• 0 التعليقات ·0 المشاركات ·89 مشاهدة
  • Pokemon TCG Pocket: Glaceon ex Guide (Deck, Strategy, Counter)
    gamerant.com
    Despite initial excitement, Glaceon ex has struggled to escape the shadow of other Eeveelutions in Pokemon TCG Pocket. Even its brief success after release, which sparked hope among fans, wasnt enough to elevate its status.
    0 التعليقات ·0 المشاركات ·68 مشاهدة
  • Best Open-World Games Featuring Supernatural Abilities, Ranked
    gamerant.com
    Open-world games are incredibly popular because, at their most fundamental level, they sell the greatest fantasy of going out into the world with total freedom and doing whatever players want. How can that feeling of freedom be enhanced? Well, supernatural abilities are a great place to start!
    0 التعليقات ·0 المشاركات ·67 مشاهدة
  • ChatGPT remains the most popular AI tool in offices worldwide, survey finds, with India leading the way
    www.techradar.com
    ChatGPT remains the most widely used AI tool in workplaces, with India leading global adoption with a 92% usage rate.
    0 التعليقات ·0 المشاركات ·65 مشاهدة
  • NYT Connections hints and answers for Saturday, March 8 (game #636)
    www.techradar.com
    Looking for NYT Connections answers and hints? Here's all you need to know to solve today's game, plus my commentary on the puzzles.
    0 التعليقات ·0 المشاركات ·64 مشاهدة
  • This Vegan Leather Passport Holder is a Masterclass in Women-Led EDC Design
    www.yankodesign.com
    A passport holder might seem like a simple objectan accessory designed to keep essentials in placebut when design is approached with intention, even the smallest details carry meaning. The Belmont Vegan Leather Passport Holder reflects that philosophy, balancing form and function with a thoughtful approach to everyday travel. Behind its creation are two women whose influence extends beyond design itselfGladys Phan, Senior Product Designer, and Krystle Koo, Brand Strategist. Together, they shape not just the physical product, but the entire experience surrounding it. On International Womens Day, we take a deep dive into how they brought their unique perspectives together to create the Belmont Passport Holder.Gladys understands that good design cant be limited to merely aestheticsits about how it works in real life. With the Belmont, she focuses on practicality without sacrificing aesthetics, ensuring that every detail serves a purpose. The structured compartments provide easy access to travel documents, while the vegan leather adds durability and a refined feel. Every stitch, pocket, and fold is thoughtfully designed for a product that feels as intuitive as it looks refined. Nothing is included without reason, and that level of precision is what sets her work apart.Designers: Gladys Phan / Krystle KooClick Here to Buy Now.Gladys Phan and Krystle KooKrystles role builds on that foundation, ensuring that the Belmont isnt just a one-off SKU but part of a larger story. Travel is about movement, identity, and the objects we take with us along the way. She shapes the way the passport holder connects with people, emphasizing values like conscious craftsmanship and intentional design. The product exists not in isolation but as an extension of a broader vision, where utility and meaning are intertwined. The messaging, branding, and identity surrounding the Belmont carry the same level of care as its physical design, creating a cohesive experience from start to finish.Gladys and Krystles collaboration brings together two distinct but deeply connected skill setsone shaping the physical product, the other defining its meaning in the world. Gladys design decisions werent made in isolation; they were informed by the broader narrative Krystle was crafting. The intuitive layout of the passport holder wasnt just about storageit was about anticipating how travelers move, how they reach for documents in transit, how an item can become second nature in their hands. Krystles strategic insight ensured that these choices werent just functional but aligned with a vision of purposeful, conscious design. Their process wasnt linearideas moved back and forth, evolving with each refinement, proving that when women collaborate, the result reflects that through a combined experience shaped by lived understanding and intention.The result is a passport holder that feels effortless to use, yet carries layers of thoughtfulness beneath its surface. Its minimalist design is intentional, removing clutter while ensuring that the essentials are exactly where they need to be. Vegan leather was chosen not just for its ethical considerations, but for its tactile quality and durability. Even the stitching speaks to longevity, reinforcing the idea that well-designed objects should withstand time and use.Beyond the product itself, this campaign highlights the broader impact of women in design. Their contributions extend across disciplines, shaping not only what we use but how we engage with the world around us. Gladys and Krystles work demonstrates that design is broadly about problem-solving, storytelling, and creating products that seamlessly fit into everyday life. The Belmont Passport Holder embodies that philosophy, proving that when design is approached with intention, even the smallest objects carry significance.Click Here to Buy Now.The post This Vegan Leather Passport Holder is a Masterclass in Women-Led EDC Design first appeared on Yanko Design.
    0 التعليقات ·0 المشاركات ·62 مشاهدة
  • Finer-CAM Revolutionizes AI Visual Explainability: Unlocking Precision in Fine-Grained Image Classification
    www.marktechpost.com
    Researchers at The Ohio State University have introduced Finer-CAM, an innovative method that significantly improves the precision and interpretability of image explanations in fine-grained classification tasks. This advanced technique addresses key limitations of existing Class Activation Map (CAM) methods by explicitly highlighting subtle yet critical differences between visually similar categories.Current Challenge with Traditional CAMConventional CAM methods typically illustrate general regions influencing a neural networks predictions but frequently fail to distinguish fine details necessary for differentiating closely related classes. This limitation poses significant challenges in fields requiring precise differentiation, such as species identification, automotive model recognition, and aircraft type differentiation.Finer-CAM: Methodological BreakthroughThe central innovation of Finer-CAM lies in its comparative explanation strategy. Unlike traditional CAM methods that focus solely on features predictive of a single class, Finer-CAM explicitly contrasts the target class with visually similar classes. By calculating gradients based on the difference in prediction logits between the target class and its similar counterparts, it reveals unique image features, enhancing the clarity and accuracy of visual explanations.Finer-CAM PipelineThe methodological pipeline of Finer-CAM involves three main stages:Feature Extraction:An input image first passes through neural network encoder blocks, generating intermediate feature maps.A subsequent linear classifier uses these feature maps to produce prediction logits, which quantify the confidence of predictions for various classes.Gradient Calculation (Logit Difference):Standard CAM methods calculate gradients for a single class.Finer-CAM computes gradients based on the difference between the prediction logits of the target class and a visually similar class.This comparison identifies the subtle visual features specifically discriminative to the target class by suppressing commonly shared features.Activation Highlighting:The gradients calculated from the logit difference are used to produce enhanced class activation maps that emphasize discriminative visual details crucial for distinguishing between similar categories.Experimental ValidationB.1. Model AccuracyResearchers evaluated Finer-CAM across two popular neural network backbones, CLIP and DINOv2. Experiments demonstrated that DINOv2 generally produces higher-quality visual embeddings, achieving superior classification accuracy compared to CLIP across all tested datasets.B.2. Results on FishVista and AircraftQuantitative evaluations on the FishVista and Aircraft datasets further demonstrate Finer-CAMs effectiveness. Compared to baseline CAM methods (Grad-CAM, Layer-CAM, Score-CAM), Finer-CAM consistently delivered improved performance metrics, notably in relative confidence drop and localization accuracy, underscoring its ability to highlight discriminative details crucial for fine-grained classification.B.3. Results on DINOv2Additional evaluations using DINOv2 as the backbone showed that Finer-CAM consistently outperformed baseline methods. These results indicate that Finer-CAMs comparative method effectively enhances localization performance and interpretability. Due to DINOv2s high accuracy, more pixels need to be masked to significantly impact predictions, resulting in larger deletion AUC values and occasionally smaller relative confidence drops compared to CLIP.Visual and Quantitative AdvantagesHighly Precise Localization: Clearly pinpoints discriminative visual features, such as specific coloration patterns in birds, detailed structural elements in cars, and subtle design variations in aircraft.Reduction of Background Noise: Significantly reduces irrelevant background activations, increasing the relevance of explanations.Quantitative Excellence: Outperforms traditional CAM approaches (Grad-CAM, Layer-CAM, Score-CAM) in metrics including relative confidence drop and localization accuracy.Extendable to multi-modal zero-shot learning scenariosFiner-CAM is extendable to multi-modal zero-shot learning scenarios. By intelligently comparing textual and visual features, it accurately localizes visual concepts within images, significantly expanding its applicability and interpretability.Researchers have made Finer-CAMs source code and colab demo available.Check outthe Paper, Github and Colab demo.All credit for this research goes to the researchers of this project. Also,feel free to follow us onTwitterand dont forget to join our80k+ ML SubReddit. Jean-marc MommessinJean-marc is a successful AI business executive .He leads and accelerates growth for AI powered solutions and started a computer vision company in 2006. He is a recognized speaker at AI conferences and has an MBA from Stanford.Jean-marc Mommessinhttps://www.marktechpost.com/author/jean-marc0000677/Inception Unveils Mercury: The First Commercial-Scale Diffusion Large Language ModelJean-marc Mommessinhttps://www.marktechpost.com/author/jean-marc0000677/CASS: Injecting Object-Level Context for Advanced Open-vocabulary semantic segmentationJean-marc Mommessinhttps://www.marktechpost.com/author/jean-marc0000677/MVGD from Toyota Research Institute: Zero Shot 3D Scene ReconstructionJean-marc Mommessinhttps://www.marktechpost.com/author/jean-marc0000677/Revolutionizing Robot Learning: How Metas Aria Gen 2 enables 400% Faster Training with Egocentric AI Recommended Open-Source AI Platform: IntellAgent is a An Open-Source Multi-Agent Framework to Evaluate Complex Conversational AI System' (Promoted)
    0 التعليقات ·0 المشاركات ·62 مشاهدة
  • How Deepseek Destroyed OpenAI, and How You Can Do it Too!
    towardsai.net
    LatestMachine LearningHow Deepseek Destroyed OpenAI, and How You Can Do it Too! 0 like March 8, 2025Share this postAuthor(s): Mohit Varikuti Originally published on Towards AI. What is PTX/ASM?This member-only story is on us. Upgrade to access all of Medium.In the rapidly evolving world of GPU computing, performance can often be the make-or-break factor in an applications success. One of the secret weapons behind high-performance frameworks like DeepSeek is the intelligent use of CUDA PTX and inline assembly (ASM). DeepSeeks remarkable efficiency and speed didnt come solely from high-level algorithm design; it was also the way DeepSeek got so good by exploiting low-level CUDA PTX/ASM optimizations to squeeze every ounce of performance from modern GPUs.In this article, well dive into CUDAs PTX (Parallel Thread Execution) language and explore how inline assembly can be used within CUDA kernels. Well look at what PTX is, how it fits into the CUDA compilation pipeline, and examine some practical code examples.CUDA PTX is an intermediate assembly-like language used by NVIDIA GPUs. Think of PTX as the assembly language for CUDA, though its higher-level than the actual machine code executed on the GPU. When you compile CUDA code using nvcc, your high-level C/C++ code is transformed into PTX code, which is then optimized and further compiled down to machine-specific binary code (SASS) for the target GPU, more specifically:Portability: PTX abstracts many hardware details, Read the full blog for free on Medium.Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming asponsor. Published via Towards AITowards AI - Medium Share this post
    0 التعليقات ·0 المشاركات ·63 مشاهدة
  • PSA: The Longer You Wait To File Your Taxes Online, The More You'll Pay
    www.cnet.com
    Our Tax Software Picks Don't stress about tax season. CNET Money editors have tested the top online tax software options to make your tax return easier so you can get your refund faster. See at TurboTax Best tax filing service for most filers TurboTax See at TurboTaxSee at H&R Block Best free tax filing service H&R Block See at H&R BlockSee at FreeTaxUSA A great free or low-cost filing option FreeTaxUSA See at FreeTaxUSASee at Cash App A free option that's great for confident filers Cash App Taxes See at Cash AppSee at TaxSlayer Best tax filing service for freelancers, gig workers and sole proprietors TaxSlayer See at TaxSlayerSee at TaxAct Best accuracy guarantee TaxAct See at TaxActSee at Jackson Hewitt Best for filing multiple state tax returns Jackson Hewitt See at Jackson HewittTax season kicks off each year when the IRS startes accepting tax returns at the end of January. But a little-known open secret that I've learned as a tax editor is that the early weeks of tax season are also when taxpayers can expect the best deals for filing their taxes online.When the calendar shifts to late February and early March, popular services like TurboTax, H&R Block, TaxSlayer and TaxAct tend to raise prices for federal and state returns. The costs of tax pro services that either offer tax return reviews or allow you to hand over your forms to a tax expert who will file your taxes on your behalf may also increase. TAX SOFTWARE DEALS OF THE WEEK H&R Block Free Simple Tax Returns eFile: $0 (save $0) TurboTax Deluxe 2024 (Federal and State, PC/Mac Download): $56 (save $24) TurboTax Premier 2024 (Federal and State, PC/Mac Download): $83 (save $32) TaxSlayer Classic Plan: $28 (save $10) Deals are selected by the CNET Group commerce team, and may be unrelated to this article. This story is part of Taxes 2025, CNET's coverage of the best tax software, tax tips and everything else you need to file your return and track your refund. As we approach the April 15 federal tax filing deadline, expect prices to increase further as taxpayers rush to get their taxes done.Waiting until the last minute to file your taxes can be costly, especially if you plan to itemize expenses or have complicated tax situations like freelance, rental or investment income or earned money cryptocurrency sales. Just about every tax service charges you more for these situations. Prices can easily eclipse $100, especially if you want help from a tax professional.For some, the cost may be worth it. Services like TurboTax and H&R Block, two CNET editor's choice picks this year, offer the best user experience of the companies on our best list in addition to tons of free tax resources, full-service options and AI tools.Can I still qualify to file my taxes for free?Yes. Filers with simple tax situations (W-2s, education costs and some basic credits) can still for free throughout tax season. Online services like TurboTax, H&R Block, TaxSlayer and TaxAct do not remove free tiers. But be sure to compare free-filing options to see if you qualify for any.In addition to tax services with tiered systems, consider filing with Cash App Taxes, which offers one federal and one state return for free no matter your tax situation.The IRS also offers two free filing options. IRS Free File is a federal program that lets you file for free with one of eight partners if you earned less than $84,000 last year. IRS Direct File, a proprietary tax filing service launched last year, also allows free returns.There is a tradeoff, though. You may not receive much contextual tax help or live support from a tax professional with free services. Make sure you're comfortable filing your taxes on your own if you select free tiers. Some options may also have form or salary restrictions.Low cost ways to file your taxes in 2025Don't qualify for free filing? Consider FreeTaxUSA, which offers a free federal return, but charges $14.99 per state return. I used this service to file my return this year. I lost out on a more premium experience compared to filng with H&R Block and TurboTax in past years, but FreeTaxUSA offers a solid question and answer model that is easy to navigate. Be prepared to enter all of your tax forms manually, however.Jackson Hewitt Online also allows you to file your taxes and multiple state returns for a flat $25 fee. Neither of these services have noteworthy form restrictions.Will CPAs raise prices too?If you're turning to an accountant for tax help, you won't fall victim to the same price hikes as tax day nears. It's rare for certified public accounts to raise their prices during tax season, said Jassen Bowman, IRS enrolled agent and tax expert."What you will see, however, are more and more tax professionals having an early cut-off date," he said.If you don't get your tax documents to them far enough in advance of the April 15 deadline, they may tell you that they'll need to file an extension to give them more time to sift through your tax file, Bowman added.But money talks. So if you're willing to pay a "rush fee" to get your taxes done on time, some firms will bump another client for you after you pay the surcharge."The way around all of this is simply to not procrastinate," Bowman said. "Get your documents together early, and book your appointment now so that you're not bumping up against the April 15 deadline." More tax tips
    0 التعليقات ·0 المشاركات ·64 مشاهدة
  • DOGE Plan to Push AI Across the US Federal Government is Wildly Dangerous
    www.techpolicy.press
    DOGE Plan to Push AI Across the US Federal Government is Wildly DangerousBen Green / Mar 6, 2025February 20, 2025Elon Musk speaking at the 2025 Conservative Political Action Conference (CPAC) at the Gaylord National Resort & Convention Center in National Harbor, Maryland. Gage Skidmore via Wikimedia CC BY-SA 2.0Last month, Elon Musks DOGE team outlined a plan to transform the United States federal government through an AI-first strategy.Led by Thomas Shedd, a former Tesla engineer now in charge of the General Services Administrations technology team, the plan is to deploy AI widely across the federal government. The overall goal is to cut the agencys budget by fifty percent. Shedd suggested using AI to analyze contracts for redundancies, root out fraud, and facilitate a reduction in the federal workforce by automating much of their work.Its no surprise that Musk, who co-founded OpenAI in 2015 and now (among his various endeavors) runs the company xAI, is advocating for an AI-driven government transformation. This ambition may seem plausible since AI tools are capable of many impressive feats, from writing essays to answering physics questions. But unleashing AI to revolutionize the federal government is a recipe for disaster.To predict how Musks proposed transformation will go, we can look at prior examples of integrating AI into government decision-making and operations. There are already many instances of AI-driven governance gone wrongparticularly in cases where the goal is to root out fraud and cut budgets.In my home state of Michigan, the Unemployment Insurance Agency (UIA) adopted an algorithm (the Michigan Integrated Data Automated System, aka MiDAS) to streamline its operations. The UIAs goals were to prevent unemployment fraud and to eliminate one-third of the existing staff by automating their work. MiDAS quickly boosted the cases of suspected welfare fraud by a factor of five, which led to a 23-fold increase in the UIAs revenues.Based on these numbers, MiDAS seems to have accomplished the UIAs goals. The only problem? Almost every accusation of fraud93% of supposed fraud caseswas incorrect. Then, even once these errors became clear, it took years of litigation for the wrongly accused to receive the money they were owed. For many people, the fraud charge remained on their criminal record for years, barring them from jobs.Around the United States and the rest of the world, similar algorithms have cut vulnerable people off healthcare and violated human rights.A central reason for these flawed AI tools is that engineers underestimate the complexity of government processes. In turn, engineers embed their superficial assumptions into software and overestimate their tools capabilities.It is a typical story that engineers walk into government departments with swashbuckling confidence, believing that technical expertise can break through existing bottlenecks and solve all the governments problems. These quick-fix proposals are typically laughable to the existing staff in those agencies, who have deep expertise in navigating the complex nuances of government processes. The engineers do not understand the intricate workflows that government agencies follow. And because they do not understand these workflows, the engineers do not know how to interpret government data and forms.In a recent study, colleagues and I asked computer scientists to develop a software tool that gives automated advice about whether people are eligible for bankruptcy relief. The computer scientists completed the task quickly, with one even noting, this seems very simple. However, when we analyzed the software tools they built, we found that errors and legal misconceptions were rampant. Despite these flaws, the computer scientists were confident that their tools would be beneficial in practice. Seventy-nine percent of them stated that they would be comfortable with their tool replacing judges in bankruptcy court.Our content. Delivered.Join our newsletter on issues and ideas at the intersection of tech & democracyThank you!You have successfully joined our subscriber list.The DOGE team is displaying this engineering hubris to an extreme degree. They are racing forward with little care for existing laws and protocols. Without spending the time to understand government operations, Shedd has asserted that AI can replace many federal workers.What will happen if these plans for an AI revolution in government move forward? Already, we are seeing evidence that DOGE is repeating the same mistakes that have plagued other instances of government automation.Most notably, DOGE will likely make many baseless judgments of waste and fraud based on AI results. This outcome will be partly due to Musks own wildly over-expansive definitions of waste and fraud. He has stated, without any evidence and contrary to other estimates, that A trillion dollars can be saved just by addressing waste, fraud, and abuse. Given this viewpoint, DOGEs AI developers will surely be more worried about avoiding false negatives (overlooking an instance of fraud) than false positives (incorrectly labeling a contract as wasteful or fraudulent), particularly in cases where they may have a conflict of interest.One cause of mistaken accusations will be misinterpreting information in government databases. Government data is notoriously complex. Understanding its meaning requires detailed knowledge of how that data is collected and used. An early instance of this error came when Musk tweeted in outrage about millions of people older than 100 supposedly receiving Social Security checks. His post included a screenshot of a simple analysis of the Social Security database, counting the number of people marked alive at different ages. Although his analysis correctly counted the data, Musk misinterpreted what it meant. Millions of individuals over 100 are marked as alive due to technical limitations and gaps in the data. Crucially, almost none of them are receiving checks. In fact, several years ago, the Social Security Administration considered updating its systems for annotating deaths but decided that the cost of the upgrade would not be worth the benefits. Had Musk simply asked Social Security experts about the data, he could have gained a correct understanding. Instead, he jumped confidently to an incorrect conclusion.Another likely cause of mistaken fraud accusations is AIs inability to interpret the meaning of contracts and the relationships between them. Governments often divide work into multiple contracts, each serving different aspects of larger programs. These contracts may look redundant to an AI tool when, in fact, they are distinct and complementary. To be sure, there can be benefits to streamlining government contracts. But, that requires carefully reconstructing contracts to serve the underlying policy goals, not sending existing contracts to the woodchipper.DOGE has already demonstrated a pattern of misinterpreting contracts, leading it to vastly overstate the savings it has generated. On February 19, DOGE posted a wall of receipts, boasting $16.5 billion in savings from canceled contracts. Their calculations were riddled with mistakes. For instance, many canceled contracts had ended years ago or were terminated under prior administrations. Since their initial post, DOGE has updated 40% of the contracts with more accurateand smallervalues. These updates included deleting the five largest contracts for which DOGE had initially claimed credit. That $16.5 billion now looks more like $8 billion, and additional errors remain uncorrected.To make matters worse, DOGE is making these decisions in an opaque and unaccountable manner. It is unclear what role AI has played in the actions that DOGE has already taken. Moving forward, an AIs classification of fraud could lead to contracts being canceled immediately without any notice or appeals process. When this action harms a person or organization, they may not be able to get a clear explanation of what happened aside from the AI said so.Of course, for Musk and his allies, these problems may be features rather than bugs. The AI will help them rapidly cut the federal budget and assert their dominance over the government. So what if many fraud accusations are incorrect and some essential government operations get waylaid?None of this is to say that AI cannot be a helpful tool for improving government services and operations. But, achieving that goal requires taking the opposite approach of DOGE.During my own time working as a data scientist in local government, I quickly learned that my technical know-how was wildly insufficient on its own. To advance socially beneficial projects, I had to recognize the limits of technology and avoid the assumption that the most cutting-edge technical tools would be the most useful in practice. I also had to appreciate the wisdom of existing staff. I relied on their guidance to make sense of government processes and datasets. I needed to view my work as augmenting the work of existing teams, not replacing them.The most unmistakable evidence that Musk and DOGE are not committed to thoughtfully developing technology that improves government is that they are dismantling the teams that have been pioneers in this work. Since it was founded in 2014, the United States Digital Service (USDS) has generated benefits such as improving healthcare access for 18 million veterans and enabling millions of rural households to access affordable high-speed internet. However, DOGE quickly fired a third of USDS staff. A few days later, another group of USDS staff resigned, writing that DOGEs actions are not compatible with the mission we joined the United States Digital Service to carry out: to deliver better services to the American people through technology and design. Then, last weekend, DOGE eliminated 18F, a group that helps federal agencies improve their technology. This move came just weeks after Shedd told 18F staff that they represented the gold standard of government technology developers.Given the frantic pace and hubris of Musk and the DOGE team, I fear that this AI incursion will do incredible damage to federal operations and the people who depend on them. But, at the very least, those of us who teach courses about technology and public policy will have a new case study of government AI gone wrong to share with our students.
    0 التعليقات ·0 المشاركات ·65 مشاهدة