The Russian Plates team visited Saint Petersburg during pre-production for the series Crime and Punishment to shoot 360-degree footage of some of the citys iconic landscapes. They shot scenic backgrounds of the Neva and Moika rivers, as well as historic streets, for an authentic visual foundation.This material was adapted later for shooting on LED screens, allowing the recording of boat rides and car journey scenes within a controlled studio environment. The easy blend of real-world visuals with virtual production technology allowed the crew to authentically recreate the atmosphere of Saint Petersburg without the challenges that come with on-location shooting.The collaboration between Plus Studio and Zoom Production, directed by Vladimir Mirzoev and cinematography by Matvey Stavitsky, demonstrates a mastery of storytelling and innovative filmmaking. The behind-the-scenes footage is exciting to watch as it reveals how such immersive scenes were created. xovirtualproductionThe post Crime and Punishment Virtual Production on xovirtualproduction appeared first on Vfxexpress.

Bread becomes a "precious and beautiful element" in thisjewellery range created by designer Cindy Xinyi Wu as an ode to the act of baking and the energetic nature of dough.The Rising collection features sculptural rings, brooches, earrings, bangles and necklaces, all with baked bread dough bulging out of intricately shaped metal frameworks.The shapes were produced by allowing the dough to expand freely within the metal bands while rising and baking.The Rising jewellery collection is made with bread doughAfter baking, the bread was dehydrated to harden its form and preserve its appearance.Wu designed the bands themselves to reference the actions of kneading, stretching and folding during the breadmaking process.She channelled years of baking experience into the Rising collection, which she created as her final-year project in the jewellery design undergraduate course at London's Central Saint Martins.The jewellery was made by allowing bread to rise freely and bake around metal frameworksAfter experimenting with baking as a child, Wu said she began to formally study the craft in 2018 and enjoyed not just the joy of eating and sharing bread but the mental retreat provided by the process of breadmaking."Baking requires my complete concentration at every step and careful observation throughout each phase," Wu told Dezeen."Immersing myself fully in the process of making bread helps me relax you could say this hobby has saved me over the years."The metal elements reference how dough looks as it is being stretched and folded"While I might be realistic, orderly, repressed, anxious and a perfectionist in my primary world, whenever I feel crises and a loss of control, baking becomes my retreat," she continued."It transports me to this second world of bread, where I can breathe again and regain my composure," Wu added."Thus, baking is not just a hobby for me; it's a vital escape from the harshness of reality and a way to alleviate stress and anxiety."The baked bread was dehydrated to preserve itHer goal with the collection was both to document her own connection with bread and to elevate the foodstuff to the position of a decorative material, on a par with gemstones."I've always perceived bread as a precious and beautiful element," Wu said."In my work, I intended for bread to take the central role traditionally occupied by gemstones in jewellery, hence my method of integrating bread with metal is influenced by the classic technique of setting gemstones in metal frameworks."Designer Cindy Xinyi Wu sees the bread as serving in the place of gemstonesThe metal frames in Wu's work take many different forms. Some are deeply textured like slabs of pulled dough, some are complex and twisted like the folds of a bun, while others provide a smooth and gleaming counterpoint.The designer said her baking studies had introduced her to the many different metal tools and moulds used to shape bread around the world, which ultimately inspired her to experiment with metal frameworks.Read: Chess set hidden in scarf, watch and rings by Louis Le Joly Senoville"Unlike traditional recipes, where moulds strictly define the bread's shape, I wanted the dough to expand freely and randomly, showcasing its own vitality and the unique aesthetic brought about by its rising," said Wu.Wu hopes that her deep affinity for the material comes across in the collection and inspires viewers to look at bread anew.Wu is a longtime baker who brought her experience to the projectShe points to characteristics like bread's unpredictability, the surprising changes brought on by factors such as temperature or length of fermentation, and the way dough "relaxes" and slumps after being shaped that give this food a unique and loveable character."The aspect of baking I love the most is the dough's expansion during fermentation, which I refer to as 'rising' also the name of my collection," said Wu."In the Rising series, I aim to capture the vitality and inherent energy of dough as it expands, documenting the unique shapes created by its free and random growth."Wu hopes the collection entices viewers to consider the unique character of bread"Watching the dough rise is fascinating; the yeast slowly fills it with air, and bubbles gradually emerge on the surface," she continued."To me, this process feels like breathing, like a pulse, with each batch of dough possessing a distinct personality. This part of the process always feels healing and recharging to me."Another product released for bread lovers this year is the baguette stamp designed by Stphane Humbert-Basset for the French post office. As well as featuring a little baguette illustration, the scratch-and-sniff stamp gives off the aroma of baked goods.The post Bread takes place of gemstones in Rising jewellery collection appeared first on Dezeen.

A screen for anyone who likes to work in the dark, and still be able to see afterwards.

Aligning large language models (LLMs) with human preferences is an essential task in artificial intelligence research. However, current reinforcement learning (RL) methods face notable challenges. Proximal Policy Optimization (PPO) and similar techniques often demand extensive online sampling, which can lead to high computational costs and instability. Offline RL methods like Direct Preference Optimization (DPO) avoid these issues but face difficulties with tasks requiring multi-step reasoning, such as solving mathematical problems or generating complex code. These methods frequently treat the generation process as a single-step problem, neglecting the long-horizon dependencies intrinsic to many reasoning tasks. Additionally, sparse reward functions, which provide feedback only at the conclusion of a reasoning sequence, make intermediate step guidance challenging.Researchers from ByteDance and UCLA have introduced Direct Q-function Optimization (DQO) to address these challenges. DQO frames the response generation process as a Markov Decision Process (MDP) and utilizes the Soft Actor-Critic (SAC) framework. By parameterizing the Q-function directly through the language model, DQO shifts the LLM alignment problem into a structured, step-by-step learning process. Unlike bandit-based methods, DQO incorporates process rewardsintermediate feedback signalsto support multi-step reasoning more effectively.A key feature of DQO is its ability to identify and optimize correct reasoning steps even within partially correct responses. For example, in mathematical problem-solving, DQO assigns higher value to accurate steps and penalizes errors, enabling incremental improvement in reasoning. This makes DQO particularly suitable for tasks requiring detailed, long-horizon decision-making.Technical Implementation and Practical AdvantagesDQOs approach is centered on parameterizing the Q-function using the language model, thereby integrating policy and value functions. The model updates its Q-function and value function based on the Soft Bellman Equation. KL-regularization ensures stable learning and helps prevent overfitting to specific samples.To handle challenges such as high bias in temporal difference errors, DQO employs -return, a mechanism that balances short-term and long-term rewards for more stable training. Importance sampling further enhances DQOs offline learning capabilities by reducing distributional shifts between the training data and the models policy.DQO offers several practical advantages. It eliminates the need for online sampling, reducing computational costs. Moreover, it can learn from unbalanced and negative samples, enhancing its robustness across various scenarios. The use of process rewards helps refine reasoning capabilities while improving alignment with task requirements.Results and InsightsExperimental evaluations of DQO on mathematical reasoning datasetsGSM8K and MATHdemonstrate its effectiveness. On the GSM8K dataset, DQO improved performance from a baseline of 59.06% to 87.26% for greedy generation and from 53.30% to 84.69% for sampling-based generation. These results surpass other baseline methods, including DPO and DRO. Similarly, on the MATH dataset, DQO outperformed baselines, achieving improvements of 1.18% in sampling and 1.40% in greedy generation.Enhancing DQO with process rewards further boosted performance, suggesting its potential to incorporate additional supervisory signals. These results underscore DQOs capability to handle multi-step reasoning tasks effectively and align LLMs with complex objectives.ConclusionDirect Q-function Optimization (DQO) offers a thoughtful approach to reinforcement learning for LLM alignment. By framing response generation as an MDP and utilizing the SAC framework, DQO addresses the limitations of existing methods. Its ability to integrate process rewards, handle unbalanced data, and stabilize training through -return and importance sampling makes it a practical solution for tasks involving multi-step reasoning.Future research could explore applying DQO to other domains, such as code generation and dialogue systems, where long-horizon decision-making is critical. As AI systems evolve to tackle increasingly complex challenges, methods like DQO will play an important role in enhancing the alignment and performance of language models.Check out the Paper. All credit for this research goes to the researchers of this project. Also,dont forget to follow us onTwitter and join ourTelegram Channel andLinkedIn Group. Dont Forget to join our60k+ ML SubReddit. Aswin Ak+ postsAswin AK is a consulting intern at MarkTechPost. He is pursuing his Dual Degree at the Indian Institute of Technology, Kharagpur. He is passionate about data science and machine learning, bringing a strong academic background and hands-on experience in solving real-life cross-domain challenges. [Download] Evaluation of Large Language Model Vulnerabilities Report (Promoted)

Dec 31, 2025Ravie LakshmananVulnerability / Cloud SecurityCybersecurity researchers have uncovered three security weaknesses in Microsoft's Azure Data Factory Apache Airflow integration that, if successfully exploited, could have allowed an attacker to gain the ability to conduct various covert actions, including data exfiltration and malware deployment."Exploiting these flaws could allow attackers to gain persistent access as shadow administrators over the entire Airflow Azure Kubernetes Service (AKS) cluster," Palo Alto Networks Unit 42 said in an analysis published earlier this month.The vulnerabilities, albeit classified as low severity by Microsoft, are listed below -Misconfigured Kubernetes RBAC in Airflow clusterMisconfigured secret handling of Azure's internal Geneva service, andWeak authentication for GenevaBesides obtaining unauthorized access, the attacker could take advantage of the flaws in the Geneva service to potentially tamper with log data or send fake logs to avoid raising suspicion when creating new pods or accounts.The initial access technique involves crafting a directed acyclic graph (DAG) file and uploading it to a private GitHub repository connected to the Airflow cluster, or altering an existing DAG file. The end goal is to launch a reverse shell to an external server as soon as it's imported.To pull this off, the threat actor will have to first gain write permissions to the storage account containing DAG files by utilizing a compromised service principal or a shared access signature (SAS) token for the files. Alternatively, they can break into a Git repository using leaked credentials.Although the shell obtained in this manner was found to be running under the context of the Airflow user in a Kubernetes pod with minimal permissions, further analysis identified a service account with cluster-admin permissions connected to the Airflow runner pod.This misconfiguration, coupled with the fact that the pod could be reachable over the internet, meant that the attacker could download the Kubernetes command-line tool kubectl and ultimately take full control of the entire cluster by "deploying a privileged pod and breaking out onto the underlying node."The attacker could then leverage the root access to the host virtual machine (VM) to burrow deeper into the cloud environment, gain unauthorized access to Azure-managed internal resources, including Geneva, some of which grant write access to storage accounts and event hubs."This means a sophisticated attacker could modify a vulnerable Airflow environment," security researchers Ofir Balassiano and David Orlovsky said. "For example, an attacker could create new pods and new service accounts. They could also apply changes to the cluster nodes themselves and then send fake logs to Geneva without raising an alarm.""This issue highlights the importance of carefully managing service permissions to prevent unauthorized access. It also highlights the importance of monitoring the operations of critical third-party services to prevent such access."The disclosure comes as the Datadog Security Labs detailed a privilege escalation scenario in Azure Key Vault that could permit users with the Key Vault Contributor role to read or modify Key Vault contents, such as API keys, passwords, authentication certificates, and Azure Storage SAS tokens.The problem is that while a user with the Key Vault Contributor role had no direct access to Key Vault data over a key vault configured with access policies, it was discovered that the role did come with permissions to add itself to Key Vault access policies and access Key Vault data, effectively bypassing the restriction."A policy update could contain the ability to list, view, update and generally manage the data within the key vault," security researcher Katie Knowles said. "This created a scenario where a user with the Key Vault Contributor role could gain access to all Key Vault data, despite having no [Role-Based Access Control] permission to manage permissions or view data."Microsoft has since updated its documentation to emphasize the access policy risk, stating: "To prevent unauthorized access and management of your key vaults, keys, secrets, and certificates, it's essential to limit Contributor role access to key vaults under the Access Policy permission model."The development also follows the discovery of an issue with Amazon Bedrock CloudTrail logging that made it difficult to differentiate malicious queries from legitimate ones made to large language models (LLMs), thereby allowing bad actors to conduct reconnaissance without raising any alert."Specifically, failed Bedrock API calls were logged in the same manner as successful calls, without providing any specific error codes," Sysdig researcher Alessandro Brucato said."The lack of error information in API responses may hinder detection efforts by generating false positives in CloudTrail logs. Without this detail, security tools may misinterpret normal activity as suspicious, leading to unnecessary alerts and potential oversight of genuine threats."Found this article interesting? Follow us on Twitter and LinkedIn to read more exclusive content we post.SHARE

Searching for the perfect online mattress? Youre in the right place.

Our Experts Written by David Watsky Our expert, award-winning staff selects the products we cover and rigorously researches and tests our top picks. If you buy through our links, we may get a commission. Reviews ethics statement Why You Can Trust CNET 16171819202122232425+ Years of Experience 14151617181920212223 Hands-on Product Reviewers 6,0007,0008,0009,00010,00011,00012,00013,00014,00015,000 Sq. Feet of Lab Space How we test CNETs expert staff reviews and rates dozens of new products and services each month, building on more than a quarter century of expertise.Table of Contents Our Picks Bragg Refreshers View details $12 at Walmart View details Sidekick Sodas View details $39 at Sidekick Sodas View details Poppi Prebiotic ACV Sodas View details $29 at Amazon View details Remedy ACV Immunity Shot View details $50 at Amazon View details Bragg Flavored ACV Shots View details $20 at Amazon View details Emergen-C ACV Packets View details $13 at Walmart View details Remedy Apple Cider Vinegar Sparkling Water (Currently out of stock) View details $30 at Amazon View details Table of Contents These flavored Apple Cider Vinegar drinks make downing the healthy elixir a whole lot more pleasant. David Watsky/CNETApple cider vinegar is popular for the amount of health benefits it offers. It's a miracle drink, with the ability to kill bacteria, improve gut health and cholesterol levels, which makes it an essential drink for those with various illnesses. But at the same time, the drink is also notorious for its taste: It tastes like death warmed up (or on ice, if you prefer it cool), and most folks can't stomach it. Thankfully, there is a selection of drinks that incorporate apple cider vinegar while providing a much better taste, including sodas, shots and more.Do you want all the potential benefits of apple cider vinegar, but are reluctant to drink it straight? Don't worry, there are other alternatives. These apple cider vinegar-based beverages are mixed with fresh fruit juice, honey and ginger -- also known for their health benefits -- and are meant to be enjoyed, not tolerated. If these ACV drinks dont hit the spot, no worries -- its not a total loss. Share the leftovers with friends or family as a thoughtful post-holiday treat. After all, its the season of giving, and nothing goes to waste. I tried seven options, including apple cider vinegar (ACV) sparkling water, a few blends of ACV and fruit juices, and some flavored shots of premixed vinegar. Below, you'll find a list of my favorite drinks and quick shots to help you get your daily apple cider vinegar fix, along with what I thought about how they taste.Read more:The ABCs of Apple Cider Vinegar: Benefits, Precautions and Proper Dosage Watch this: Taste-Testing 6 Apple Cider Vinegar Drinks 01:32 Best apple cider vinegar drinks $12 at Walmart Braggs Refreshers contain apple cider vinegar along with juice, lemon and other natural ingredients to create a refreshing beverage anchored by healthy apple cider vinegar. The beverages come in four flavors, but I liked the lemon ginger refresher best. The 40-calorie refreshers taste a lot like kombucha and are best consumed cold or poured over ice. $39 at Sidekick Sodas Sidekicks are not your typical sugar-packed soda. These fruity and tart sodas mix healthy ACV with plum, strawberry and mango juice and a dash of New Zealand honey. All three recipes I tried were excellent; not too sweet and perfectly refreshing. I enjoyed the plum strawberry the most and found the sparkling beverage was best when poured over ice.Sidekicks aren't cheap. You can order a 12-pack from the website for $39. This fun and flashy line of low-calorie sodas is marketed as aiding in digestive health and boasts apple cider vinegar as the third ingredient after water and cane sugar. Poppi prebiotic sodas come in several flavors; I tried lemon-strawberry and ginger-lime and found both delicious. The apple cider vinegar flavor is tamer than in some of the other beverages I tried, and with 4 grams of sugar, the sodas aren't too sweet and have just 25 calories.A 12-pack of Poppi's healthier ACV sodas costs $30 at Walmart or $8 for a four-pack at Target.

Sparklight may have wide coverage in Nampa, but Quantum Fibers multi-gig plan steals the spotlight. Check out the top broadband options in this Idaho city.

You can trust VideoGamer. Our team of gaming experts spend hours testing and reviewing the latest games, to ensure you're reading the most comprehensive guide possible. Rest assured, all imagery and advice is unique and original. Check out how we test and review games hereDemon Slayer is a collaboration that many Fortnite players would love to have. Chapter 6 has brought many new collabs and amazing skins to the game, yet Epic Games wont slow down anytime soon. There have been a lot of rumblings regarding the popular anime, yet the Fortnite developer hasnt revealed anything about it yet.In this article, we will take a look at the the rumored partnership in Fortnite. Well check where this rumor comes from and if the Demon Slayer collaboration will really come to the game.When will the Fortnite x Demon Slayer collaboration come out?Before we start, it should be noted that this Fortnite anime collaboration has not been confirmed by Epic Games yet. However, like most of other information about the future content, the collaboration was leaked. This time, Hypex and ShiinaBR leaked the partnership, and both of them have an amazing track record with leaks.The collaboration was leaked in late November 24, a week before the release of Fortnite Chapter 6. However, these anime skins have come out since then, which is why many players have questioned the legitimacy of the leak.The Fortnite x Demon Slayer collaboration should be released soon. Image by VideoGamerWhile these two leakers are generally trustworthy, its important to note that theyve never revealed the potential release date of the Fortnite x Demon Slayer collaboration. They simply said that it would come soon, which likely refers to Chapter 6 Season 1. With the season ending in late February 2025, we are likely still a few weeks away from the collab.There is also a chance that Epic Games scrapped this content. However, this is an unlikely scenario considering how much traction it has gained over social media. The company is on a winter break right now, but we can expect more leaks and possibly collaboration teasers in early January, after the next Fortnite update. FortnitePlatform(s):Android, iOS, macOS, Nintendo Switch, PC, PlayStation 4, PlayStation 5, Xbox One, Xbox Series S/XGenre(s):Action, Massively Multiplayer, Shooter9VideoGamerSubscribe to our newsletters!By subscribing, you agree to our Privacy Policy and may receive occasional deal communications; you can unsubscribe anytime.Share

Since burning through tens of billions of dollars on its flop metaverse concept and laying off thousands in the aftermath of that gamble, tech giant Meta has strained to reinvent itself as a company poised to capitalize on the overhyped AI revolution. Last year, for example, founder and CEO Mark Zuckerberg unveiled animated AI chatbots modeled as alter egos of celebrities including Snoop Dogg, MrBeast, Paris Hilton, and Kendall Jenner. But licensing the voices and likenesses of famous people did little to endear Meta to the younger demographics it wants to turn into loyal users of Facebook and Instagram. It junked the bots widely ridiculed as creepy, corny, and without real purpose less than a year later. This move coincided with the debut of its AI Studio, where users can create their own chatbots, including wholly fictional characters or, in the case of popular creators who want followers to feel more connected, an AI that can message with your audience on your behalf, mimicking your tone and expressions. Now, Meta is planning to take the next step: integrating these AI creations as Facebook and Instagram users in themselves. As reported by the Financial Times, the hope is that these semi-independent custom avatars will prove more engaging to the young people who are crucial to the survival of Metas flagship social networks. We expect these AIs to actually, over time, exist on our platforms, kind of in the same way that accounts do, Connor Hayes, Metas vice-president of product for generative AI, told FT. Theyll have bios and profile pictures and be able to generate and share content powered by AI on the platform...thats where we see all of this going.The prospect seems an unusual one for websites predicated on the idea of human-to-human interaction. While Metas other AI offerings, including photo-editing tools, a ChatGPT-like text bot, and forthcoming software for producing AI-generated video, have more readily obvious appeal, its far from certain that an Instagram or Facebook populated by virtual apparitions would prove enticing to newcomers or satisfying for long-time users. Meta revealed to FT that users have created hundreds of thousands of characters since AI Studio launched in July, but that most of them remain private. The company declined to comment further to Rolling Stone. Whatever the reaction, such a change to the fundamental structure of Metas platforms could be transformative. Recent months have brought a wave of alarm about our growing attachments to AI, from student overreliance on it in the classroom and its ability to spread inappropriate content or dangerous misinformation to a recent lawsuit from a mother of a teen who claims his intimate involvement with a Character.AI bot led to his suicide. That company did not comment on the pending litigation but told CNN that it was heartbroken by the tragic loss of one of our users. (AI dating programs are a burgeoning business particularly fraught with emotional hazard.)The envisioned change also comes at a time when Facebook in particular is dominated by AI-generated spam images commonly referred to as slop. These pictures range from visually attractive though nonexistent landscapes and houses to depictions of babies, soldiers, the American flag, and Jesus Christ, presumably meant to evoke a response from older and conservative-leaning audience. This low-effort engagement farming allows the individuals posting the images many of them in the Global South to monetize the pages or sell related merchandise. Through payments to content creators who develop large followings for slop, Meta effectively incentivizes the content.Some of that engagement no doubt comes from spam bots, which are automated to like, share, and comment on posts in order to drive interactions. But the implications of a Facebook where the majority of the activity is bots bouncing off one another are hard to grasp, even at a moment when artificially gamed exchanges are common on the social network. Behind all this, Meta has been working to streamline its path to AI leadership while steadily rolling out more AI features to its 3 billion monthly active users. In January, Zuckerberg announced the merger of its Fundamental AI Research (FAIR) division with its GenAI product team, saying the move would accelerate their work in this field and was an important step on the way to creating a so-called artificial general intelligence. An AGI is a hypothetical autonomous system capable of teaching itself and surpassing human intelligence.Experts told Rolling Stone at the time that such a breakthrough is still rather far-fetched, while Metas efforts to leverage user data in order to engineer its AI tech posed significant privacy risks. They also questioned Zuckerbergs promise of transparency in the process and goal of responsibly open-sourcing an AGI making the code available to the public noting that Meta has offered little insight into its existing models and there is no real framework for safely releasing the blueprint for an (as yet theoretical) AGI. All told, then, the latest AI spin from Meta sounds like more of the same: a mix of vague hype, dubious business models, and a baffling sense of what people really want from a social media network. If, in a few years, Instagram and Facebook are just places for AI bots to hang out, it stands to reason that the humans may find other ways to communicate.