Agregátor RSS

No warning, no opt-out, and critic claims ... no consent

Research from a leading academic shows Android users have advertising cookies and other gizmos working to build profiles on them even before they open their first app.…

Threat hunters are calling attention to a new highly-targeted phishing campaign that singled out "fewer than five" entities in the United Arab Emirates (U.A.E.) to deliver a previously undocumented Golang backdoor dubbed Sosano. The malicious activity was specifically directed against aviation and satellite communications organizations, according to Proofpoint, which detected it in late October Ravie Lakshmananhttp://www.blogger.com/profile/[email protected]

V pondělí v 5 odpoledne kalifornského času zveřejnil Tim Cook krátký teaser s textem „Něco je ve vzduchu“. Verzálkami psaný nápis AIR pak protne něco, co by se dalo považovat za tenké hliníkové zařízení. A internet začal řešit, co to všechno znamená. [TWITTER:id='1896589954517701057'||] Zkrátím ...

Chinese AI startup DeepSeek has claimed its V3 and R1 models achieve a theoretical daily cost-profit ratio of 545%, highlighting cost implications for enterprises adopting similar models from other cloud providers.

In a GitHub post published over the weekend, DeepSeek estimated its daily inference cost for V3 and R1 models at $87,072, assuming a $2 per hour rental for Nvidia’s H800 chips.

Theoretical daily revenue was pegged at $562,027, implying a 545% cost-profit ratio and over $200 million in potential annual revenue.

[ Related: More DeepSeek news and analysis ]

However, the company noted actual earnings are significantly lower due to free web and app access, lower V3 model costs, and discounted developer rates during off-peak hours.

This is the first time the Hangzhou-based company has disclosed profit margins from inference tasks, where trained AI models generate responses or perform functions like chatbot interactions.

Authenticity and enterprise impact

Analysts say DeepSeek’s focus on scalability and efficiency is notable, but caution that it is too early to view its claims as an industry benchmark applicable to companies in or outside China.

“Also, in theory versus practice, there is a significant difference, as cost metrics are also highly subjective to geography, resources, and revenue generation,” said Neil Shah, partner & co-founder at Counterpoint Research. “However, we don’t know the purpose of these public claims, but they will definitely put pressure on Western companies to at least reveal and/or internally optimize their costs.”

If accurate, DeepSeek’s profitability despite deep discounts would signal a sustainable low-cost AI model, potentially pressuring rivals to cut prices while prompting enterprises to reassess vendor choices and long-term AI strategies.

“There are no US-based AI firms of scale that are profitable right now,” said Hyoun Park, CEO and chief analyst at Amalgam Insights. “Open AI is not even close and both Microsoft and Google are spending billions of dollars to enter the market.”

Park noted that while DeepSeek’s figures are theoretical and difficult to verify, one thing is clear – DeepSeek has massively reduced the cost of inference.

“DeepSeek’s AI models, when hosted on established cloud platforms such as AWS and Microsoft Azure, can offer enterprises a balance of performance, governance, and affordability,” said Abhiram Srivasta, senior analyst at Everest Group. “These models are reportedly more cost-efficient than those from leading US AI firms, requiring significantly less compute power, which translates to lower operational costs.” 

Threat to US companies

DeepSeek’s claims of cost efficiency and its open-source approach could intensify competition in the AI market, particularly for US firms investing heavily in proprietary models.

The company currently offers its models as open source, allowing US-based enterprises to audit and modify them.

As long as the deployment does not rely on Chinese-hosted infrastructure, there may not be any significant barriers to global adoption. “Given that many current models are good enough for established generative AI use cases, DeepSeek is absolutely a threat to US based AI model builders,” Park said. “AI developers, focusing on theoretical artificial general intelligence are likely to be quickly surpassed by those making more practical agentic models that can get work done and provide interaction visibility.” 

Internet service providers (ISPs) in China and the West Coast of the United States have become the target of a mass exploitation campaign that deploys information stealers and cryptocurrency miners on compromised hosts. The findings come from the Splunk Threat Research Team, which said the activity also led to the delivery of various binaries that facilitate data exfiltration as well as offer Ravie Lakshmananhttp://www.blogger.com/profile/[email protected]

Snahu (ať už jakýmkoli způsobem) poskytnout americké továrny Intelu společnosti TSMC vnímají bývalí ředitelé Intelu jako velmi zásadní chybu a navrhují jiné řešení…

Tlakoměr Omron M2 Intelli IT+ stojí v průměru okolo 1500 Kč, ale momentálně jej Alza , Datart a Pilulka nabízejí za 1299 Kč . Na trhu jsou i levnější alternativy, ale ve prospěch Omronu hraje dobrá reputace japonského výrobce, který profi vybavení dodává i do ordinací. Na tento model je navíc ...

Says the biz trying to sell us stuff to catch that, admittedly

High-profile deepfake scams that were reported here at The Register and elsewhere last year may just be the tip of the iceberg. Attacks relying on spoofed faces in online meetings surged by 300 percent in 2024, it is claimed.…

** S fotografickými koncepty se v posledních dnech roztrhl pytel ** Xiaomi to zkouší s vyměnitelnými objektivy, které berou energii z telefonu ** Přicvaknou se přes MagSafe, s telefonem komunikují přes laser

Představte si erupci sopky tak mohutnou, že vyvrhne do atmosféry stovky megatun síry a popela, zastíní Slunce a ochladí planetu. Přesně tak vypadal výbuch supervulkánu Los Chocoyos v dnešní Guatemale. Byla to jedna z největších sopečných událostí čtvrtohor. Vědci z Univerzity St Andrews díky ...

Stížnosti uživatelů a recenzentů na černé obrazovky a pády karet řady GeForce RTX 5000 komentovala Nvidia příslibem ovladačů, které problém opraví. Skutečně vyšly…

** Koncept telefonu od Infinixu se umí nabíjet ze slunce ** Maximální dobíjecí výkon je 2 watty ** Další technologie zase změní barvu zad podle nastavené tapety

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) on Monday added five security flaws impacting software from Cisco, Hitachi Vantara, Microsoft Windows, and Progress WhatsUp Gold to its Known Exploited Vulnerabilities (KEV) catalog, based on evidence of active exploitation. The list of vulnerabilities is as follows - CVE-2023-20118 (CVSS score: 6.5) - A command injection [email protected]

Google has released its monthly Android Security Bulletin for March 2025 to address a total of 44 vulnerabilities, including two that it said have come under active exploitation in the wild. The two high-severity vulnerabilities are listed below - CVE-2024-43093 - A privilege escalation flaw in the Framework component that could result in unauthorized access to "Android/data," "Android/obb," Ravie Lakshmananhttp://www.blogger.com/profile/[email protected]

Enhancing API security for financial institutions

Partner Content  Open banking has revolutionized financial services, empowering consumers to share their financial data with third-party providers, including fintech innovators.…

Mixed messages from Pentagon, CISA as Trump gets pally with Putin and Kremlin strikes US critical networks

Comment  America's cybersecurity chiefs in recent days have been sending mixed messages about the threat posed by Russia in the digital world.…

The device also mimics lemonade and coffee—but fried eggs? Not so much.

“That Cajun blackened shrimp recipe looks really good,” I tell my husband while scrolling through cooking videos online. The presenter describes it well: juicy, plump, smoky, a parade of spices. Without making the dish, I can only imagine how it tastes. But a new device inches us closer to recreating tastes from the digital world directly in our mouths.

Smaller than a stamp, it contains a slurry of chemicals representing primary flavors like salty, sweet, sour, bitter, and savory (or umami). The reusable device mixes these together to mimic the taste of coffee, cake, and other foods and drinks.

Developed by researchers at Ohio State University, the device has a tiny gum-like strip linked to a liquid reservoir. It releases each taste component in a gel and pumps the resulting blend onto the tongue. The system is wireless and includes a sensor to control the chemical mixture. In a demonstration, one person dipped the sensor into some lemonade in San Francisco and transferred a facsimile of the taste to people wearing the devices in Ohio in real-time.

Complex flavor profiles—say, a fried egg—are harder to simulate. And it’s likely awkward to have a device dangling on your mouth. But the work brings us a little closer to adding a new sense to virtual and augmented reality and making video games more immersive.

“This will help people connect in virtual spaces in never-before-seen ways,” study author Jinghua Li said in a press release. “This concept is here, and it is a good first step to becoming a small part of the metaverse.”

Gaming aside, future iterations of the device could potentially help people who have lost their sense of taste, including those living with long Covid or traumatic brain injuries.

What’s Taste, Anyways?

We can taste food thanks to a variety of chemicals stimulating our taste buds. There are five main types of taste bud, each specializing in a different taste. When we chew food, our taste buds send electrical signals to the brain where they combine into a host of flavors—the bitterness of coffee, tanginess of a cup of orange juice, or richness of a buttery croissant.

But taste isn’t an isolated sensation. Smells, textures, memories, and emotions also come into play. One spoon of comfort food can take you back to happy days as a child. That magic is hard to replicate with a few spurts of chemical flavor and is partly why taste is so hard to recreate in digital worlds, wrote the team.

Virtual and augmented reality have mainly focused on audio and visual cues. Adding smell or taste could make experiences more immersive. An early version of the idea, dubbed Smell-O-Vision, dates back nearly a century when scents were released in theaters to heighten the film experience. It’s still employed in 4DX theaters today.

Cinema isn’t the only industry looking for a multi-sensory upgrade. At this year’s CES, a trailer for Sony’s hit game, The Last of Us, showed the technology at work in an immersive, room-size version of the game where players could smell the post-apocalyptic world.

Taste is harder to recreate. Older methods activated taste buds with electrical zaps to the tongue. While participants could detect very basic tastes, hooking your tongue up to electrodes isn’t the most comfortable setup.

More recently, a Hong Kong team developed a lollipop-like device that produces nine tastes embedded in food-safe gels. An electrical zap releases the chemicals, and upping the voltage delivers a stronger flavor. The approach is an improvement, but holding a lollipop in your mouth while gaming for hours is still awkward.

Tasty Interface

The new device offers a neater solution. Dubbed e-Taste, it has two main components: a sensing platform to analyze the taste profile of a food or drink and an actuator to deliver a mixture of liquid chemicals approximating the sampled taste.

The actuator, a cube the size of a shirt button and a gum-like strip, hangs on the lower teeth. The cube stores chemicals mimicking each of five tastes—glucose for sweet and citric acid for sour, for example—in separate chambers. A tiny pump, activated by an electrical zap, pushes the liquids onto a gel strip where they mingle before being pumped onto the tongue. Each pump is the equivalent of a drop of water, which is enough to activate taste buds.

A person using the device holds the strip inside their mouth with the cube dangling outside. Once the sensor captures a food or drink’s flavor profile—say, equal amounts of sweet, sour, salty, and savory—it wirelessly transmits the data to the actuator which releases the final taste mixture for roughly 45 minutes—plenty of time to experience a virtual foodie session.

Eat Digital Cake

After training e-Taste to understand which chemical mixtures best approximate various foods, the team asked 10 volunteers to name the food they were tasting from a list of possibilities.

Roughly 90 percent could pick out lemonade and gauge its sourness. Most could also identify the taste of cake. But not all foods were so easily mimicked. Participants struggled to name umami-heavy dishes, such as fried eggs or fish stew.

Rather than a bug in the device, however, this is an expected result. Taste is highly subjective. Our tolerance to spice or sourness varies largely.

Then there’s the weirdness of a virtual setup. We eat and drink with our eyes open and smell food too. One participant said that tasting coffee through the device without seeing a normal coffee maker led to some confusion. Scientists have long known the color of food is essential to our perception of flavor. Smell and texture are also crucial. The smell of a good southern barbeque joint sets expectations—even before we’ve tasted anything.

The team is exploring ways to enhance the experience by adding these senses. Shrinking the device is also on the menu.

Although the team developed e-Taste to enhance gaming, people could use something like it to sample food across the globe or items when grocery shopping online. Doctors could use it to detect if people have lost their sense of taste, an early indication of multiple diseases, including viral infections and Alzheimer’s disease. And more sophisticated versions could one day augment taste in people who’ve lost it.

The post You Can Taste Cake in Virtual Reality With This New Device appeared first on SingularityHub.

Část věcí bude při podávání daňového přiznání oproti předchozímu roku jinak. Na které změny musíte letos pamatovat?