Starlink operator SpaceX claims that Amazon violated orbital debris requirements by launching satellites into initial altitudes that are too high, increasing the risk of collision with other satellites and spacecraft. SpaceX, which recently reported two Starlink satellite failures that created new space debris, yesterday accused Amazon and its launch partner Arianespace of negligence that "needlessly and significantly increases risk to other operational systems and inhabited spacecraft."

Amazon Leo, formerly known as Kuiper Systems, is launching satellites into low-Earth orbits (LEO) to compete against Starlink's much larger constellation of broadband satellites. Amazon denied that its launch altitudes violate any requirements or impose a safety risk and said SpaceX itself helped Amazon launch satellites into a similar altitude last year when Amazon used SpaceX as a launch partner.

SpaceX only objected to the launch parameters after moving its Starlink satellites into nearby altitudes, Amazon said. Changing the altitude of a recent Leo launch would have delayed it by months, according to Amazon. Both Amazon and SpaceX have accused each other of using FCC proceedings to delay the other's satellite launches at various times over the years.

Read full article

Comments

OpenAI might be pulling back on video generation, but Google is forging ahead with a major AI update to its Vids editing product. The company's latest video and audio models are now integrated with the tool, and you can choose from various controllable avatars to appear in generated videos. Your creations are also easier to share on YouTube now.

Veo 3.1 is the biggest part of the Vids upgrade. Google first deployed this updated model in Gemini late last year , promising a substantial improvement in realism and consistency. While Google has pitched Veo as a tool for filmmakers, that's not how it positions Vids. Google suggests using the AI tools in Vids to create animated party flyers, business sizzle reels, or a video greeting card. You can use Vids for free, but you won't be able to generate very many videos without an AI subscription.

If you're not paying for any AI access on your account, you only get 10 video generations per month. AI Pro subscribers can get 50 videos, and those paying for Google's spendy AI Ultra plan (either personal or enterprise) get 1,000 videos per month. Like most other Veo implementations, the videos are eight seconds long and 720p resolution.

Read full article

Comments

Octopuses are one of the most alien creatures on Earth. The lack of bones makes them amazing shapeshifters, most of them can change color like chameleons, and they pump blue copper-based blood through their bodies using three distinct hearts. They rely on a decentralized nervous system, where two-thirds of their neurons reside in their arms, allowing each limb to independently taste, touch, and make decisions for itself.

Now, a team of scientists led by Pablo S. Villar, a molecular biologist at Harvard University, for the first time took a close look at octopuses' sex life. It turned out it was just as weird.

Love in the dark

The deep ocean is a challenging place to find a partner, especially since octopuses are solitary animals that wander the seafloor alone, mating only during highly infrequent encounters. The exact mechanics of their reproduction when they do find each other have long puzzled biologists. We knew that male octopuses don't rely on flashy plumage or complex mating calls and that they use a specialized appendage called the hectocotylus—basically a modified tentacle—to identify females.

Read full article

Comments

The details of how animal life began are a bit murky. Most of the groups familiar today are present in the Cambrian, a period when they rapidly diversified, with familiar features evolving alongside bizarre creatures with no obvious modern equivalents. There are hints that some forms of present animal life predated the Cambrian. But most of the organisms we've found in Ediacaran deposits have no obvious relationship to anything we're familiar with.

The complete absence of these creatures in later strata suggest they might have vanished in a mass-extinction event that cleared the way for the explosion of Cambrian species. But a new series of fossils found at a site in China includes examples of groups that flourished in the Cambrian living side-by-side with a few Ediacaran species. The deposits suggest that there might have been a gradual shift into the Cambrian.

Ediacaran and more

The newly described fossils, described by a team from Yunnan University and Oxford University, come from just south of Kunming, near Fuxian Lake. The rocks they're in are part of the larger Dengying Formation, within a segment that's known to include deposits from the Edicaran, which ranged from 635 to 540 million years ago. They come from close to the end of the period, only about 7 million years before the first clearly Cambrian deposits.

Read full article

Comments

The cost of high-performance GPUs, typically $8,000 or more, means they are frequently shared among dozens of users in cloud environments. Two new attacks demonstrate how a malicious user can gain full root control of a host machine by performing novel Rowhammer attacks on high-performance GPU cards made by Nvidia.

The attacks exploit memory hardware’s increasing susceptibility to bit flips, in which 0s stored in memory switch to 1s and vice versa. In 2014 , researchers first demonstrated that repeated, rapid access—or “hammering”—of memory hardware known as DRAM creates electrical disturbances that flip bits. A year later , a different research team showed that by targeting specific DRAM rows storing sensitive data, an attacker could exploit the phenomenon to escalate an unprivileged user to root or evade security sandbox protections. Both attacks targeted DDR3 generations of DRAM.

From CPU to GPU: Rowhammer's decade-long journey

Over the past decade, dozens of newer Rowhammer attacks have evolved to, among other things:

Read full article

Comments

On Wednesday, the International Renewable Energy Agency (IRENA) released its numbers on what was built in 2025. And much as we saw in the US , solar power is the primary driver of change. The numbers show that the world installed an average of 1.4 gigawatts of solar capacity every day last year, for a total of 511 GW. That brings the total solar capacity up to 2.4 Terawatts, making it the largest single source of renewable capacity by far, at more than a Terawatt above either wind or hydro.

Obviously, the actual power generated will be less than the rated capacity. And because solar panels have become so cheap, the economics now favor installing panels in areas that get far less sunlight—places in which photovoltaics would have been a questionable decision just five years earlier. So we're likely to see the energy produced for each unit of capacity (termed the capacity factor) decline over the coming years.

How much of a factor is that? 2025's power-generation numbers are not yet available, but data from 2024 shows photovoltaics generating 7 percent of the world's power, with wind at 8 percent and nuclear at 9. That's despite having 1.9 times as much solar capacity as wind capacity. Still, despite the lower capacity factor, solar is catching up fast. As these numbers don't include concentrated solar power or last year's data, it's possible that solar has already become the second-largest source of carbon-free electricity (after hydropower). If not, we're certain to see that happen before the decade is out.

Read full article

Comments

Google's Gemini AI models have improved by leaps and bounds over the past year, but you can only use Gemini on Google's terms. The company's Gemma open-weight models have provided more freedom, but Gemma 3, which launched over a year ago , is getting a bit long in the tooth. Starting today, developers can start working with Gemma 4 , which comes in four sizes optimized for local usage. Google has also acknowledged developer frustrations with AI licensing, so it's dumping the custom Gemma license.

Like past versions of its open-weight models, Google has designed Gemma 4 to be usable on local machines. That can mean plenty of things, of course. The two large Gemma variants, 26B Mixture of Experts and 31B Dense, are designed to run unquantized in bfloat16 format on a single 80GB Nvidia H100 GPU. Granted, that's a $20,000 AI accelerator, but it's still local hardware. If quantized to run at lower precision, these big models will fit on consumer GPUs.

Google also claims it has focused on reducing latency to really take advantage of Gemma's local processing. The 26B Mixture of Experts model activates only 3.8 billion of its 26 billion parameters in inference mode, giving it much higher tokens-per-second than similarly sized models. Meanwhile, 31B Dense is more about quality than speed, but Google expects developers to fine-tune it for specific uses.

Read full article

Comments

When it comes to automotive bragging rights, a good Nürburgring Nordschleife lap time is right up there with the best of them. And today, those bragging rights belong to Ford. The automaker revealed that its GT Mk IV, an evolution of the mid-engined supercar it created in 2016, is now the fastest production car to ever lap the 12.9-mile (20.8-km) race track in Germany, with a time of 6 minutes, 15.997 seconds set by Frédéric Vervisch.

The century-old racetrack in Germany's Eifel region was built during the Great Depression as a way to create jobs but also to provide Germany's car industry with a place to test its products. In addition to races, it was—and remains—open to the public for leisure driving.

Well, for some definition of leisure: The place isn't known as the Green Hell for nothing, with hundreds of feet of elevation change across 12.9 miles (20.8 km) and between 73–170 corners, depending how you count them. After years of driving it online, I got my first laps there last summer and can report that in real life, it is bumpy and narrow, and I'd need another hundred laps or so before I started to feel properly comfortable.

Read full article

Comments

An Anthropic-backed DMCA effort to remove its recently leaked Claude Code client source code from GitHub this week resulted in the accidental removal of many legitimate forks of its official public code repository. While that overzealous takedown has now been reversed, Anthropic still faces an extreme uphill battle in limiting the spread of its recently leaked code.

The DMCA notice that GitHub received late Tuesday focuses on a repository containing the leaked source code originally posted by GitHub user nirholas ( archived here ) and nearly 100 specifically named forks of that repository. In a note appended to that request, though, GitHub said it had acted to take down a network of 8,100 similar forked repositories because "the submitter alleged that all or most of the forks were infringing to the same extent as the parent repository."

That expanded takedown affected many repositories that didn't contain leaked code but instead forked Anthropic's official public Claude Code repository , which the company shares to encourage public bug reports and fixes . Many coders took to social media to complain about being swept up in the DMCA dragnet despite not sharing any leaked code.

Read full article

Comments