Ars Technica Mar 4, 07:17 PM
After a rocky six years, Sony cancels future single-player PC game releases Titles like Ghost of Yotei will remain exclusive to Sony's hardware.
Ars Technica Mar 4, 06:44 PM
MacBook Neo hands-on: Apple build quality at a substantially lower price The Neo won't be for everyone, but Apple has managed to preserve a premium feel.
Ars Technica Mar 4, 07:00 PM
Re-creating the complex cuisine of prehistoric Europeans SEM analysis of pottery residues showed people combined fish with a wide variety of plants when cooking.
Ars Technica Mar 4, 06:50 PM
The US Senate empowers NASA to fully engage in lunar space race "Our bill authorizes critical funding for, and gives strategic direction to, the agency."
VentureBeat Mar 4, 08:18 PM
Black Forest Labs' new Self-Flow technique makes training multimodal AI models 2.8x more efficient To create coherent images or videos, generative AI diffusion models like Stable Diffusion or FLUX have typically relied on external "teachers"—frozen encoders like CLIP or DINOv2—to provide the semantic understanding they couldn't learn on their own.
But this reliance has come at a cost: a "bottleneck" where scaling up the model no longer yields better results because the external teacher has hit its limit.
Today, German AI startup Black Forest Labs (maker of the FLUX series of AI image models) has announced a potential end to this era of academic borrowing with the release of Self-Flow, a self-supervised flow matching framework that allows models to learn representation and generation simultaneously.
By integrating a novel Dual-Timestep Scheduling mechanism, Black Forest Labs has demonstrated that a single model can achieve state-of-the-art results across images, video, and audio without any external supervision.
The technology: breaking the "semantic gap"
The fundamental problem with traditional generative training is that it's a "denoising" task. The model is shown noise and asked to find an image; it has very little incentive to understand what the image is, only what it looks like.
To fix this, researchers have previously "aligned" generative features with external discriminative models. However, Black Forest Labs argues this is fundamentally flawed: these external models often operate on misaligned objectives and fail to generalize across different modalities like audio or robotics.
The Labs' new technique, Self-Flow, introduces an "information asymmetry" to solve this. Using a technique called Dual-Timestep Scheduling, the system applies different levels of noise to different parts of the input. The student receives a heavily corrupted version of the data, while the teacher—an Exponential Moving Average (EMA) version of the model itself—sees a "cleaner" version of the same data.
The student is then tasked not just with generating the final output, but with predicting what its "cleaner" self is seeing—a process of self-distillation where the teacher is at layer 20 and the student is at layer 8. This "Dual-Pass" approach forces the model to develop a deep, internal semantic understanding, effectively teaching itself how to see while it learns how to create.
Product implications: faster, sharper, and multi-modal
The practical results of this shift are stark. According to the research paper, Self-Flow converges approximately 2.8x faster than the REpresentation Alignment (REPA) method, the current industry standard for feature alignment. Perhaps more importantly, it doesn't plateau; as compute and parameters increase, Self-Flow continues to improve while older methods show diminishing returns.
The leap in training efficiency is best understood through the lens of raw computational steps: while standard "vanilla" training traditionally requires 7 million steps to reach a baseline performance level, REPA shortened that journey to just 400,000
Wired Mar 4, 10:00 PM
What AI Models for War Actually Look Like While companies like Anthropic debate limits on military uses of AI, Smack Technologies is training models to plan battlefield operations.
The Verge Mar 5, 02:21 AM
Did Live Nation punish a venue by taking Billie Eilish away? John Abbamondi had orders to let the CEO of Ticketmaster down easy.
In April 2021, Abbamondi was the CEO of BSE Global, the company that ran Brooklyn arena the Barclays Center. BSE Global's existing Ticketmaster contract would expire at the end of September, and Abbamondi and his team had evaluated proposals from SeatGeek, AXS, and Ticketmaster. The economics of Ticketmaster offer, according to Abbamondi, "was nowhere near as good as the other two." SeatGeek's technology was "superior" to Ticketmaster's on balance, on top of better financial terms including an equity stake in the company, the arena decided. It clinched their decision to go …
Read the full story at The Verge.
Ars Technica Mar 4, 10:14 PM
Large genome model: Open source AI trained on trillions of bases System can identify genes, regulatory sequences, splice sites, and more.
Ars Technica Mar 4, 10:54 PM
TerraPower gets OK to start construction of its first nuclear plant Plant won't be done until 2030 at the earliest, and it still needs an operating license.
Ars Technica Mar 4, 10:32 PM
Space Command chief throws cold water on the question of UAPs in space "I am not aware of anything that is extraterrestrial, other than comets and things like that."
The Verge Mar 4, 11:52 PM
Tim Sweeney signed away his right to criticize Google until 2032 Epic CEO Tim Sweeney might be one of the most outspoken people in the history of the world. He fought two of the world's most valuable and powerful companies almost all the way to the US Supreme Court, insulting them again and again: "crooked," "deceitful," "insanely sneaky," calling Android a "fake open platform," calling both companies "gangster-style businesses that will do anything they think they can get away with," telling me how Google's Project Hug was "an astonishingly corrupt effort at a massive scale."
But Google has finally muzzled Tim Sweeney. It's right there in a binding term sheet for his settlement with Google.
On March 3 …
Read the full story at The Verge.
The Verge Mar 5, 01:09 AM
A new video from the White House mixes Call of Duty footage with actual video of Iran strikes A screenshot of the Call of Duty footage in the White House’s video.
On Wednesday, the White House posted a video of actual military strikes on Iran in the style usually seen in Call of Duty highlight videos, and started the video with a clip from Call of Duty. The real-life footage of missiles and other munitions hitting targets in Iran shows clips seen in other Trump administration videos, like this one posted to the U.S. Central Command X account.
Courtesy of the Red, White & Blue. pic.twitter.com/kTO0DZ56IJ
- The White House (@WhiteHouse) March 4, 2026
As noted by The Washington Post's Drew Harwell, the animation at the start appears to be from Call of Duty: Modern Warfare III when a player activates a …
Read the full story at The Verge.
Bloomberg Technology Mar 7, 03:34 AM
Nintendo Files Lawsuit Seeking Refunds on Trump’s Tariffs Nintendo Co. filed a lawsuit in the US Court of International Trade on Friday, seeking a refund for duties it paid to the US government as a result of President Donald Trump’s Tariffs.
CNBC Technology Mar 7, 12:03 AM
5 unresolved questions hanging over the Anthropic–Pentagon fracas: 'It's all very puzzling' Plenty of unknowns remain after Defense Secretary Pete Hegseth designated Anthropic a "Supply-Chain Risk to National Security" on Friday.
Bloomberg Technology Mar 6, 11:16 PM
Oracle, OpenAI End Plans to Expand Flagship Data Center Oracle and OpenAI have scrapped plans to expand a flagship artificial intelligence data center in Texas after negotiations dragged over financing and OpenAI’s changing needs. The collapsed talks created an opening for Meta Platforms Inc. to step in and consider leasing the planned expansion site in Abilene, Texas, from developer Crusoe, according to people familiar with the matter. Nvidia Corp., the leading AI chipmaker, helped facilitate Meta’s discussions with the developer, said the people, who asked not to be identified because the talks are private. The shifting plans underscore the complexity of building out AI data centers, which are expected to cost in the tens of billions of dollars and require cooperation from a wide swath of partners. Bloomberg Tech Co-Host Ed Ludlow joins Bloomberg Businessweek Daily to discuss. He speaks with Carol Massar and Tim Stenovec. (Source: Bloomberg)
TechCrunch Mar 6, 10:41 PM
X is testing a new ad format that connects posts with products An ad test on X promotes Musk's Starlink beneath original content.
TechCrunch Mar 6, 10:35 PM
Nintendo sues the US government for a refund on tariffs This lawsuit comes after a Supreme Court decision struck down some of the president's sweeping tariffs, which had impacted Nintendo and thousands of other companies.
Bloomberg Technology Mar 7, 12:46 AM
Musk’s Spam Frustration Was ‘Confusing’ to Ex-Twitter Executives Two former top executives at Twitter Inc. sought to beat back Elon Musk’s narrative at a jury trial that they lied to him about the makeup of the platform’s user base when he was purchasing the company in 2022.
Bloomberg Technology Mar 7, 01:48 AM
Rebellions Eyes Competition With Nvidia, AMD in AI Chips Rebellions CEO Sunghyun Park discusses how his AI semiconductor startup plans to compete with the likes of Nvidia and AMD in the AI chip space. He speaks with Haslinda Amin from the sidelines of 'IMF Conference; Asia 2050'. (Source: Bloomberg)
CNBC Technology Mar 7, 12:01 AM
Anthropic CEO says 'no choice' but to challenge Trump admin's supply chain risk designation in court Anthropic said even with the designation, the government can't forbid it from working with companies in other capacities.
CNBC Technology Mar 7, 12:00 AM
Anthropic officially told by DOD that it's a supply chain risk even as Claude used in Iran The formal declaration will require defense vendors and contractors to certify that they don’t use Anthropic’s models in their work with the Pentagon.
The Verge Mar 7, 02:59 AM
DJI will pay $30K to the man who accidentally hacked 7,000 Romo robovacs The DJI Romo robot vacuums. | Image: DJI
On Valentine's Day, I brought you a story that's since made headlines all around the world: How one man, just trying to steer his DJI robot vacuum with a PlayStation gamepad, discovered an entire network of 7,000 remote-control DJI robots ready to let him peek into other people's homes.
To be clear, DJI had already begun addressing some of the related vulnerabilities before the man, Sammy Azdoufal, showed The Verge just how much he could access. But it wasn't clear whether DJI would pay him for his discovery, particularly after how it treated security researcher Kevin Finisterre back in 2017 - or how soon DJI might fully patch the additiona …
Read the full story at The Verge.
Bloomberg Technology Mar 6, 11:33 PM
Blue Owl Leads $750 Million Debt Deal for Vista Nexthink Buyout Blue Owl Capital Inc. led a $750 million debt financing for Vista Equity Partners’ buyout of Nexthink SA, underscoring the private lender’s continued appetite for software deals even as liquidity concerns and unease over AI-driven disruption weigh on the market.
Bloomberg Technology Mar 7, 03:53 AM
China Turns to AI as Jobseeker Numbers Top Belgium’s Population China will harness artificial intelligence to create jobs as the number of university graduates poised to enter the labor market this year will reach 12.7 million — more than the population of countries such as Belgium.