Feed aggregator

It is both staggering and utterly unsurprising to get proof positive that Health and Human Services Secretary Robert F. Kennedy Jr. bald-faced lied to the Senate during his confirmation hearing. Honestly, with this administration, not lying is probably a rarer occurrence, but nonetheless. The Guardian and The Associated Press got their mitts on letters proving that Kennedy’s claim that a…

Source

A cartoon by Pedro Molina. Related | ICE finds heinous new way to spend millions and terrorize people…

Source

Shots were fired at the Hot Chips conference in California this week as Nvidia announced its new Groq-3-based LPX racks had entered production, with early tests showing the systems churning an eye-watering 3,400 tokens a second in Gemma 4 31B. That's four times faster than rival Cerebras. A day later, Cerebras fired back, touting nearly equivalent performance from its next-gen CS-4 accelerators revealed last week. These top-line performance figures make inference feel instantaneous relative to the chatbots we've grown accustomed to over the past four years. But the two companies are essentially arguing over numbers their customers will probably never see in production. To be clear, neither is lying. If you wanted to recreate these results, you certainly could — Artificial Analysis, the team responsible for both sets of benchmarks, knows what they're doing — but beyond a marketing gimmick, no inference-as-a-service model operator in their right mind would run either system this way. Not unless you've somehow figured out how to make a profit serving one request at a time. In reality, these numbers are more like the top speed on a race car. It makes for great marketing but you probably aren't driving that fast on a regular basis, and if you did, you wouldn't get very far before your tank runs dry. These aren't the numbers that matter Unsurprisingly, the economics of "premium" or "ultra-low latency" inference are a bit more nuanced, but ultimately come down to how efficiently you can scale that performance at the rightmost end of the Pareto frontier. We've shown the graphic below before. It charts the performance characteristics of how different Nvidia B300 configurations perform across a Pareto front. As you can see, GPUs are great for high-throughput, low-interactivity applications, but run out of steam quite quickly as per-user generation rates climb. Chips like Nvidia's Groq-3 LPUs or Cerebras' dinner plate-sized accelerators offer massive bandwidth thanks to their SRAM-heavy architectures, so they can extend and push the Pareto curve out to the right. Unfortunately, the same thing that makes them so fast also limits their ability to scale that performance to large numbers of users. To understand why we need to walk through a little math. We're going to pick on Gemma 4 31B, in part because we have rough numbers for both Nvidia and Cerebras, but also because it's a small model and therefore presents a best-case scenario. Each Groq-3 accelerator is equipped with 500 MB of on-chip memory capacity. Nvidia achieved its 3,400 tok/s per user run at 8-bit precision. So to run a 31 billion-parameter model, we need about 31 GB of memory or about 64 LPUs. But that's just to hold the model's weights. Since it's running as a standalone system, we also have to account for the KV caches generated during the prefill stage in which the prompt is processed. According to LMcache's KV calculator, at 8-bit precision, a 100,000-token input sequence is going to chew up another 8 GB of memory, and that's per sequence. So for every additional 100,000 token input, that's another 16 LPUs worth of memory. By our estimate, that means the most a single LPX rack with its 256 LPUs can manage on its own before running out of memory is a batch size of 12 at a 100,000 token input length each. The numbers look a little different for Cerebras' CS4, but the conclusion is largely unchanged. Each CS4 rack contains three massive WSE-3T accelerators with 44 GB of SRAM apiece, totaling 132 GB of capacity. That leaves the system a bit more headroom, but we're still looking at a maximum batch size of 12. To be clear, that's not a compute limitation. The systems simply don't have enough memory to support larger batches without adding additional racks or cutting the prompt size. Sure, we're oversimplifying a bit. We're assuming batch inference with fixed input sequences, not concurrent users hitting the systems asynchronously, and with different input and output sequences each time. For a chatbot, especially one with a limited context window — think the model's short term memory — a single CS4 or LPX rack could conceivably serve far more users than the 12 noted here. Lifting the curve Even if both systems can hit 3,400 tok/s, the better platform is going to be the one that delivers better scale at a lower cost and power consumption. The benchmark figures touted by Nvidia and Cerebras this week are really measuring the top speed the systems could run on, not the speed customers will actually run them. In fact, customers probably won’t run these systems on their own. These SRAM-heavy architectures work a lot better as decode accelerators, which is exactly why Nvidia spent $20B to license Groq's tech and poach its engineering team. It's also why Cerebras recently teamed up with AWS and AMD. We've discussed these kinds of heterogeneous compute architectures in the past, but in a nutshell they involve GPUs or other compute heavy XPUs, like Amazon's Trainium chips, with something like Groq's LPUs, Cerebras' waferscale accelerators, or SambaNova's SN-series parts. This combination largely eliminates, or at least minimizes, the KV cache and compute overheads that made scaling Cerebras' and Groq's chips so difficult. As decode accelerators, they only really need to hold the model weights, as almost everything else can get on just fine on a GPU. Again, top line AI performance makes for great headlines, but the far right of the Pareto is rarely where you'll want to be. As is usually the case with Pareto curves, the sweet spot usually presents itself as a notable hump or knee pushing toward the upper right corner of the chart. Here you'll be trading some per-user interactivity for better concurrency, throughput, and economics. Combining GPUs with Cerebras or Groq accelerators is what Nvidia, AMD, and AWS are all counting on to make "premium inference" cost-effective. There's a better benchmark for this We've yet to see benchmarks combining Cerebras' or Groq's chips with Nvidia, AMD, or AWS' respective GPUs and XPU platforms, but these are the metrics that are actually going to matter. SemiAnalysis' InferenceX offers a better representation of whether or not adding a CS4 or LPX rack would even help, as we can see exactly how each system performs across the full Pareto, ranging from high-volume low-interactivity "bulk" tokens on the left, to ultra-low-latency but also lower-volume tokens on the right. In fact, while Nvidia and Cerebras opted for Artificial Analysis' simpler test methodology, OpenAI actually opted for InferenceX to show how its spicy new Jalapeño accelerators perform across bulk and premium inference regimes. So why didn't Nvidia and Cerebras do the same? If we had to guess, it's because their LPX and CS4 racks are still best suited toward low-latency inference and wouldn't look as impressive on the left and middle sections of the Pareto curve. We fully expect that both companies will share InferenceX results for their combined GPU and LPX/CS4 systems when they're ready, but no marketer ever got a bonus for making their product look worse than it really is, so we'll just have to wait. In any case, the next time you see someone boasting of some impossibly fast inference accelerator, it's worth asking where on the Pareto that number came from. ®

The U.S. military is experiencing a “beyond critical” shortage of advanced missile interceptors in Europe largely driven by President Donald Trump’s war with Iran, a U.S. defense official in Europe and a NATO official told The Associated Press, raising concerns about vulnerabilities in NATO countries to a potential Russian attack. The U.S. defense official said the most concerning shortfall…

Source

⚡ Quick Take

• LEGO has revealed its latest deluxe Star Wars set, Darth Vader’s Super Star Destroyer Executor.
• This massive set is LEGO’s longest ever Star Wars build in size, and contains 6,130 separate pieces.
• Along with Darth Vader, this set comes with 6 bounty hunter minfigures.
• Nerdist’s take: This LEGO build ranks alongside the Death Star among the best Star Wars sets ever.

The Star Wars franchise is no stranger to amazing and iconic spaceships. Sure, the good guys might have the best ones, as it’s hard to beat the Millennium Falcon and the X-Wings. But those Imperials are no slouches when it comes to cool ships. Case in point, Darth Vader’s flagship, the Super Star Destroyer Executor, which debuted in The Empire Strikes Back. Now, that iconic starship has earned the honor of being LEGO’s longest Star Wars brick-built model released so far, measuring over 53.5 inches (136 cm) in length. And just try to get John Williams’ “Imperial March” out of your head as you gaze upon this bad boy. We know we can’t! You can check out several images of the new 6,130-piece LEGO Star Wars Super Star Destroyer build down below in our gallery.

LEGO Star Wars Executor Super Star Destroyer Click To View Gallery The LEGO Group/Lucasfilm Lucasfilm/The LEGO Group The LEGO Group/Lucasfilm Lucasfilm/The LEGO Group The LEGO Group/Lucasfilm Lucasfilm/The LEGO Group The LEGO Group/Lucasfilm Lucasfilm/The LEGO Group The LEGO Group/Lucasfilm Lucasfilm/The LEGO Group The LEGO Group/Lucasfilm Lucasfilm/The LEGO Group The LEGO Group/Lucasfilm Lucasfilm/The LEGO Group

The LEGO Star Wars Executor set comes with a few smaller ships for scale, including a standard Imperial Star Destroyer. In The Empire Strikes Back, George Lucas used the standard Star Destroyer, which seemed so huge to audiences in A New Hope, to show the sheer scale of the new model. Also included in this build is an even smaller Millennium Falcon, attached to the Super Star Destroyer’s rear. This, of course, echoes Han Solo’s maneuver as he attached the Falcon to the Executor to evade their detection. After which he then floated away “with the rest of the garbage” when they dumped their trash into space.

The LEGO Group

Of course, no LEGO set is complete without some minifigures. And it comes with all six bounty hunters from Empire; Boba Fett, Dengar, Bossk, IG-88, 4-LOM, and Zuckuss. And, of course, Darth Vader. This new set represents the first time LEGO has released all of these characters together as minifigures. Sadly, this one doesn’t come with the string of terrified Imperial officers Vader terrorizes throughout the film. No dead Admiral Ozzel or Captain Needa, or always fretting Admiral Piett. We’ll leave them to the LEGO customizers. The Super Star Destroyer Executor will become available for purchase for $799.99, starting October 1 via LEGO Stores and LEGO.com/Star-Wars.

Eric Diaz is a staff writer at Nerdist and a Swiss Army knife of fandoms. Eric loves DC, Marvel, horror movies, and even Star Wars and Star Trek equally.

The post LEGO Reveals Massive STAR WARS Super Star Destroyer Set appeared first on Nerdist.

ph @rossa_di_sera_caffe

House Speaker Mike Johnson told a room full of GOP acolytes that the Republican midterm convention is not a desperate attempt to engage voters but a chance to remind them of President Donald Trump’s disastrous stewardship of the country. Johnson recalled a phone conversation in which Trump ordered him to hold a “midterm convention,” remarking, it’s the “first time in history any party’s ever…

Source

Pajaro Valley Unified School District officials and its teachers union announced that they have reached a tentative agreement after nearly a year of contentious negotiations.

The post Pajaro Valley Unified teachers reach tentative agreement with district appeared first on Lookout Santa Cruz.

CRPx0, a cybercrime crew that has rapidly evolved from a scam service to a ClickFix-delivered ransomware and crypto-theft business over the summer, claims its victim count jumped from fewer than 10 in June to 48 organizations on its clear-web leak site at the time of publication. Keep in mind: criminals aren’t always the most trustworthy bunch, so take their claims with a healthy dose of salt. Still, the ransomware biz’s expanding operations, unique payload, and white-label hacking service make it one to watch, and a few recent analyses provide tips for defenders to keep the crooks out of their IT environments. Rakesh Krishnan, a threat-intel analyst who writes about cybercrime investigations on TheRavenFile blog, was one of the first researchers to publish details about CRPx0 at the beginning of the month, including previously unreleased malware samples. The operators offer a hacking service, providing “complete database extraction” from victim organizations and “optional public leak coordination upon request.” This platform also advertises full network compromise, “from initial access, through lateral movement, to full domain compromise,” plus persistent access across the victim’s infrastructure. A second, white-label, ransomware-as-a-service platform makes it really easy for wannabe crooks to get into the data-theft and extortion biz. CRPx0 builds, deploys, and configures everything from command-and-control infrastructure to the negotiation panel and malware, allowing its criminal customers to bring their own brand identity to their operations - and originally offered these services for a $10,000 one-time fee. At first, the operators promised to allow affiliates to keep 100 percent of their profits - this is unheard of in ransomware-as-a-service operations. This has since shifted to a 70-30 model, with affiliates receiving 70 percent of the extortion payments (after a one-time $333 enrollment fee) and the remaining 30 percent going to the operators. CRPx0 rules prohibit affiliates from infecting Commonwealth of Independent States (CIS) member countries and organizations based in these countries. This is a pretty common rule among Russia-based ransomware operations, and in an earlier interview with The Register, Recorded Future threat intelligence analyst Allan Liska called it the “first rule of ransomware club: you don't attack organizations in the Commonwealth of Independent States.” The gang also prefers Monero (XMR) payments, rather than Bitcoin (BTC). ClickFix ransomware delivery Affiliates can also customize their own ClickFix payload delivery. The operators offer two lures, a fake Windows Update and a fake Google reCAPTCHA, to socially engineer victims into executing the initial command, according to a Ransom-ISAC research team analysis published on Thursday. The Windows lure tricks a victim into pasting a PowerShell command into the Run dialog. This drops a DLL stager chain and ultimately deploys Python-based ransomware. The macOS lure, however, uses a curl|bash command that downloads portable Python and the ransomware directly. The lures and the rest of CRPx0’s malware run on Windows and macOS, and according to the researchers, there are four payload formats: “the two HTML lures plus a standalone DLL and a standalone EXE, both of which discard the social engineering step entirely.” All four deliver the same ransomware: a 1,769-line Python script that steals high-value files before encrypting them with AES-128-CBC (Fernet). The malware moves laterally via WMI/schtasks, and delivers a ransom note that gives victims a 48-hour deadline to pay up - or see their files leaked. 'Complete, professional offensive control center' On August 23, the CRPx0 operators published a v3.0 update note on the group’s clearnet leak site, promising “a complete, professional offensive control center for managing compromised remote machines from a single web dashboard.” It provides crims with tools to steal valuable files, credentials, and wallet recovery phrases and keys, while “watching stolen cryptocurrency wallet addresses flow in.” This service also provides scripts to run remote commands, and the control panel sets up “automated attack reactions that fire on their own when something valuable happens on a target.” As the operators note: “Everything is built to be operated by a human with no technical background: point-and-click panels, plain-language rules, and clear status indicators. The underlying attack engine is hidden behind a clean, dark-themed interface.” CRPx0’s hacking and ransomware services, enabling everything from crypto theft to encryptors and full network compromise, “could be a strategic move to attract new recruits, or a scam targeting a range of affiliate hopefuls seeking cybercrime services,” according to an August 12 analysis from Jade Brown, a threat researcher at Bitdefender. What defenders should do Still, “other threat actors may attempt to adopt similar techniques,” she warns. “This is a reminder that organizations should balance detection capabilities in preparation for different types of compromises, configuring technologies to detect and block malicious behavior that aligns with both crypto theft and encryption processes.” The Ransomware-ISAC team says defenders should prioritize five actions, in this order. “The first three cost nothing and blunt the entire ClickFix class of attack, not just CRPx0,” they note. First: remove the Run dialog for standard users - this will entirely block the Windows path. For macOS users: restrict Terminal via MDM for non-technical staff. Next, the threat-intel analysts advise defenders to alert on RunMRU writes containing powershell, curl or long base64 strings. “Every ClickFix victim leaves a trace at HKCU\Software\Microsoft\Windows\CurrentVersion\Explorer\RunMRU,” they wrote. “This is the highest-fidelity ClickFix detection available and it is trivial to deploy.” The team also lists adversary network indicators and suggests blocking those. Make sure to hunt for indicators and malicious behavior in the pre-encryption exfiltration window. “Data leaves before a single file is encrypted, so .crpx0 extensions and ransom notes are a post-mortem indicator, not a warning,” according to the Ransomware-ISAC. Finally, confirm backups are unreachable from the account that would be compromised, and treat anything reachable with the victim's credentials as destroyed. ®

Wyoming’s Republican Governor Mark Gordon is sounding the alarm about the Trump administration’s “aggressive” and “irregular” elections monitoring.

Speaking at a canvassing board meeting Wednesday, Gordon expressed concerns about the Department of Justice’s surprise efforts to monitor election sites in the state during primary elections last week, WyoFile reported.

“To my knowledge, the U.S. Attorney was unaware of it, and it sounded like the behavior of the observers was a bit aggressive,” Gordon said.

“It just was irregular, and I have asked the attorney general to start looking into, you know, kind of the nature of the visits and all that sort of stuff,” he said.

It’s typical for federal observers to monitor elections, but Gordon claimed that the behavior of those at last week’s voting seemed to overstep the authority Congress had granted to them. “I’m not saying anything went wrong or anything else, and I certainly think the conduct of this election was phenomenal,” he said.

“I feel very strongly about federal government coming in and taking our guns,” Gordon said. “I don’t like that. And I don’t like federal government coming in and taking our votes either.”

In a statement to WyoFile Wednesday evening, Debra Lee, the chief election officer for Laramie County, recounted two monitors’ behavior after they arrived at a polling station in Wyoming’s most populous district. Lee said she was not aware that they were coming, and they did not offer credentials until requested.

The two monitors arrived at the busy polling station and asked staff questions that appeared to be “outside the scope of federal voting rights laws,” including questions about poll worker training and new election equipment. They also asked to examine their tabulators, according to Lee.

When Lee discovered the monitors intended to interview voters in order to “assess compliance with federal civil rights and voting rights laws,” the election official denied their request. “We do not allow interviews to be conducted in polling places,” she said.

“This was a poorly managed exercise and one that showed complete disregard for our state laws and administration of our elections,” she concluded.

The controversy follows a set of primary elections where candidates claiming elections are safe as is did extraordinarily well in Wyoming.

Monday’s Supreme Court ruling has set off a new court scramble to block the mail ballot restrictions for this November’s election. By Dion Nissenbaum for Votebeat Monday’s U.S. Supreme Court ruling in favor of President Donald Trump’s attempt to restrict mail ballots via executive order is procedural and narrow, but it’s fueling broad confusion and uncertainty as the November…

Source

Picture this: You’ve just undergone one of the most harrowing experiences of your life, and you’re trapped in the rubble of a collapsed building. You’re injured, air is running out, and human rescuers can’t reach you. But crawling through the rubble to administer life-saving drugs and reassure you comes the last thing you’d expect: a swarm of cyborg cockroaches bearing tiny syringes. Unfortunately for those who’d rather find solace in the arms of a fluffy, drooly St. Bernard, this tale is anything but science fiction. It’s not even a new idea: Cockroaches fitted with electronics that allow rescue workers to control their movements from a distance have been on the research docket for biotech researchers for some time. What’s new in this latest experiment from researchers at Australia’s University of Queensland is outfitting the, erm, little buggers with devices remote rescuers can use to shoot a drug dart into trapped patients. “A lot of people might not like the sight of a giant cockroach scurrying towards them, but if you’re trapped in rubble or stuck in a cave and need help, it could make a real difference between life and death,” PhD candidate and lead research paper author Hai Nhan Le told the University of his phobia-triggering project. Giant is right – the species chosen for the project, the giant burrowing cockroach, is the world’s heaviest roach species, and can grow up to 8 centimeters in length, or around 3.1 inches. Known to be docile, they’re popular pets in their native Australia, and – lucky for anyone coming face to face with one while trapped in a pile of rubble – they’re known to be relatively clean and disease-free, too. The giant burrowing cockroach was chosen for its size, per the paper, as they needed to be large enough to “integrate centimetre-scale apparatuses for … real world scenarios.” The “paraborgs,” as the team calls their subjects, were outfitted with electrodes that gave researchers the ability to make the roaches turn left and right, control their speed, and immobilize them in order to deliver drugs without the roach moving while scientists tried to take aim. Then they were fitted with various add-ons specific to their roles in the paramedic swarm –one built to search, another equipped with a single launchable drug syringe, for example. Outfitting the roaches with their cyborg backpacks isn’t fatal, and the team said that their subjects were able to live just as long as other roaches once their harnesses were removed. *shudder* Well, at least they're good at their jobs The Queensland team had overwhelming success in getting their paraborgs to end up where they wanted them - 100 percent successful, in fact, the paper said. As for drug injection, the auto-injection mechanism succeeded 90 percent of the time, and paraborgs able to get within 15 cm of a target (5.9 inches) were able to successfully inject drugs 95 percent of the time. The only trouble spot for the experiment was getting the robo roaches to sit still, with only a 72 percent success rate in stabilizing the cockroaches for injection targeting. That’s still a pretty good rate of success, leaving the team to predict that swarms of specially-outfitted paraborgs could be swarming over disaster sites in a matter of a few short years. “Some could carry cameras and environmental sensors, while others could carry specialised emergency or medical equipment,” UQ robotics engineering lab director and lecturer Dr. Thang Vo-Doan said of the project. “Hopefully, within the next 5 to 10 years, we could see cyborg insect rescue teams deployed to help people in real emergencies.” Hopefully large bugs bearing medical aid isn’t something that’ll send you into a panic. ®

A cartoon by Clay Bennett. Related | Remembering Dolly…

Source

Eight large and colorful sculptures that were inspired by Mexican folk art have been installed around the city of Salinas as part of a world travelling exhibit.

“Alebrijes y Nahuales: Fantastic Animals from Mexico” touches on magical realism through colorful mythical creatures, all created by artists from Oaxaca and central Mexico.

Three of the sculptures are stationed at the Northridge Mall.

WORLD TRAVELER “The Armadillo” has been positioned in front of Sherwood Hall. (Tarmo Hannula/The Pajaronian)

Standing nearly 20-feet tall, the pieces are equipped with solar lighting, allowing their intricate craftsmanship to stand out both day and night.

Other locations include the Star Center, at Main Street and Blanco Road, Vineyard Church, 1122 East Alisal St., Sherwood Hall, 940 North Main St., Salinas Regional Soccer Complex, 1440 Constitution Blvd., and the Steinbeck Center, 1 Main St.

The sculptures will be up until Sept. 4.

FROM MEXICO “The Dog” now keeps watch in front of the Star Market in Salinas. (Tarmo Hannula/The Pajaronian)

Florida Governor Ron DeSantis’s administration “misappropriated” $10 million from a state Medicaid settlement to lobby against weed.

According to a Florida grand jury report, which was first obtained by CBS News, the money moved from a state Medicaid settlement, to a charity tied to the governor’s wife, Casey DeSantis, to a pair of dark-money PACs, and finally into a campaign to help kill a marijuana legalization ballot measure opposed by the governor.

It is not clear who directed the transfer of funds, which were meant to help poor children receive health insurance, the grand jury noted. As a result, the grand jury found “insufficient evidence to charge anyone criminally.”

“Nobody will take responsibility for deciding the $10 million of taxpayer money would go to Hope Florida,” the jury concluded. “While we can’t prove who is responsible, we can plainly see that taxpayer money was misused for political purposes and we would like to see changes made to prevent this from happening again.”

The grand jury’s report is dated January 28, 2026, and is still officially sealed.

The Hope Florida Foundation is the fundraising arm of a charity, Hope Florida, started by Casey DeSantis in 2021. The organization intends to connect low-income families with churches and local groups to keep them off public assistance. In its own words, Hope Florida goes “beyond government to create a pathway for Floridians to live up to their God-given potential and accomplish the American Dream.”

The $10 million transferred to the charity was initially part of a Medicaid settlement between the state of Florida and Centene Corporation, a for-profit health care company, over claims of over-billing. In February 2023, Centene Corporation offered to pay the entire $67,048,611 settlement to the state. More than a year and a half later, the settlement was split up: “On September 27, 2024, State of Florida officials agreed that Centene would pay $57,048,611 to the State of Florida and $10,000,000 to Hope Florida Foundation, Inc.,” the grand jury’s report read. “Within days, the $10,000,000 was transferred to other organizations for political purposes because money was spent on a political campaign and went to the Republican Party of Florida.”

According to the grand jury report, “after the money was transferred to Hope Florida, it was quickly funneled from there into two political action committees”—$5 million to Secure Florida’s Future and $5 million to Save Our Society From Drugs, both “dark money” groups that are not required to release detailed financial reports. The two groups then used the cash to bankroll Keep Florida Clean, another PAC created specifically to oppose a ballot initiative to legalize marijuana. Although the marijuana amendment was popular across the state, it ultimately failed.

The report goes on to implicate a series of officials in Florida state government who may have had knowledge of the transaction, including Republican Senate candidate Ashley Moody, who was the state’s attorney general at the time. Moody allegedly “authorized” her chief deputy, John Guard, to sign the settlement agreement diverting $10 million to Hope Florida.

When asked about the grand jury’s report on Thursday, DeSantis claimed the only crime that was “apparent was whoever leaked the grand jury report.”

Centers for Medicare and Medicaid Services Administrator Mehmet Oz is at it again spinning anti-vax conspiracies. This time, he’s casting doubt on the two measles deaths in Pennsylvania earlier this week. “We don’t know exactly what happened in Pennsylvania yet,” Oz told Fox Business. “We don’t know if these children died from measles or they had another condition that was already attacking…

Source

ALT