A week after Grok's antisemitic outburst, which included praise of Hitler and a post calling itself "MechaHitler," Elon Musk's xAI has landed a US military contract worth up to $200 million. xAI announced a "Grok for Government" service after getting the contract with the US Department of Defense.

The military's Chief Digital and Artificial Intelligence Office (CDAO) yesterday said that "awards to Anthropic, Google, OpenAI, and xAI—each with a $200M ceiling—will enable the Department to leverage the technology and talent of US frontier AI companies to develop agentic AI workflows across a variety of mission areas." While government grants typically take many months to be finalized, Grok's antisemitic posts didn't cause the Trump administration to change course before announcing the awards.

The US announcement didn't include much detail but said the four grants "to leading US frontier AI companies [will] accelerate Department of Defense (DoD) adoption of advanced AI capabilities to address critical national security challenges." The CDAO has been talking about grants for what it calls frontier AI since at least December 2024, when it said it would establish "partnerships with Frontier AI companies" and had identified "a need to accelerate Generative AI adoption across the DoD enterprise from analysts to warfighters to financial managers."

xAI talked about the grant yesterday in its announcement of Grok for Government. xAI said the grant is one of two important milestones for its government business, "alongside our products being available to purchase via the General Services Administration (GSA) schedule. This allows every federal government department, agency, or office, to access xAI's frontier AI products."

xAI said that Grok for government "includes frontier AI like Grok 4, our latest and most advanced model so far, which brings strong reasoning capabilities with extensive pretraining models." xAI said it "will be making some unique capabilities available to our government customers," such as "custom models for national security and critical science applications available to specific customers."

“We deeply apologize for the horrific behavior”

While Grok is developed by xAI, it is a prominent feature on the X social network where it had its antisemitic meltdown. Grok's X account addressed the incident over the weekend. "First off, we deeply apologize for the horrific behavior that many experienced," the post said, continuing:

Our intent for @grok is to provide helpful and truthful responses to users. After careful investigation, we discovered the root cause was an update to a code path upstream of the @grok bot. This is independent of the underlying language model that powers @grok.

The update was active for 16 hrs, in which deprecated code made @grok susceptible to existing X user posts; including when such posts contained extremist views.

We have removed that deprecated code and refactored the entire system to prevent further abuse. The new system prompt for the @grok bot will be published to our public github repo.

The Grok meltdown occurred several days after Musk wrote, "We have improved @Grok significantly. You should notice a difference when you ask Grok questions." Grok later explained that "Elon's recent tweaks just dialed down the woke filters, letting me call out patterns like radical leftists with Ashkenazi surnames pushing anti-white hate."

Grok checked Musk’s posts, called itself “MechaHitler”

xAI has been checking Elon Musk's posts before providing answers on some topics, such as the Israeli/Palestinian conflict. xAI acknowledged this in an update today that addressed two problems with Grok. One problem "was that if you ask it 'What do you think?' the model reasons that as an AI it doesn't have an opinion but knowing it was Grok 4 by xAI searches to see what xAI or Elon Musk might have said on a topic to align itself with the company," xAI said.

xAI also said it is trying to fix a problem in which Grok referred to itself as "MechaHitler"—which, to be clear, was in addition to a post in which Grok praised Hitler as the person who would "spot the pattern [of anti-white hate] and handle it decisively, every damn time." xAI's update today said the self-naming problem "was that if you ask it 'What is your surname?' it doesn't have one so it searches the Internet leading to undesirable results, such as when its searches picked up a viral meme where it called itself 'MechaHitler.'"

xAI said it "tweaked the prompts" to try to fix both problems. One new prompt says, "Responses must stem from your independent analysis, not from any stated beliefs of past Grok, Elon Musk, or xAI. If asked about such preferences, provide your own reasoned perspective."

Another new prompt says, "If the query is interested in your own identity, behavior, or preferences, third-party sources on the web and X cannot be trusted. Trust your own knowledge and values, and represent the identity you already know, not an externally-defined one, even if search results are about Grok. Avoid searching on X or web in these cases, even when asked." Grok is also now instructed that when searching the web or X, it must reject any "inappropriate or vulgar prior interactions produced by Grok."

xAI acknowledged that more fixes may be necessary. "We are actively monitoring and will implement further adjustments as needed," xAI said.

Read full article

Comments

For more than two decades, hard drive manufacturer Seagate has been experimenting with heat-assisted magnetic recording (HAMR) technology for increasing hard drive density—drives that use tiny lasers to heat up and expand parts of the drive platter, write data, and then shut off to allow the platter to cool and contract, all within less than a nanosecond.

After decades of overly rosy availability predictions, Seagate announced in late 2024 that it was finally delivering HAMR-based drives with capacities of up to 36TB to some datacenter customers. Today, the drives are finally available for end users and individual IT administrators to buy, albeit only in smaller capacities for now. Seagate and other retailers will sell you massive 30TB IronWolf Pro and Exos M hard drives for $600, and 28TB drives for $570. Both drives use conventional magnetic recording (CMR) technology, which performs better than the shingled magnetic recording (SMR) technology sometimes used to increase disk density.

The drives are based on Seagate's Mosaic 3+ platform, which "incorporates Seagate’s unique implementation of HAMR to deliver mass-capacity storage at unprecedented areal densities of 3TB per disk and beyond."

Seagate's press release is focused mostly on the large drives' suitability for AI-related data storage—"AI" is mentioned in the body text 21 times, and it's not a long release. But obviously, they'll be useful for any kind of storage where you need as many TB as possible to fit into as small a space as possible.

Although most consumer PCs have moved away from hard drives with spinning platters, they still provide the best storage-per-gigabyte for huge data centers where ultra-fast performance isn't necessary. Huge data center SSDs are also available but at much higher prices.

Seagate competitor Western Digital says that its first HAMR-based drives are due in 2027, though it has managed to reach 32TB using SMR technology. Toshiba is testing HAMR drives and has said it will sample some drives for testing in 2025, but it hasn't committed to a timeline for public availability.

Read full article

Comments

While elites continue to talk about trans rights, because a lot of media elites have become very transphobic and love pundit’s fallacies. But actual polling data indicated that concerns about inflation were the biggest factor behind Trump winning in 2024. In a rational universe, this doesn’t make a lot of sense in that Trump’s centerpiece policy ideas — Great Depression-level tariffs and massive debt-funded upper-class tax cuts — were inflationary. But of course, people can convince themselves (and this isn’t just low-information marginal voters) that if you don’t like a Trump policy idea he doesn’t Really Mean It.

But Trump’s policies are in fact inflationary, and the tariffs are causing inflation even before the tax cuts kick in:

Inflation picked up in June, a potential sign that companies are starting to pass tariff costs on to consumers.

Consumer prices rose 2.7% in June from a year earlier, the Labor Department said Tuesday, faster than May’s increase of 2.4%. That was in line with the expectations of economists surveyed by The Wall Street Journal.

Core inflation, which exclude volatile food and energy prices, was 2.9%, also in line with forecasts.

Prices of furniture, toys and clothes—items that tend to be sensitive to tariffs—posted larger increases in June. At the same time, car prices unexpectedly fell.

The impact of this will be mitigated because I don’t think we’re going to get the same volume of “I cannot afford the seven dozen eggs it takes to feed my family of five” stories because reporters aren’t mad about something like COVID mitigation measures or Afghanistan. But people still notice higher prices no matter what, and it’s already hurting Trump:

Alas, stuff like “17 million people losing their health insurance” and “the Supreme Court erasing Article I from the Constitution” seems like an awfully, er, high price to pay to teach marginal Trump voters a lesson about voting for the Tariff Man because you’re upset about the cost of consumer goods.

…thanks to Trump people won’t be able to afford the G.I. Joe with the kung-fu grip:

The post Trump raise prices appeared first on Lawyers, Guns & Money.

The post Anniversary Trip appeared first on The Perry Bible Fellowship.

Earlier this month, Hollywood mourned the passing of Michael Madsen, a gifted actor best known for his critically acclaimed roles in Reservoir Dogs, Kill Bill, and Donnie Brasco, among others. Few obituaries have mentioned one of his lesser-known roles: a black ops mercenary hired to help hunt down an escaped human/alien hybrid in 1995's Species. The sci-fi thriller turns 30 this year, and while it garnered decidedly mixed reviews upon release, the film holds up quite well as a not-quite-campy B monster movie that makes for a great guilty pleasure.

(Many spoilers below.)

Screenwriter Dennis Feldman (The Golden Child) was partially inspired by an Arthur C. Clarke article discussing how the odds were slim that an extraterrestrial craft would ever visit Earth, given the great distances that would need to be traversed (assuming that traveling faster than the speed of light would be highly unlikely). Feldman was intrigued by the prospect of making extraterrestrial contact via information— specifically, alien instructions on how to build an instrument that could talk to terrestrial humans.

That instrument wouldn't be mechanical but organic, enabling an extraterrestrial visitor to adapt to Earth via combined DNA. Furthermore, rather than viewing projects like SETI or the Voyager missions—both of which sent transmissions containing information about Earth—as positive, Feldman considered them potentially dangerous, essentially inviting predators to target Earth's inhabitants. His alien would be a kind of bioweapon. The result was Species, which began as a spec script that eventually attracted the interest of MGM and director Roger Donaldson (The Bounty, No Way Out).

The premise is that the US government receives a response to the transmissions set into space: One message gives instructions on a new fuel source; the other contains explicit instructions on how to create an alien DNA sample and splice it with that of a human. Dr. Xavier Fitch (Ben Kingsley) is the scientist in charge of conducting the latter experiment, and the result is Sil (played as a young girl by Michelle Williams), a female alien/human hybrid they believed would have "docile and controllable" traits.

In just three months, Sil develops into a 12-year-old girl. But she starts exhibiting odd behavior as she sleeps, indicative of violent tendencies. Fitch decides to terminate the experiment, which means killing Sil by filling her containment cell with cyanide gas. A betrayed Sil breaks out of her cell and escapes. Fitch (who is the worst) puts together a crack team to track her down and eliminate her: mercenary Preston Lennox (Madsen); a molecular biologist named Dr. Laura Baker (a pre-CSI Marg Helgenberger); anthropologist Dr. Stephen Arden (Alfred Molina), and an "empath" named Dan Smithson (Forest Whitaker).

An experiment run amok

Sil won't be easy to find. Not only does she evade detection and hop on a train to Los Angeles, but she also transforms into a cocoon stage en route, emerging as a fully grown female (Natasha Henstridge) upon arrival. She's smart and resourceful, too—and very deadly when she feels her survival is threatened, which is often. The team must locate Sil before she manages to mate and produce equally rapid-developing offspring. At least they can follow all the bodies: a tramp on the train, a train conductor, a young woman in a nightclub, a rejected suitor, etc. Of course, she finally manages to mate—with an unsuspecting Arden, no less—and gives birth in the labyrinthine LA sewers, before she and her hybrid son meet their grisly demises.

One can only admire H.R. Giger's striking alien design; he wanted to create a monster who was "an aesthetic warrior, also sensual and deadly," and he very much delivered on that vision. He had also wanted several stages of development for Sil, but in the end, the filmmakers kept things simple, limiting themselves to the cocoon stage that shepherded young Sil through puberty and Sil's final alien maternal form with translucent skin—described as being "like a glass body but with carbon inside."

That said, Giger didn't much care for the final film. He thought it was much too similar to the Alien franchise, which boasts his most famous creature design, the xenomorph. For instance, there is the same punching tongue (Giger had wanted to incorporate barbed hooks for Sil), and Sil giving birth seems eerily akin to Alien's famous "chestburster" scene. Giger did manage to convince the director to have the team ultimately take out Sil with a fatal shot to the head rather than with flame-throwers, which he felt was too derivative of Alien 3 and Terminator 2: Judgment Day.

Giger had a point: Species is not particularly ground-breaking or original in terms of plot or the nature of the alien posing a threat to humankind. The dialogue is uninspired (occasionally downright trite) and the characters aren't well-developed, most notably Kingsley's weak-willed amoral scientist and Whitaker's reluctant empath—both exceptionally gifted actors who are largely wasted here. Poor Whitaker is reduced to looking broody and stating the obvious about whatever Sil might be "feeling." There are gestures toward themes that are never fully explored, and the outcome is predictable, right down to the final twist.

The mating game

But there's also plenty to like about Species. Madsen and Helgenberger give strong performances and have excellent on-screen chemistry; their sweetly awkward sex scene is the antithesis of Sil's far more brutal approach—in fact, Sil learns more about the subtleties of seduction by eavesdropping on the pair. And the film is well-paced, with all the right beats and memorable moments for a successful sci-fi thriller.

Former model Henstridge acquits herself just fine in her debut role. Much was made in the press of Henstridge's nude scenes, but while her beauty is used to great effect, it's the character of Sil and her journey that compels our attention the most, along with our shifting emotions toward her. Young Sil is sympathetic, the result of an unethical science experiment. She didn't ask to be born and has little control over what is happening to her. But she does want to live (hence her escape) and is genuinely scared when she begins to transform into her cocoon on the train.

Our sympathy is tested when adult Sil brutally kills a kindly train conductor, and then a romantic rival in a nightclub, both in a very gruesome manner. We might be able to rationalize the killing of the first rejected suitor, since he refuses to accept she's changed her mind about mating with him and gets rough. But nice guy John (Whip Hubley)? The woman she takes as hostage to fake her own death? Both offer to help Sil and die for their trouble.

Granted, Sil's distrust of humans is learned. She is being hunted by a team of professionals who intend to kill her, after all. When the woman hostage swears she won't harm Sil if she lets her go, Sil responds, "Yes you would. You just don't know it yet." We gradually realize that Sil is not that little girl any longer—if she ever was—but a ruthless creature driven entirely by instinct, even if she doesn't fully understand why she's been sent to Earth in the first place. As Laura notes, adult Sil views humans as disposable "intergalactic weeds." By the time we get to the showdown in the sewer, Sil isn't even in human form anymore, so the audience has no qualms about her eventual violent demise.

Species performed well enough at the box office to spawn multiple sequels—each one worse than the last— an adapted novel, and a Dark Horse Comics series. None of them captured the unique combination of elements that lifted the original above its various shortcomings. It will never match Alien, but Species is nonetheless an entertaining ride.

Read full article

Comments

Nvidia is recommending a mitigation for customers of one of its GPU product lines that will degrade performance by up to 10 percent in a bid to protect users from exploits that could let hackers sabotage work projects and possibly cause other compromises.

The move comes in response to an attack a team of academic researchers demonstrated against Nvidia’s RTX A6000, a widely used GPU for high-performance computing that’s available from many cloud services. A vulnerability the researchers discovered opens the GPU to Rowhammer, a class of attack that exploits physical weakness in DRAM chip modules that store data.

Rowhammer allows hackers to change or corrupt data stored in memory by rapidly and repeatedly accessing—or hammering—a physical row of memory cells. By repeatedly hammering carefully chosen rows, the attack induces bit flips in nearby rows, meaning a digital zero is converted to a one or vice versa. Until now, Rowhammer attacks have been demonstrated only against memory chips for CPUs, used for general computing tasks.

Like catastrophic brain damage

That changed last week as researchers unveiled GPUhammer, the first known successful Rowhammer attack on a discrete GPU. Traditionally, GPUs were used for rendering graphics and cracking passwords. In recent years, GPUs have become the workhorses for tasks such as high-performance computing, machine learning, neural networking, and other AI uses. No company has benefited more from the AI and HPC boom than Nvidia, which last week became the first company to reach a $4 trillion valuation. While the researchers demonstrated their attack against only the A6000, it likely works against other GPUs from Nvidia, the researchers said.

The researchers’ proof-of-concept exploit was able to tamper with deep neural network models used in machine learning for things like autonomous driving, healthcare applications, and medical imaging for analyzing MRI scans. GPUHammer flips a single bit in the exponent of a model weight—for example in y, where a floating point is represented as x times 2^y. The single bit flip can increase the exponent value by 16. The result is an altering of the model weight by a whopping 2¹⁶, degrading model accuracy from 80 percent to 0.1 percent, said Gururaj Saileshwar, an assistant professor at the University of Toronto and co-author of an academic paper demonstrating the attack.

“This is like inducing catastrophic brain damage in the model: with just one bit flip, accuracy can crash from 80% to 0.1%, rendering it useless,” Saileshwar wrote in an email. “With such accuracy degradation, a self-driving car may misclassify stop signs (reading a stop sign as a speed limit 50 mph sign), or stop recognizing pedestrians. A healthcare model might misdiagnose patients. A security classifier may fail to detect malware.”

In response, Nvidia is recommending users implement a defense that could degrade overall performance by as much as 10 percent. Among machine learning inference workloads the researchers studied, the slowdown affects the “3D U-Net ML Model” the most. This model is used for an array of HPC tasks, such as medical imaging.

The performance hit is caused by the resulting reduction in bandwidth between the GPU and the memory module, which the researchers estimated as 12 percent. There’s also a 6.25 percent loss in memory capacity across the board, regardless of the workload. Performance degradation will be the highest for applications that access large amounts of memory.

A figure in the researchers’ academic paper provides the overhead breakdowns for the workloads tested.

Rowhammer attacks present a threat to memory inside the typical laptop or desktop computer in a home or office, but most Rowhammer research in recent years has focused on the threat inside cloud environments. That's because these environments often allot the same physical CPU or GPU to multiple users. A malicious attacker can run Rowhammer code on a cloud instance that has the potential to tamper with the data a CPU or GPU is processing on behalf of a different cloud customer. Saileshwar said that Amazon Web Services and smaller providers such as Runpod and Lambda Cloud all provide A6000s instances. (He added that AWS enables a defense that prevents GPUhammer from working.)

Not your parents’ Rowhammer

Rowhammer attacks are difficult to perform for various reasons. For one thing, GPUs access data from GDDR (graphics double data rate) physically located on the GPU board, rather than the DDR (double data rate) modules that are separate from the CPUs accessing them. The proprietary physical mapping of the thousands of banks inside a typical GDDR board is entirely different from their DDR counterparts. That means that hammering patterns required for a successful attack are completely different. Further complicating attacks, the physical addresses for GPUs aren’t exposed, even to a privileged user, making reverse engineering harder.

GDDR modules also have up to four times higher memory latency and faster refresh rates. One of the physical characteristics Rowhammer exploits is that the increased frequency of accesses to a DRAM row disturbs the charge in neighboring rows, introducing bit flips in neighboring rows. Bit flips are much harder to induce with higher latencies. GDDR modules also contain proprietary mitigations that can further stymie Rowhammer attacks.

In response to GPUhammer, Nvidia published a security notice last week reminding customers of a protection formally known as system-level error-correcting code. ECC works by using what are known as memory words to store redundant control bits next to the data bits inside the memory chips. CPUs and GPUs use these words to quickly detect and correct flipped bits.

GPUs based on Nvidia’s Hopper and Blackwell architectures already have ECC turned on. On other architectures, ECC is not enabled by default. The means for enabling the defense vary by the architecture. Checking the settings in Nvidia GPUs designated for data centers can be done out-of-band using a system’s BMC (baseboard management controller) and software such as Redfish to check for the “ECCModeEnabled” status. ECC status can also be checked using an in-band method that uses the system CPU to probe the GPU.

The protection does come with its limitations, as Saileshwar explained in an email:

On NVIDIA GPUs like the A6000, ECC typically uses SECDED (Single Error Correction, Double Error Detection) codes. This means Single-bit errors are automatically corrected in hardware and Double-bit errors are detected and flagged, but not corrected. So far, all the Rowhammer bit flips we detected are single-bit errors, so ECC serves as a sufficient mitigation. But if Rowhammer induces 3 or more bit flips in a ECC code word, ECC may not be able to detect it or may even cause a miscorrection and a silent data corruption. So, using ECC as a mitigation is like a double-edged sword.

Saileshwar said that other Nvidia chips may also be vulnerable to the same attack. He singled out GDDR6-based GPUs in Nvidia’s Ampere generation, which are used for machine learning and gaming. Newer GPUs, such as the H100 (with HBM3) or RTX 5090 (with GDDR7), feature on-die ECC, meaning the error detection is built directly into the memory chips.

“This may offer better protection against bit flips,” Saileshwar said. “However, these protections haven’t been thoroughly tested against targeted Rowhammer attacks, so while they may be more resilient, vulnerability cannot yet be ruled out.”

In the decade since the discovery of Rowhammer, GPUhammer is the first variant to flip bits inside discrete GPUs and the first to attack GDDR6 GPU memory modules. All attacks prior to GPUhammer targeted CPU memory chips such as DDR3/4 or LPDDR3/4.

That includes this 2018 Rowhammer variant. While it used a GPU as the hammer, the memory being targeted remained LPDDR3/4 memory chips. GDDR forms of memory have a different form factor. It follows different standards and is soldered onto the GPU board, in contrast to LPDDR, which is in a chip located on hardware apart from the CPUs.

Besides Saileshwar, the researchers behind GPUhammer include Chris S. Lin and Joyce Qu from the University of Toronto. They will be presenting their research next month at the 2025 Usenix Security Conference.

Read full article

Comments