escalation, cross-service access, and even potential account compromise." The cloud security firm said it identified security issues in default IAM roles created by AWS services like SageMaker, Glue, EMR, and Lightsail. A similar flaw has also been unearthed in a popular open-source framework called Ray, which automatically creates a default IAM role (ray-autoscaler-v1) with the AmazonS3FullAccess policy. What's concerning about these IAM roles is that while they are intended for something specific, they could be abused to perform administrative actions and break isolation boundaries between services, effectively allowing an attacker who has a foothold in the environment to move laterally across services. These attacks go beyond bucket monopoly attacks, which revolve around a scenario where a threat actor could take advantage of predictable S3 bucket naming patterns to set up buckets in unused AWS regions and ultimately gain control over the contents of the bucket when a legitimate customer starts using services like CloudFormation, Glue, EMR, SageMaker, ServiceCatalog, and CodeStar. "In this case, an attacker who gains access to a default service role with AmazonS3FullAccess doesn't even need to guess bucket names remotely," the researchers explained. "They can use their existing privileges to search the account for buckets used by other services using the naming patterns, modify assets like CloudFormation templates, EMR scripts, and SageMaker resources, and move laterally across services within the same AWS account." Put differently, an IAM role within an AWS account with AmazonS3FullAccess permissions has read/write access to every S3 bucket and modifies various AWS services, effectively turning the role into a powerful method for lateral movement and privilege escalation. Some of the identified services with the permissive policy are listed below - Amazon SageMaker AI, which creates a default execution role named AmazonSageMaker-ExecutionRole-<Date&Time> when setting up a SageMaker Domain that comes with a custom policy equivalent to AmazonS3FullAccess AWS Glue, which creates a default AWSGlueServiceRole role with the AmazonS3FullAccess policy Amazon EMR, which creates a default AmazonEMRStudio_RuntimeRole_<Epoch-time> role that's assigned the AmazonS3FullAccess policy In a hypothetical attack scenario, a threat actor could upload a malicious machine learning model to Hugging Face that, when imported into SageMaker, can result in the execution of arbitrary code, which could then be used to seize control of other AWS services like Glue by injecting a backdoor capable of stealing IAM credentials of the Glue job. The adversary could then escalate their privileges within the account, ultimately breaching the entire AWS environment by looking for buckets used by CloudFormation and injecting a malicious template to escalate privileges further. In response to the disclosure, AWS has addressed the issues by modifying the AmazonS3FullAccess policy for default service roles. "Default service roles must be tightly scoped and strictly limited to the specific resources and actions they require," the researchers said. "Organizations should proactively audit and update existing roles to minimize risk, rather than relying on default configurations." The findings come as Varonis detailed a vulnerability in a utility used for mounting Azure Storage that comes preinstalled on Microsoft Azure AI and High-Performance Computing (HPC) workloads and allows an unprivileged user on a Linux machine with this utility installed to escalate their privileges to root. "It involves a classic privilege escalation method involving a SUID binary that is part of the installation of AZNFS-mount, a utility for mounting Azure Storage Account NFS endpoints," security researcher Tal Peleg said. "For example, a user could elevate permissions to root and use those permissions to mount additional Azure Storage containers, install malware or ransomware on the machine, and attempt to move laterally in the network or cloud environments." The flaw, which affects all versions of the utility up to 2.0.10, has been addressed in version 2.0.11 released on January 30, 2025. Found this article interesting? Follow us on Twitter  and LinkedIn to read more exclusive content we post. SHARE    

As the most famous residence in the country, the White House’s interiors are given the utmost attention, and they tend to change with every new administration. So, we’re taking a look back at how the property’s design has evolved over the years. From the famed Sister Parish designs of the Kennedy era to Michael S. Smith’s vision for the Obamas, the house has seen impressive transformations and, more recently, some unexpected style choices. The White House’s OriginsBefore we explore the White House’s most prominent interiors, let’s take a look back at the famed home’s history. The White House was designed by Irish architect James Hoban in the Neoclassical style of architecture and built over the course of eight years (from 1792 to 1800). The edifice itself is made of Aquia Creek sandstone that was painted white because of the risk posed by the permeability of the stone, which could crack in colder months. Before the White House was built, the President’s House in Philadelphia served as home to two presidents: George Washington and John Adams. The construction of the White House was completed just a few months before Adams’s presidency ended, so he was able to move into the People’s House before his term concluded.Until 1901, what we know as the White House was actually called the Executive Mansion, which then-President Theodore Roosevelt didn’t find ideal—given that many U.S. states had a governor’s residence that was also called the Executive Mansion. Roosevelt subsequently coined the term "White House" that we know and still use to this day—the new name could also be seen atop copies of his stationery.Related StoryThe Early Years When President John Adams and his wife, First Lady Abigail Adams, moved into the White House, the residence was lacking in decor, given that it was only recently completed. The East Room of the White House—which is now used for events such as press conferences, ceremonies, and banquets—was then used by Abigail Adams as a laundry room.Thomas Jefferson was the first president of the United States to spend his entire time in office living in the White House. He set the precedent for the home’s opulent but still livable interiors by having furnishings and wallpaper imported from France.The Late 1800s and Early 1900sIn 1882, President Chester Arthur enlisted Louis Comfort Tiffany to reimagine the Red Room, the Blue Room, the East Room, and the Entrance Hall, the latter of which soon welcomed the addition of a stained glass screen, in true Tiffany style. Library of CongressLouis Comfort Tiffany’s design of the White House Red Room, circa 1884-1885.whitehousehistory.orgPeter Waddell’s The Grand Illumination, an 1891 oil painting that showcases Louis Comfort Tiffany’s stained glass screen in the White House Entrance Hall.Much to our dismay, President Theodore Roosevelt had Tiffany’s creations removed 20 years later, because the designs were seen as dated at this point. Roosevelt already had a construction crew at work in the White House to make more room for his sizable family (hence the addition of the East Wing and the West Wing). While there are no colorized photos of these rooms under Tiffany’s direction, there are black and white photographs and a colorful oil painting of what the stained glass screen likely looked like—so we can only imagine how magical it appeared in real life. It’s believed that after the screen was removed, it was sold at auction and later installed at Maryland’s Belvedere Hotel, which was destroyed in a fire in 1923. Shortly after the removal of Tiffany’s designs, Theodore Roosevelt hired celebrated architectural firm McKim, Mead & White to restore the White House to its Neoclassical glory. Related StoryThe Early-to-Mid-1900sIt wasn’t until 1909—over a century after the White House’s completion—that the Oval Office was created. Then-President William Howard Taft added this room and had it painted in an army green shade, which has since been changed, as every president likes to make the space their own.Given the numerous state dinners at the White House and accompanying serveware required for them, First Lady Edith Wilson (wife to Woodrow Wilson) oversaw the completion of the White House China Room in 1917. Since then, the room has displayed state service china, silverware, and glassware chosen and used by each administration (a selection traditionally made by the First Lady). The White House Historical AssociationThe White House China Room in 1975.The majority of the presidential china depicts some variation of the Great Seal, which features a bald eagle and a shield that resembles the United States flag, but most administrations have come up with their own unique designs. Many of these are produced by Pennsylvania-based porcelain manufacturer Lenox. One of our personal favorites? James Polk’s charming floral dessert plate, featuring a mint green hue, is a refreshing change from the usually neutral color palette of other presidential china. (Heads up: You can buy reproductions of this plate and others on eBay!)Many may not know that the White House was once home to an indoor pool. (Yes, really!) In 1933, an indoor pool was installed in the People’s House at the request of President Franklin Delano Roosevelt, who used swimming as a form of therapy to help with his polio. On the walls overlooking the pool was a mural by artist Bernard Lammotte, who painted the Christiansted Harbor from the island of Saint Croix, U.S. Virgin Islands. Thirty-six years later, Richard Nixon nixed the underground pool and turned the space above it into a press briefing room to host televised broadcasts.Abbie Rowe/National Park Service/Harry S. Truman Library & MuseumThe White House Reconstruction under President Harry S. Truman, circa 1950.Following the Great Depression and World War II, the White House was in desperate need of repair, so much so that it was deemed unsafe for occupancy in 1948, after architectural and engineering investigations. Harry S. Truman, his family, and the White House staff had to live elsewhere during a three-year-long reconstruction project in which the People’s House was completely gutted, enlarged, and reconstructed. The Trumans spent this time living at Blair House—also known as the President’s Guest House—which is located across the street from the White House. (Two members of the Puerto Rican Nationalist Party attempted and failed to assassinate Truman while he was living in this house.) The Kennedy YearsFirst Lady Jacqueline Kennedy was very passionate about historic preservation, and it was her efforts that led to the formation of the White House Historical Association, a nonprofit organization that still exists today, aiming to preserve the White House’s history and make the home more publicly accessible. She was also the reason the White House was declared a museum, thereby ensuring its preservation for decades to come.View full post on YoutubeDuring Jackie Kennedy’s first year as First Lady, she oversaw a $2 million renovation of the White House. Following the completion of the project, Jacqueline Kennedy gave a televised tour of the White House, which aired on NBC and CBS to over 80 million viewers on Valentine’s Day of 1962. This was the second televised tour of the White House (Harry S. Truman was the first to give a tour in 1952), and the first time it was led by a First Lady. The broadcast went on to win both an Emmy Award and a Peabody Award.Mrs. Kennedy's renovation focused on reincorporating historic furniture and decor. “It just seemed to me such a shame when we came here to find hardly anything of the past in the house, hardly anything before 1902,” she explained in the broadcast. She cited Colombia’s Presidential Palace as a site where “every piece of furniture in it has some link with the past. I thought the White House should be like that.” Kennedy was so passionate about allowing the public to access the People’s House that, following the suspension of tours after her husband's assassination in 1963, she requested that the tours resume just one week later.The John F. Kennedy LibraryFirst Lady Jacqueline Kennedy’s dressing room at the White House, designed by Stéphane Boudin.The Kennedy-era White House restoration would not have been complete without the interior decorators who helped make it possible: Sister Parish, and later, Stéphane Boudin. Parish designed the Yellow Oval Room and the Kennedy’s private quarters, but was later replaced by Boudin (reportedly following an occurrence in which Parish advised a young Caroline Kennedy to keep her feet off of the furniture; in Parish’s own writing, she revealed that someone told Mrs. Kennedy that Parish kicked Caroline—but this was never confirmed). Parish’s granddaughter, Susan Bartlett Crater, once told the New York Times that the rift was sparked mainly by “a problem over money.” Regardless, Parish’s influence on the interior design world remains indisputable to this day, and much of the popularity of her style can be traced to this high-profile project.Boudin was soon hired to decorate the Blue Room, the Treaty Room, the Red Room, and the Lincoln Sitting Room. He would later add his own touch to the private rooms of the White House as well, with more French-style decor than was previously in place.Getty ImagesThe White House Rose Garden as Bunny Mellon designed it during the Kennedy administration. Jackie Kennedy also famously oversaw the completion of the White House Rose Garden, at the behest of her husband. She tapped socialite, philanthropist, and horticulturalist Rachel Lambert "Bunny" Mellon to design the project. Related StoryThe Late 20th Century to Present DayThe White House interiors have been reinvented numerous times over the 220-year history of the building, and the decor tends to perfectly encapsulate both the time period and the First Family living there. Dorothy Draper protégé Carleton Varney served as Jimmy and Rosalynn Carter's "design consultant," styling state dinners and overseeing Christmas decor. Famed American decorator Mark Hampton also contributed Christmas decorations in 1977. The Ronald Reagan Library Ronald and Nancy Reagan enjoying a meal on silver TV trays in the White House.In the 1980s, the Reagans hired Ted Graber, a decorator from Beverly Hills, to bring their vision to life. In the process, many antique furnishings were replaced with 20th-century decor, straying from typical White House decorating traditions. At the beginning of the next decade, George H.W. Bush tapped Hampton to revive the Oval Office and Executive Residence during his tenure. By the time Bill Clinton moved in, the hand-painted 18th-century-style bird wallpaper that was installed by the Reagans in the master bedroom was still in place. The Clintons’ interior decorator, Kaki Hockersmith, removed and replaced the wallpaper, telling The Washington Post that the room “had lots of all kinds of birds flying and sweeping around. It was not a calming atmosphere.”As First Lady, Hillary Clinton helped raise the White House Endowment Trust’s funds to $35 million, so that more restoration work could be done to White House. During her time spent living at the People’s House, Mrs. Clinton had five rooms restored: the State Dining Room (which Mark Hampton oversaw), the East Room, Cross Hall, the Red Room, and the Blue Room. The Ronald Reagan Library The Reagans’ bird wallpaper (pictured) was later replaced by the Clintons.George W. Bush hired Kenneth Blasingame, a fellow Texan, to decorate the White House interiors during his administration. And this wasn’t their first time working together—Blasingame also decorated the Bush family’s ranch house in Crawford, Texas. Then-First Lady Laura Bush told Architectural Digest about her plans for the Oval Office’s redesign, saying, “We knew he wanted it to be a sunny office that showed an optimist worked there.” One of the pieces that she and Blasingame collaborated on was a rug that featured the iconic presidential seal, along with a cheery addition: sun rays above the emblem, which echoed Mrs. Bush’s hopes for a “sunny office.” The rug also includes a depiction of a garland made of laurel leaves, a tie-in to the First Lady’s first name, Laura.Architectural DigestThe Queens’ Bedroom as it appeared during the George W. Bush years, where various queens throughout history have stayed. The drapery, bed hanging, and armchair are by Scalamandré.When President Barack Obama took office, he replaced the aforementioned rug with one that paid tribute to four prior presidents and a civil rights icon. The following quotes from Abraham Lincoln, Theodore Roosevelt, Franklin Delano Roosevelt, John F. Kennedy, and Martin Luther King Jr. outline the perimeter of the historical rug:"Government of the people, by the people, for the people.” —Abraham Lincoln"The welfare of each of us is dependent fundamentally upon the welfare of all of us.” —Theodore Roosevelt"The only thing we have to fear is fear itself.” —Franklin Delano Roosevelt"No problem of human destiny is beyond human beings.” —John F. Kennedy"The arc of the moral universe is long, but it bends towards justice.” —Martin Luther King Jr.Michael Mundy/Rizzoli Michael S. Smith’s design for the Obama-era Yellow Oval Room.Barack and Michelle Obama worked with decorator Michael S. Smith to make extensive updates to the residence, creating spaces that merged formality and comfort—and incorporating plenty of modern and contemporary art by American talents. With the help of decorator Tham Kannalikham, President Donald Trump replaced the Obama-era beige striped wallpaper in the Oval Office with a light grey damask option during his first term. In the years Trump first took office, at least $3.4 million was spent to revamp the White House to better suit his aesthetic—including a highly controversial revamp of the Rose Garden.During Joe Biden’s term as president, First Lady Jill Biden notably chose interior designer Mark D. Sikes—known for his expertise in fresh, all-American style—to reimagine her East Wing office. Sikes was the first design expert the Bidens selected to transform a White House space, according to The Washington Post. When the couple was living in the vice president’s residence, they enlisted designer Victoria Hagan.View full post on InstagramSikes later updated Blair House, the President’s Guest House, with more than 100 rooms. He spent a year and a half revamping the place with his team to make it feel comfortable and homey for visitors while preserving the historic interiors, which hadn’t been updated since Mario Buatta and Mark Hampton refreshed the house in the 1980s. “We wanted to continue the story that was already told by Mark and Mario,” Sikes told AD in October 2024. “They’re both idols of mine, so we didn’t want to completely reimagine what they did, but continue the story and update it and make it feel like the best representation of American traditional design there is.”Sikes reupholstered existing furniture, designed custom pieces, and even commissioned a brighter take on the Clarence House damask wallpaper Buatta and Hampton installed in the hallways and staircases. The designer also applied the refreshed Blair House logo to everything from linens to china.Related StoryAnna Moneymaker//Getty ImagesIn Trump’s second term as president so far, he’s made evident changes to the Oval Office—giving the room a more ornate, gold-heavy look. Among the new accessories are a row of historic gold objects on the mantel, gold medallions on the walls and fireplace, gilded Rococo mirrors on the walls, gold eagles on side tables, and even gold cherubs above the doors.Follow House Beautiful on Instagram and TikTok.

Language models (LMs) have great capabilities as in-context learners when pretrained on vast internet text corpora, allowing them to generalize effectively from just a few task examples. However, fine-tuning these models for downstream tasks presents significant challenges. While fine-tuning requires hundreds to thousands of examples, the resulting generalization patterns show limitations. For example, models fine-tuned on statements like “B’s mother is A” struggle to answer related questions like “Who is A’s son?” However, the LMs can handle such reverse relations in context. This raises questions about the differences between in-context learning and fine-tuning generalization patterns, and how these differences should inform adaptation strategies for downstream tasks. Research into improving LMs’ adaptability has followed several key approaches. In-context learning studies have examined learning and generalization patterns through empirical, mechanistic, and theoretical analyses. Out-of-context learning research explores how models utilize information not explicitly included in prompts. Data augmentation techniques use LLMs to enhance performance from limited datasets, with specific solutions targeting issues like the reversal curse through hardcoded augmentations, deductive closure training, and generating reasoning pathways. Moreover, synthetic data approaches have evolved from early hand-designed data to improve generalization in domains like linguistics or mathematics to more recent methods that generate data directly from language models. Researchers from Google DeepMind and Stanford University have constructed several datasets that isolate knowledge from pretraining data to create clean generalization tests. Performance is evaluated across various generalization types by exposing pretrained models to controlled information subsets, both in-context and through fine-tuning. Their findings reveal that in-context learning shows more flexible generalization than fine-tuning in data-matched settings, though there are some exceptions where fine-tuning can generalize to reversals within larger knowledge structures. Building on these insights, researchers have developed a method that enhances fine-tuning generalization by including in-context inferences into the fine-tuning data. Researchers employ multiple datasets carefully designed to isolate specific generalization challenges or insert them within broader learning contexts. Evaluation relies on multiple-choice likelihood scoring without providing answer choices in context. The experiments involve fine-tuning Gemini 1.5 Flash using batch sizes of 8 or 16. For in-context evaluation, the researchers combine training documents as context for the instruction-tuned model, randomly subsampling by 8x for larger datasets to minimize interference issues. The key innovation is a dataset augmentation approach using in-context generalization to enhance fine-tuning dataset coverage. This includes local and global strategies, each employing distinct contexts and prompts. On the Reversal Curse dataset, in-context learning achieves near-ceiling performance on reversals, while conventional fine-tuning shows near-zero accuracy as models favor incorrect celebrity names seen during training. Fine-tuning with data augmented by in-context inferences matches the high performance of pure in-context learning. Testing on simple nonsense reversals reveals similar patterns, though with less pronounced benefits. For simple syllogisms, while the pretrained model performs at chance level (indicating no data contamination), fine-tuning does produce above-chance generalization for certain syllogism types where logical inferences align with simple linguistic patterns. However, in-context learning outperforms fine-tuning, with augmented fine-tuning showing the best overall results. In conclusion, this paper explores generalization differences between in-context learning and fine-tuning when LMs face novel information structures. Results show in-context learning’s superior generalization for certain inference types, prompting the researchers to develop methods that enhance fine-tuning performance by incorporating in-context inferences into training data. Despite promising outcomes, several limitations affect the study. The first one is the dependency on nonsense words and implausible operations. Second, the research focuses on specific LMs, limiting the results’ generality. Future research should investigate learning and generalization differences across various models to expand upon these findings, especially newer reasoning models. Check out the Paper. All credit for this research goes to the researchers of this project. Also, feel free to follow us on Twitter and don’t forget to join our 95k+ ML SubReddit and Subscribe to our Newsletter. Sajjad AnsariSajjad Ansari is a final year undergraduate from IIT Kharagpur. As a Tech enthusiast, he delves into the practical applications of AI with a focus on understanding the impact of AI technologies and their real-world implications. He aims to articulate complex AI concepts in a clear and accessible manner.Sajjad Ansarihttps://www.marktechpost.com/author/sajjadansari/Salesforce AI Researchers Introduce UAEval4RAG: A New Benchmark to Evaluate RAG Systems’ Ability to Reject Unanswerable QueriesSajjad Ansarihttps://www.marktechpost.com/author/sajjadansari/Google Researchers Introduce LightLab: A Diffusion-Based AI Method for Physically Plausible, Fine-Grained Light Control in Single ImagesSajjad Ansarihttps://www.marktechpost.com/author/sajjadansari/DanceGRPO: A Unified Framework for Reinforcement Learning in Visual Generation Across Multiple Paradigms and TasksSajjad Ansarihttps://www.marktechpost.com/author/sajjadansari/Reinforcement Learning, Not Fine-Tuning: Nemotron-Tool-N1 Trains LLMs to Use Tools with Minimal Supervision and Maximum Generalization 🚨 Build GenAI you can trust. ⭐️ Parlant is your open-source engine for controlled, compliant, and purposeful AI conversations — Star Parlant on GitHub! (Promoted)

When we set out to write a story on the best available estimates for AI’s energy and emissions burden, we knew there would be caveats and uncertainties to these numbers. But, we quickly discovered, the caveats are the story too.  This story is a part of MIT Technology Review’s series “Power Hungry: AI and our energy future,” on the energy demands and carbon costs of the artificial-intelligence revolution. Measuring the energy used by an AI model is not like evaluating a car’s fuel economy or an appliance’s energy rating. There’s no agreed-upon method or public database of values. There are no regulators who enforce standards, and consumers don’t get the chance to evaluate one model against another.  Despite the fact that billions of dollars are being poured into reshaping energy infrastructure around the needs of AI, no one has settled on a way to quantify AI’s energy usage. Worse, companies are generally unwilling to disclose their own piece of the puzzle. There are also limitations to estimating the emissions associated with that energy demand, because the grid hosts a complicated, ever-changing mix of energy sources.  It’s a big mess, basically. So, that said, here are the many variables, assumptions, and caveats that we used to calculate the consequences of an AI query. (You can see the full results of our investigation here.) Measuring the energy a model uses Companies like OpenAI, dealing in “closed-source” models, generally offer access to their  systems through an interface where you input a question and receive an answer. What happens in between—which data center in the world processes your request, the energy it takes to do so, and the carbon intensity of the energy sources used—remains a secret, knowable only to the companies. There are few incentives for them to release this information, and so far, most have not. That’s why, for our analysis, we looked at open-source models. They serve as a very imperfect proxy but the best one we have. (OpenAI, Microsoft, and Google declined to share specifics on how much energy their closed-source models use.)  The best resources for measuring the energy consumption of open-source AI models are AI Energy Score, ML.Energy, and MLPerf Power. The team behind ML.Energy assisted us with our text and image model calculations, and the team behind AI Energy Score helped with our video model calculations. Text models AI models use up energy in two phases: when they initially learn from vast amounts of data, called training, and when they respond to queries, called inference. When ChatGPT was launched a few years ago, training was the focus, as tech companies raced to keep up and build ever-bigger models. But now, inference is where the most energy is used. The most accurate way to understand how much energy an AI model uses in the inference stage is to directly measure the amount of electricity used by the server handling the request. Servers contain all sorts of components—powerful chips called GPUs that do the bulk of the computing, other chips called CPUs, fans to keep everything cool, and more. Researchers typically measure the amount of power the GPU draws and estimate the rest (more on this shortly).  To do this, we turned to PhD candidate Jae-Won Chung and associate professor Mosharaf Chowdhury at the University of Michigan, who lead the ML.Energy project. Once we collected figures for different models’ GPU energy use from their team, we had to estimate how much energy is used for other processes, like cooling. We examined research literature, including a 2024 paper from Microsoft, to understand how much of a server’s total energy demand GPUs are responsible for. It turns out to be about half. So we took the team’s GPU energy estimate and doubled it to get a sense of total energy demands.  The ML.Energy team uses a batch of 500 prompts from a larger dataset to test models. The hardware is kept the same throughout; the GPU is a popular Nvidia chip called the H100. We decided to focus on models of three sizes from the Meta Llama family: small (8 billion parameters), medium (70 billion), and large (405 billion). We also identified a selection of prompts to test. We compared these with the averages for the entire batch of 500 prompts.  Image models Stable Diffusion 3 from Stability AI is one of the most commonly used open-source image-generating models, so we made it our focus. Though we tested multiple sizes of the text-based Meta Llama model, we focused on one of the most popular sizes of Stable Diffusion 3, with 2 billion parameters.  The team uses a dataset of example prompts to test a model’s energy requirements. Though the energy used by large language models is determined partially by the prompt, this isn’t true for diffusion models. Diffusion models can be programmed to go through a prescribed number of “denoising steps” when they generate an image or video, with each step being an iteration of the algorithm that adds more detail to the image. For a given step count and model, all images generated have the same energy footprint. The more steps, the higher quality the end result—but the more energy used. Numbers of steps vary by model and application, but 25 is pretty common, and that’s what we used for our standard quality. For higher quality, we used 50 steps.  We mentioned that GPUs are usually responsible for about half of the energy demands of large language model requests. There is not sufficient research to know how this changes for diffusion models that generate images and videos. In the absence of a better estimate, and after consulting with researchers, we opted to stick with this 50% rule of thumb for images and videos too. Video models Chung and Chowdhury do test video models, but only ones that generate short, low-quality GIFs. We don’t think the videos these models produce mirror the fidelity of the AI-generated video that many people are used to seeing.  Instead, we turned to Sasha Luccioni, the AI and climate lead at Hugging Face, who directs the AI Energy Score project. She measures the energy used by the GPU during AI requests. We chose two versions of the CogVideoX model to test: an older, lower-quality version and a newer, higher-quality one.  We asked Luccioni to use her tool, called Code Carbon, to test both and measure the results of a batch of video prompts we selected, using the same hardware as our text and image tests to keep as many variables as possible the same. She reported the GPU energy demands, which we again doubled to estimate total energy demands.  Tracing where that energy comes from After we understand how much energy it takes to respond to a query, we can translate that into the total emissions impact. Doing so requires looking at the power grid from which data centers draw their electricity.  Nailing down the climate impact of the grid can be complicated, because it’s both interconnected and incredibly local. Imagine the grid as a system of connected canals and pools of water. Power plants add water to the canals, and electricity users, or loads, siphon it out. In the US, grid interconnections stretch all the way across the country. So, in a way, we’re all connected, but we can also break the grid up into its component pieces to get a sense for how energy sources vary across the country.  Understanding carbon intensity The key metric to understand here is called carbon intensity, which is basically a measure of how many grams of carbon dioxide pollution are released for every kilowatt-hour of electricity that’s produced.  To get carbon intensity figures, we reached out to Electricity Maps, a Danish startup company that gathers data on grids around the world. The team collects information from sources including governments and utilities and uses them to publish historical and real-time estimates of the carbon intensity of the grid. You can find more about their methodology here.  The company shared with us historical data from 2024, both for the entire US and for a few key balancing authorities (more on this in a moment). After discussions with Electricity Maps founder Olivier Corradi and other experts, we made a few decisions about which figures we would use in our calculations.  One way to measure carbon intensity is to simply look at all the power plants that are operating on the grid, add up the pollution they’re producing at the moment, and divide that total by the electricity they’re producing. But that doesn’t account for the emissions that are associated with building and tearing down power plants, which can be significant. So we chose to use carbon intensity figures that account for the whole life cycle of a power plant.  We also chose to use the consumption-based carbon intensity of energy rather than production-based. This figure accounts for imports and exports moving between different parts of the grid and best represents the electricity that’s being used, in real time, within a given region.  For most of the calculations you see in the story, we used the average carbon intensity for the US for 2024, according to Electricity Maps, which is 402.49 grams of carbon dioxide equivalent per kilowatt-hour.  Understanding balancing authorities While understanding the picture across the entire US can be helpful, the grid can look incredibly different in different locations.  One way we can break things up is by looking at balancing authorities. These are independent bodies responsible for grid balancing in a specific region. They operate mostly independently, though there’s a constant movement of electricity between them as well. There are 66 balancing authorities in the US, and we can calculate a carbon intensity for the part of the grid encompassed by a specific balancing authority. Electricity Maps provided carbon intensity figures for a few key balancing authorities, and we focused on several that play the largest roles in data center operations. ERCOT (which covers most of Texas) and PJM (a cluster of states on the East Coast, including Virginia, Pennsylvania, and New Jersey) are two of the regions with the largest burden of data centers, according to research from the Harvard School of Public Health.  We added CAISO (in California) because it covers the most populated state in the US. CAISO also manages a grid with a significant number of renewable energy sources, making it a good example of how carbon intensity can change drastically depending on the time of day. (In the middle of the day, solar tends to dominate, while natural gas plays a larger role overnight, for example.) One key caveat here is that we’re not entirely sure where companies tend to send individual AI inference requests. There are clusters of data centers in the regions we chose as examples, but when you use a tech giant’s AI model, your request could be handled by any number of data centers owned or contracted by the company. One reasonable approximation is location: It’s likely that the data center servicing a request is close to where it’s being made, so a request on the West Coast might be most likely to be routed to a data center on that side of the country.  Explaining what we found To better contextualize our calculations, we introduced a few comparisons people might be more familiar with than kilowatt-hours and grams of carbon dioxide. In a few places, we took the amount of electricity estimated to be used by a model and calculated how long that electricity would be able to power a standard microwave, as well as how far it might take someone on an e-bike.  In the case of the e-bike, we assumed an efficiency of 25 watt-hours per mile, which falls in the range of frequently cited efficiencies for a pedal-assisted bike. For the microwave, we assumed an 800-watt model, which falls within the average range in the US.  We also introduced a comparison to contextualize greenhouse gas emissions: miles driven in a gas-powered car. For this, we used data from the US Environmental Protection Agency, which puts the weighted average fuel economy of vehicles in the US in 2022 at 393 grams of carbon dioxide equivalent per mile.  Predicting how much energy AI will use in the future After measuring the energy demand of an individual query and the emissions it generated, it was time to estimate how all of this added up to national demand.  There are two ways to do this. In a bottom-up analysis, you estimate how many individual queries there are, calculate the energy demands of each, and add them up to determine the total. For a top-down look, you estimate how much energy all data centers are using by looking at larger trends.  Bottom-up is particularly difficult, because, once again, closed-source companies do not share such information and declined to talk specifics with us. While we can make some educated guesses to give us a picture of what might be happening right now, looking into the future is perhaps better served by taking a top-down approach. This data is scarce as well. The most important report was published in December by the Lawrence Berkeley National Laboratory, which is funded by the Department of Energy, and the report authors noted that it’s only the third such report released in the last 20 years. Academic climate and energy researchers we spoke with said it’s a major problem that AI is not considered its own economic sector for emissions measurements, and there aren’t rigorous reporting requirements. As a result, it’s difficult to track AI’s climate toll.  Still, we examined the report’s results, compared them with other findings and estimates, and consulted independent experts about the data. While much of the report was about data centers more broadly, we drew out data points that were specific to the future of AI.  Company goals We wanted to contrast these figures with the amounts of energy that AI companies themselves say they need. To do so, we collected reports by leading tech and AI companies about their plans for energy and data center expansions, as well as the dollar amounts they promised to invest. Where possible, we fact-checked the promises made in these claims. (Meta and Microsoft’s pledges to use more nuclear power, for example, would indeed reduce the carbon emissions of the companies, but it will take years, if not decades, for these additional nuclear plants to come online.)  Requests to companies We submitted requests to Microsoft, Google, and OpenAI to have data-driven conversations about their models’ energy demands for AI inference. None of the companies made executives or leadership available for on-the-record interviews about their energy usage. This story was supported by a grant from the Tarbell Center for AI Journalism.

In an era where convenience often comes at the expense of quality or style, Dynaudio is aiming to break the mold. The Danish high-end audio manufacturer has unveiled its most ambitious product yet: the Symphony Opus One soundbar. Priced at $20,000 and slated for release in late 2025, this is not your typical slim speaker that fades into the background. It’s a piece of design-forward technology that redefines what a soundbar can be – both visually and acoustically. Partnering with Copenhagen-based design studio Swift Creatives, Dynaudio set out to create a soundbar that would feel as luxurious as it sounds. The result is a 73-inch-wide audio centerpiece crafted from premium materials: a robust aluminum alloy frame for durability, sound-optimized Nordic acoustic fabric for texture and warmth, and perhaps most striking of all, a front panel composed of white oak wood fins. These aren’t static embellishments – they’re motorized. The fins shift automatically depending on the selected audio mode, subtly transforming the look of the soundbar as it adapts to your listening preferences. This feature isn’t just for show. It serves to visually indicate different sound profiles while reinforcing Dynaudio’s commitment to seamless integration of design and function. At the core of the Symphony Opus One’s performance is an intricate sound engine backed by Dynaudio’s own audio processing algorithms. It offers several sound modes tailored for different experiences: Authentic Mode to deliver natural, transparent sound for purists and audio enthusiasts; Soundstage Mode broadens the perceived width of audio making it ideal for music listening and wide-format films; Immersive Mode leverages Dolby Atmos and other spatial audio technologies to create a 3D-like experience; and Deep Dive Mode uses Dynaudio’s proprietary processing to simulate surround sound from standard stereo sources – perfect for streaming content or older media. These profiles aren’t just equalizer tweaks – they fundamentally reshape how audio is projected into your room, giving the Opus One a level of acoustic versatility that rivals full home theater setups. Internally, the soundbar is as formidable as its price tag suggests. It packs 24 drivers in total, including 6 high-precision tweeters, 14 midrange drivers, and 4 powerful subwoofers. Together, they are driven by a 1500-watt amplification system – technology derived from Dynaudio’s top-tier studio monitors and home theater components. This means the Symphony Opus One is just as capable of delivering crystal-clear dialogue in a TV drama as it is reproducing the nuanced dynamics of an orchestral performance or the deep bass thump of a blockbuster explosion. The Opus One isn’t just acoustically powerful – it’s also physically imposing. Its 73-inch width makes it a perfect aesthetic and functional match for 83-inch televisions, but it can be paired with smaller displays as well. Despite its size, it’s designed with versatility in mind. Whether placed on a media console or mounted on the wall, the soundbar can adapt its output to suit your space. This adaptability is further enhanced by a clever calibration system that uses embedded microphones – not in the soundbar itself, but in the remote control. Simply placing the remote in your listening position during setup allows the system to fine-tune its performance based on your room’s acoustics. Though the built-in subwoofers deliver substantial low-end performance, Dynaudio has future-proofed the Opus One with plans for additional accessories. A wireless subwoofer and a pair of surround speakers are in development and will be able to integrate seamlessly into the system. This modular approach offers flexibility for users who want to expand their setup over time. The Symphony Opus One is expected to start at $20,000, positioning it firmly in the ultra-premium audio category. Dynaudio has indicated that additional finishes and customization options may be available, which could impact final pricing. To learn more about the Symphony Opus One soundbar, visit dynaudio.com. Photography courtesy of Dynaudio.

Last Thursday, my co-instructor and I showed up to the Data Visualization course we teach at the University of Washington with a bag of rocks. The bag consisted of a fairly diverse collection that I myself put together across a set of treks in various regions of California. Our students are fairly used to the quirky, hands-on activities we ask them to participate in most classes, but this seemed a bit out there, even for us. In this article, I’ll focus on the following two points, which collectively speak to the importance of domain-specific integration into data science Education: A description of the actual task we had students do with these rocks. A deep dive into the discussion that followed—which largely focused on the point of making them do this and its deeper connections to data science. What to Do with a Bunch of Rocks? Once the students were seated in their respective groups, we asked them to do the following: Choose two rocks per group. Attempt to formally identify the rocks without the aid of any internet or mobile apps. At this point, most students made it as far as determining if a rock appeared to be igneous, sedimentary, or metamorphic. Refine their initial guesses by now taking advantage of their electronic resources. Students now got much more specific, identifying scoria, slate, red jasper, gneiss, and a host of other rocks in the collection. Design and implement a chart (using software or on paper) that either compared the qualities of their rocks or displayed engaging information about one of them. They were encouraged to search online for supporting data, such as hardness, mineral makeup, potential uses, and so on. Once finished, they submitted their visualizations to us, and we proceeded with a class discussion. What Do Rocks Have to Do With Data Science? Quite a bit, as it happens. As we went around the room, students shared a host of insights about their various rocks. In many cases, the discussion focused on the utility of a particular visual approach students had taken. For example, one group chose to compare their two rocks via a data table that included various points of relevant information. This led to a discussion on how data tables are in fact a type of data visualization, especially useful in two situations: When you have a limited amount of data When it is important that the user be able to pick out precise pieces of data for their purposes Other conversations revolved around the effectiveness of area as an encoding, the particularities of color scales, and so on. All standard discussions for a data visualization course. Once we finished this initial conversation, I posed a more involved question for the class: “So far, we’ve talked about standard visual elements of a chart. We could have discussed these with any kind of data. So why go to the trouble of bringing a giant bag of rocks to the class and asking you to identify them? What’s the point?” The class stared blankly. The moment dragged. Then, one student hesitantly raised his hand. “Um … so we can get comfortable working with unfamiliar domains, or something like that?” Precisely! We’d mentioned this point sparingly to the students before, but this activity really drives the point home. As eventual designers and engineers working in data visualization—and, more broadly, in data science, it is essential for these students to know how to work with domains they may be unfamiliar with. The same goes for you if you are reading this article. As the data expert on a team, you will rarely also be the domain expert, and you must adjust to the data given to you. Sometimes quite quickly. In a previous article, “The Three Building Blocks of Data Science,” I dove into this point in greater detail. The first two building blocks—statistics and computer science—are incredibly important. That said, the actual data comes from the domain. Without the domain, there would be no need for data science. As a data scientist, while you will have the support of a domain expert, you will still need to design solutions and write code corresponding to data you may be deeply unfamiliar with. As such, it is incredibly important to gain exposure to this reality as part of one’s data science education. My co-instructor and I teach in a design and engineering department, with students largely interested in pursuing fields such as UI/UX research and data engineering. We chose to make them work with rocks precisely because we knew they were unlikely to know too much about them (at least at the level of detail needed) beforehand. And that lack of prior knowledge made all the difference. Final Thoughts If you’re reading this, I’m guessing you’re training to be a data scientist, or interested in doing so. Perhaps you already are one and are just rounding out your knowledge. Whatever your position may be, my point remains the same: Every chance you get, expose yourself to new data. By its very nature, literally every field, every discipline, every topic known to man has some kind of data, and an associated group of people interested in gaining insights about it. And the person they turn to for help might just be you. The post I Teach Data Viz with a Bag of Rocks appeared first on Towards Data Science.

Limited access Under anti-vaccine advocate RFK Jr, FDA to limit access to COVID-19 shots FDA will require big, pricy trials for approvals for healthy kids and adults >65. Beth Mole – May 20, 2025 3:18 pm | 16 U.S. Secretary of Health and Human Services Robert F. Kennedy Jr. testifies before the Senate Committee on Health, Education, Labor, and Pensions on Capitol Hill on May 20, 2025 in Washington, DC. Credit: Getty | Tasos Katopodis U.S. Secretary of Health and Human Services Robert F. Kennedy Jr. testifies before the Senate Committee on Health, Education, Labor, and Pensions on Capitol Hill on May 20, 2025 in Washington, DC. Credit: Getty | Tasos Katopodis Story text Size Small Standard Large Width * Standard Wide Links Standard Orange * Subscribers only   Learn more Under the control of anti-vaccine advocate Robert F. Kennedy Jr., the Food and Drug Administration is unilaterally terminating universal access to seasonal COVID-19 vaccines; Instead, only people who are age 65 years and older and people with underlying conditions that put them at risk of severe COVID-19 will have access to seasonal boosters moving forward. The move was laid out in a commentary article published today in the New England Journal of Medicine, written by Trump administration FDA Commissioner Martin Makary and the agency's new top vaccine regulator, Vinay Prasad. The article lays out a new framework for approving seasonal COVID-19 vaccines, as well as a rationale for the change—which was made without input from independent advisory committees for the Food and Drug Administration and the Centers for Disease Control and Prevention. Normally, the FDA's VRBPAC (Vaccines and Related Biological Products Advisory Committee) and the CDC's ACIP (Advisory Committee on Immunization Practices) would publicly review, evaluate, and discuss vaccine approvals and recommendations. Typically, the FDA's scope focuses on licensure decisions, made with strong influence from VRBPAC, while the CDC's ACIP is principally responsible for influencing the CDC's more nuanced recommendations on usage, such as for specific age or risk groups. These recommendations shape clinical practice and, importantly, health insurance coverage. Makary and Prasad appear to have foregone those norms, even though VRBPAC is set to meet this Thursday to discuss COVID-19 vaccines for the upcoming season. Restrictions In the commentary, Markary and Prasad puzzlingly argue that the previous universal access to COVID-19 vaccines was patronizing to Americans. They describe the country's approach to COVID boosters as a "one-size-fits-all" and write that "the US policy has sometimes been justified by arguing that the American people are not sophisticated enough to understand age- and risk-based recommendations. We reject this view." Previously, the seasonally updated vaccines were available to anyone age 6-months and up. Further, people age 65 and older and those at high risk were able to get two or more shots, based on their risk. So, while Makary and Prasad ostensibly reject the view of Americans as being too unsophisticated to understand risk-based usage, the pair are installing restrictions to force their own idea of risk-based usage. Even more puzzlingly, in an April meeting of ACIP, the expert advisors expressed clear support for shifting from universal recommendations for COVID-19 boosters to recommendations based on risk. Specifically, advisors were supportive of urging boosters for people age 65 and older and people who are at risk of severe COVID-19—the same restrictions that Makary and Prasad are forcing. The two regulators do not mention this in their NEJM commentary. ACIP would also likely recommend a primary series of seasonally matched COVID-19 vaccines for very young children who have not been previously exposed to the virus or vaccinated. ACIP will meet again in June, but without a permissive license from the FDA, ACIP's recommendations for risk-based usage of this season's COVID-19 shots are virtually irrelevant. And they cannot recommend usage in groups the FDA licensure does not cover. It's unclear if a primary series for young children will be available and, if so, how that will be handled moving forward. New vaccine framework Under Makary and Prasad's new framework, seasonally updated COVID-19 vaccines can continue to be approved annually using only immunology studies—but the approvals will only be for people age 65 and over and people who are at high risk. These immunology studies look at antibody responses to boosters, which offer a shorthand for efficacy in updated vaccines that have already been through rigorous safety and efficacy trials. This is how seasonal flu shots are approved each year and how COVID boosters have been approved for all people age 6 months and up—until now. Moving forward, if a vaccine maker wants to have their COVID-19 vaccine also approved for use in healthy children and healthy adults under age 65, they will have to conduct large, randomized placebo-controlled studies. These may need to include tens of thousands of participants, especially with high levels of immunity in the population now. These trials can easily cost hundreds of millions of dollars, and they can take many months to complete. The requirement for such trials will make it difficult if not impossible for drug makers to conduct them each year and within a timeframe that will allow for seasonal shots to complete the trial, get regulatory approval, and be produced at scale in time for the start of respiratory virus season. Makary and Prasad did not provide any data analysis or evidence-based reasoning for why additional trials would be needed to continue seasonal approvals. In fact, the commentary had a total of only eight references, including an opinion piece Makary published in Newsweek and a New York Times article. "We simply don’t know whether a healthy 52-year-old woman with a normal BMI who has had COVID-19 three times and has received six previous doses of a COVID-19 vaccine will benefit from the seventh dose," they argue in their commentary. Their new framework does not make any mention of what will happen if a more dangerous SARS-CoV-2 variant emerges. It also made no mention of vaccine usage in people who are in close contact with high-risk groups, such as ICU nurses or family members of immunocompromised people. Context Another lingering question from the framework is how easy it will be for people dubbed at high risk to get access to seasonal shots. Makary and Prasad lay out a long list of conditions that would put people at risk of severe COVID-19 and therefore make them eligible for a seasonal booster. The list includes: obesity; asthma, lung diseases; HIV; diabetes; pregnancy; gestational diabetes; heart conditions; use of corticosteroids; dementia; physical inactivity; mental health conditions, including depression; and smoking, current or former. The FDA leaders estimate that between 100 million and 200 million Americans will fit into the category of being at high risk. It's unclear what such a large group of Americans will need to do to establish eligibility every year. In all, the FDA's move to restrict and hinder access to seasonal COVID-19 vaccines is in line with Kennedy's influential anti-vaccine advocacy work. In 2021, prior to taking the role of the country's top health official, Kennedy and the anti-vaccine organization he founded, Children's Health Defense, petitioned the FDA to revoke authorizations for COVID-19 vaccines and refrain from issuing any approvals. Ironically, Makary and Prasad blame the country's COVID-19 policies for helping to erode Americans' trust in vaccines broadly. "There may even be a ripple effect: public trust in vaccination in general has declined, resulting in a reluctance to vaccinate that is affecting even vital immunization programs such as that for measles–mumps–rubella (MMR) vaccination, which has been clearly established as safe and highly effective," the two write, including the most full-throated endorsement of the MMR vaccine the Trump administration has issued yet. Kennedy continues to spread misinformation about the vaccine, including the false and debunked idea that it causes autism. "Against this context, the Food and Drug Administration seeks to provide guidance and foster evidence generation," Makary and Prasad write. Beth Mole Senior Health Reporter Beth Mole Senior Health Reporter Beth is Ars Technica’s Senior Health Reporter. Beth has a Ph.D. in microbiology from the University of North Carolina at Chapel Hill and attended the Science Communication program at the University of California, Santa Cruz. She specializes in covering infectious diseases, public health, and microbes. 16 Comments

The Voice of America (VOA) Building in Washington, D.C. could be up for sale, White House senior adviser Kari Lake recently told reporters. The stripped classicist ensemble from 1939 by Charles Zeller Klauder—formerly called the Wilbur J. Cohen Federal Building—is “obscenely expensive” to lease, Lake said. The VOA Building is 1 million square feet and located just off the National Mall. It hosts media entities that played important roles in the Cold War like Radio Free Europe, Radio Free Asia, and the Office of Cuba Broadcasting. It’s also historically had the Agency for Global Media, and the Department of Health and Human Services. The Trump administration has proposed closing the VOA, but that shutdown was temporarily stopped by a federal judge. Instead of leaning upon the VOA and Agency for Global Media, Lake said, the White House will filter media through the One America News Network. U.S. Department of Housing and Urban Development in Washington, D.C. (F. Delventhal/Flickr/CC BY 2.0) The announcement comes not long after U.S. Housing and Urban Development (HUD) secretary Scott Turner said he wants to sell HUD headquarters, the Robert C. Weaver Building designed by Marcel Breuer, another famous Brutalist complex. HUD has also drastically downsized staff and support for social programs. Turner recently told Fox News that HUD headquarters is the “ugliest building in D.C.” The FBI also recently shared its plans to vacate its headquarters in the J. Edgar Hoover Building. FBI Director Kash Patel said in an interview with Fox Business, “This FBI is leaving the Hoover Building because this building is unsafe for our workforce.” The Hoover Building, a behemoth Brutalist structure, was designed by Charles F. Murphy and Associates in the 1960s. The General Services Administration (GSA) was ordered to sell off approximately 500 “non-core” buildings it owns and/or leases last February. GSA subsequently scaled back its plans, but is still trying to downsize its property portfolio. On its website, GSA lists several of the properties it wants to get rid of, and counting. A federal courthouse in Los Angeles, the U.S. Custom House in Philadelphia, Goodfellow Federal Center in St. Louis, and others are on the list. To raise awareness and appreciation for the architectural style, the National Building Museum extended its exhibition Capital Brutalism through June 30. Aileen Fuchs, National Building Museum president and executive director, said the exhibition’s aim is to provide “additional historical context for Brutalist architecture,” and it’s meant to inspire “visitors to rethink their perceptions of beauty and equity as we consider the future of these Brutalist buildings.”

Humpback Whales Give Birth Much Farther South Than Previously Thought, Study Finds Researchers in Australia found records of humpback whale calves more than 900 miles farther south than expected A mother and baby whale spotted swimming near Kiama, New South Wales, Australia.  Vanessa Risku / Instagram: droning_my_sorrows Each year, humpback whales migrate predictably between cold, nutrient-rich feeding grounds and warm, tropical calving grounds, where they give birth and care for their young. Or so scientists thought. Researchers in Australia recorded calves being born much farther south than previously thought, which challenges long-held assumptions about humpback migration patterns. They shared their findings in a new paper published yesterday in the journal Frontiers in Marine Science. Humpback whales (Megaptera novaeangliae) live in all of the world’s oceans. These behemoth marine mammals are highly migratory, swimming thousands of miles each year. The longest humpback whale migration ever recorded was 8,106 miles, from South America to Africa, though more typical distances are around 5,000 miles. In the Southern Hemisphere, humpbacks head to the krill-rich waters of the Southern Ocean around Antarctica every summer, then venture up the eastern and western coasts of Australia toward the warmer waters north of the continent every winter. Researchers long believed that humpbacks only gave birth to their calves in the tropical waters north of 28 degrees latitude on Australia’s east side and north of 23 degrees latitude on the west side. “Our general concept of humpback whale ecology is that they feed at high latitudes in the poles, breed in the tropics, and they have this migration that's driven between the two needs,” says lead author Jane McPhee-Frew, a marine guide and biologist at the University of New South Wales, to the Australian Broadcasting Corporation’s Ellen Phiddian. But, for the new study, scientists found more than 200 records of calves spotted south of these cut-offs. Some were seen as far south as Tasmania and New Zealand, at around 43 degrees latitude, which is between 807 and 932 miles farther south than expected. And that’s just where the sightings stopped—it’s possible humpbacks are giving birth even farther south. “Eventually, we just ran out of land to see them from, so we don’t actually know where the limit is,” McPhee-Frew says in a statement from the university. The scientists gathered the baby humpback whale sightings from a variety of sources, including migration surveys, stranding reports and citizen science observations from tourism boats. In total, the data included 209 observations consisting of 11 births, 41 strandings and 168 live calves; the scientists believe the reports represent at least 169 individual calves. Though the stranding data went back as far as 1991, most of the opportunistic observations occurred more recently, from 2016 forward. Roughly two-thirds of the total opportunistic observations occurred in 2023 and 2024. Since this is the first study of its kind, the scientists aren’t sure whether this is a new trend or just something no one had noticed before. “It may be the case that this has always been happening and we’ve just not documented it well or been paying attention,” study co-author Vanessa Pirotta, a wildlife scientist at Macquarie University, tells the Guardian’s Petra Stock. “Or it may be something new is happening because waters are becoming warmer further south of those traditional tropical areas, which means that they’re more favorable for a humpback whale to have a calf.” The findings were not necessarily a surprise, either, because humpbacks are known to feed, mate and give birth along their migratory corridor, says Olaf Meynecke, a marine scientist at Griffith University who was not involved with the study, to the Australian Broadcasting Corporation. However, scientists “don't know to what extent they are doing this,” he adds. The findings also raise additional questions. The biggest one? If humpbacks don’t need to be in tropical waters to give birth, then why are they migrating to do it? “I don't know, but it's exciting,” McPhee-Frew tells the Australian Broadcasting Corporation. One possibility is that whales born in warmer waters have better chances of survival. These regions have fewer predators, like orcas, and the waters tend to be calmer. The warmer climate may also be beneficial for baby whales—called neonates—with poor temperature regulation. In addition, calves born farther south must still migrate northward with their mothers. This long journey is not only taxing, but it also takes them through busy shipping lanes and past urbanized areas, which opens them up to risks like boat strikes, entanglement and pollution. “Newborns are like Great Dane puppies,” says study co-author Tracey Rogers, also a biologist at the University of New South Wales, in the statement. “They have those long, enormous fins that they need to grow into, and they’re not very strong swimmers. So they rest a lot of the time on their mum’s back.... It’s heartbreaking to think of these young whales traveling through busy ports and dangerous shipping lanes with those long, clumsy fins.” Moving forward, researchers might be able to answer some of these questions by following calves born in cold waters and comparing their health and longevity to those born in warm waters. Scientists are also curious about other effects, like whether calves born in colder waters also return to these same areas when they’re ready to reproduce. In the meantime, the scientists say their findings could have important conservation implications. For instance, with calves being born farther south than previously thought, policymakers may want to consider expanding marine protected areas or launching awareness-raising campaigns to help keep them safe. “We can't be in a situation where we're putting any age of whales—especially baby whales—in a situation where they're getting caught in nets, being exposed to chemicals, being hit by boats and being harassed,” McPhee-Frew says in the statement. Get the latest stories in your inbox every weekday.