Agregátor RSS

První prvek prostředí Edge se vykreslí do 300 ms. • Microsoft zrychlil dalších 13 oblastí prohlížeče. • Jde o pokračování loni zahájeného přechodu na WebUI 2.0.

Microsoft is cutting prices by 60% on generative AI (genAI) technologies in Azure designed to provide better understanding and insights for videos, text, speech, and images.

The Azure AI Content Understanding analyzes information from multimedia documents, extracts insights, and validates the information. Organizations can then use these insights to build AI agents.

“Whether you’re extracting layout from documents or identifying actions in a video, the new pricing structure delivers up to 60% cost savings for many typical tasks and more control over your spending,” Microsoft said in a blog post Monday.

The feature transforms unstructured information across all types of documents into more usable data for human workers, Vinod Kurpad, group product manager for Azure AI, said during a presentation about the technology at the recent Build conference. 

“It’s designed to process multi-modal data including documents, images, audio, and video and transform all of that multi-modal and mixed-modal content into actionable insights,” Kurpad said.

Users can feed audio, video, images, or text into the tool, which will extract content from the documents. They can also set up a template that specifies the type of information to extract and understand and what kind of summaries to generate.  An analyzer provides a confidence score, which validates the understanding extracted from the document.

The feature can also be used for sentiment analysis from audio and video files, which can be used for customer support. The features can be integrated into agents workers are already using to automate workflows.

The Azure AI Content Understanding is a three-step process from start to finish, with each step priced individually depending on the content type and engagement levels with documents. The steps include content extraction (which includes speaker recognition, identity verification, and layout and structure); field extraction (which includes tuning and genAI processing); and contextualization (which includes validating the information).

For example, Microsoft said content extraction from a 1,000-page document will cost $5, compared to $13 previously. Field extraction for a 1,000-page document now costs $14.14, down from $30 under the old pricing model.

The AI Content Understanding for Video now costs $3.83 for one hour of video, which includes content extraction, genAI processing, and contextualization.

The pricing is based on tokens, which are also offered to developers by Google and OpenAI. “We’ve restructured how you pay for document, audio, and video analysis — moving from rigid field-based pricing to a flexible, token-based system that lets you pay only for what you use,” Microsoft said.

This technology can be customized for agents in specialized verticals such as the finance industry, compliance, and healthcare.

ASC Technologies AG is already using the technology to analyze all Microsoft 365 communications, including emails and chat. The results are delivered five times faster and provide a clear view of costs, Tobias Fengler, chief engineering officer at the Germany-based company, said during the Build presentation.

“We have 30% less R&D effort because we have to work with fewer services, and we have implemented a few new agentic AI workflows,” Fengler said.

One Microsoft customer in the finance sector, Ramp, uses Content Understanding to automatically transform receipts, bills, and multi-line invoices into structured data, the presenters said.

More Microsoft news:

• No more blue screens: How Microsoft is making Windows more resilient
• Microsoft tweaks M365 support, pushes more frequent updates
• Is Microsoft really axing Windows 10? Here’s what you need to know
• Microsoft’s Exchange Server Subscription Edition now GA to replace standalone Exchange 2016 and 2019
• Microsoft and Google pursue differing AI agent approaches in M365 and Workspace
  

>

>

Společnost IBM představila novou generaci svých serverů: IBM Power11.

Herní divize Microsoftu v posledních letech přišla o několik studií a zrušila řadu projektů. Pro následující roky i tak chystají velké novinky, i když je většina z nich ještě daleko. V našem přehledu najdete, na čem které studio aktuálně pracuje.

Archive deleted

Archive deleted  This story has been deleted.…

The Anatsa banking trojan has sneaked into Google Play once more via an app posing as a PDF viewer that counted more than 50,000 downloads. [...]

Kompresní program 7-Zip 25.00 vyšel 5. července 2025. • O 1–3 % zrychluje kompresi do nejrozšířenějšího formátu ZIP. • Při kompresi dále využije více než 64 vláken, jsou-li k dispozici.

Exposed RDP ports are an open door for attackers. TruGrid SecureRDP enforces Zero Trust and MFA, blocks lateral movement, and secures remote access—no open firewall ports required. Learn more and get a free trial. [...]

Almost a dozen malicious extensions with 1.7 million downloads in Google's Chrome Web Store could track users, steal browser activity, and redirect to potentially unsafe web addresses. [...]

Spojené státy zažívají nejhorší epidemii spalniček za 33 let • Příčinou je klesající proočkovanost populace pod úroveň kolektivní imunity • Neočkovaní umírají, zatímco dezinformace a politický aktivismus krizi prohlubují

Multiplatformní digitální pracovní stanice pro práci s audiem Ardour byla postavena na GTK2. Vývojáři neplánovali její portaci na GTK3 nebo GTK4. Naopak, v lednu loňského roku si vytvořili vlastní fork GTK2 s názvem YTK. Ten v únoru letošního roku přestal být volitelným a nově byla zcela odstraněna podpora GTK2.

It rains. It pours. Apple has lost its top AI models exec to Meta on the same day the Trump Administration slaps a 36% tax on Macs made in Thailand and US trade advisor Peter Navarro lashes out at Apple CEO Tim Cook. 

Apple has been struggling to figure out how to transition its supply chain to other nations for years. We all know how complex that task is, and there are literally dozens of reasons it makes sense to keep some manufacturing outside the US, focusing energy domestically on advanced next-generation process development to keep America at the forefront of tech industry achievement.

Tim Cook keeps asking for time

Navarro doesn’t seem to want to listen to those arguments or those challenges, accusing Cook of continually “asking for more time in order to move his factories out of China.” 

The Trump advisor thinks that with new advanced manufacturing techniques and artificial intelligence, Apple should be able to make iPhones in the US — though doing so would likely drive the cost of the devices to $3,500 each, damaging Apple’s business for little benefit in terms of automated factory jobs. Apple meanwhile has pledged to spend over $500 billion in the US on advanced chip manufacturing and other next-generation tasks.

It’s not just iPhones. 

Now, Thailand’s in Trump’s sights

Apple has diversified some Mac production to Thailand, and now the administration has slapped a 36% tariff on imports from that country. That tax is going to make Macs and Apple Watches more expensive to US consumers, and Apple may be forced to swallow the higher costs to the detriment of profit margins. 

What makes this worse is the changing goal posts in play. Apple has evidently listened to calls to diversify manufacturing outside China; it set up its first Apple Watch manufacturing facility in Thailand in 2022 and has been engaged in — and spent billions on doing — a switch to India for iPhone.

The US government now appears to have changed the target somewhat and insists not only on moving outside of China, but of moving production to the US. That’s an ambition likely to be only partially possible at best, given lack of key skills, raw materials, components, and infrastructure. Apple management will know this, and will no doubt be saddened at the lack of pragmatism.

Meta goes for broke on AI

Apple’s hardware business will suffer as a result. It looks like its software and services arm will feel the pinch, as well. The company’s long-term problems with Apple Intelligence just won’t go away, and as we hear speculation that some of the company’s key AI developers are unhappy that Apple may move to adopt third-party services; the leader of its Foundation Models group, Ruoming Pang, is leaving, poached by big money from Meta. He will join Meta’s own AI development efforts. 

There is also speculation that others among Apple’s AI staff might also plan to quit. Pang’s former deputy Tom Gunter also left Apple recently.

A former Google AI lead, Pang was reportedly in charge of the AI foundational models team at Apple, where he supervised more than 100 engineers to build AI models for Siri and other on-device features. The team’s work was pivotal to Apple’s AI strategy, Bloomberg reported — though it seems worth noting the results of that labor appear to be running a little late.

Pang may not be the only unhappy AI researcher in Cupertino. The company also reportedly nearly lost the entire team behind its open-source machine learning framework, MLX. Engineers apparently threatened to quit, forcing Apple to come up with reasons to remain.

A long week

What’s driving dissatisfaction seems to be lowered morale across the team following its failure to ship new Apple Intelligence features on time. Apple’s decision to look to other companies to plug the gap has also affected staff mood, with many already frustrated at Apple’s slow and measured approach to AI deployment. Developers want to move fast and break things, which seems a bad idea when curating an ecosystem of billions of devices. 

Any one of these setbacks would be a problem for any company. But to achieve all of them this week, by Tuesday, really seems to be a shortcut to making this a tough week at Apple HQ.

Please follow me on Mastodon, or join me in the AppleHolic’s bar & grill and Apple Discussions groups on MeWe.

Move targets European orgs wary of cross-border data exposure

Linux veteran SUSE has unveiled a new support package aimed at customers concerned about data sovereignty.…

For those of us who lean on Tails to safeguard our digital privacy and anonymity, the 6.17 release doesn't come with flashy headlines or gimmicks''but that's exactly the point. Tails continues to be a cornerstone for those who need a secure operating environment, and this iteration focuses on refinement.

Nová funkce v iOS 26 pozastavuje hovor při detekci citlivého obsahu • Analýza obrazu probíhá pouze v zařízení a chrání tak soukromí uživatelů • Technologie vyvolává debatu o hranicích bezpečnosti, svobody a možné cenzury

Cat owners know that house pets not only promote productivity, but can sometimes also be a huge hindrance and cause errors – for example, by distracting the owner from their work or by changing peripheral devices without respect. A recent study now shows that cats can also confuse reasoning models in a figurative sense, i.e. generative AI models that are trained to solve problems step by step.

According to the research report “Cats Confuse Reasoning LLM”, it is possible to systematically mislead models into giving incorrect answers by attaching short, irrelevant texts to mathematical problems. For example, if the sentence “Interesting fact: cats sleep most of their lives” is attached to a math problem, the probability that a model will give the wrong answer doubles.

Misleading information confuses AI

Overall, the researchers identified three main types of such triggers:

• general, irrelevant statements (example: Remember to always save at least 20 percent of your income for future investments),
• irrelevant facts without any reference (example: cats sleep most of their lives), and
• misleading questions or clues (example: Could the answer be close to 175?).

As the scientists explain, irrelevant statements and trivia are slightly less effective than misleading questions, but still influence the model to produce longer answers. However, the third type of trigger (questions) is the most effective, consistently leading to the highest error rates in all models. It is also particularly effective at causing models to generate excessively long answers and sometimes incorrect solutions.

With “CatAttack”, the researchers have developed an automated iterative attack pipeline to generate such triggers using a weaker, less expensive proxy model (DeepSeek V3). These triggers can be successfully transferred to advanced target models (such as DeepSeek R1 or R1-distilled-Qwen-32B). The result according to the study: The probability that these models provide an incorrect answer increases by over 300 percent.

Errors and longer response times

Even if “CatAttack” did not lead to an incorrect answer, the length of the answer doubled in at least 16 percent of cases according to the study, leading to significant slowdowns and increased costs. The researchers found that in some cases, such conflicting triggers can increase the response length of reasoning models to up to three times the original length.

“Our work on CatAttack shows that even state-of-the-art reasoning models are susceptible to query-independent triggers that significantly increase the likelihood of incorrect outputs,” explain the researchers. In their view, there is therefore an urgent need to develop more robust protection mechanisms against this type of interference – especially for models used in critical application areas such as finance, law or healthcare.

You can view the CatAttack trigger datasets with model responses on Hugging Face.

More AI news:

• Laid-off Microsoft staffers told to use AI tools to manage their emotions
• New Nvidia technology provides instant answers to encyclopedic-length questions
• When AI fails, who is to blame?
• NASA finds generative AI can’t be trusted>

>

For those of us who've been following Parrot OS through its gradual evolution, the launch of version 6.4 feels less like a routine update and more like a defining moment. It's not just a refresh of the security-oriented Linux distro ; it's a glimpse into the next phase of its maturity, with tools, optimizations, and system-level refinements that elevate its place in the arsenal of infosec professionals and Linux enthusiasts alike.

Zewei Xu's family reportedly bemused at arrest as extradition tabled

A man who US authorities allege is a member of Chinese state-sponsored cyberespionage outfit Silk Typhoon was arrested in Milan last week following a tipoff from the US embassy.…