Articles

Browse all AI articles. Use filters to narrow down results.

Showing 17981 to 18000 of 31030 entries.

Most Recent Articles

Business Insider
newsbusiness insideropenaiclaudeanthropicgoogleagentagents

Mark Cuban and Anton Osika at a conference in Paris. Bloomberg/Getty Images Mark Cuban says Lovable's add-on services change how it competes with the labs. Lovable's CEO says more customers are viewing the tool as their "AI cofounder." Vibe-coding firms face major competition from AI labs like Anthropic and OpenAI. Lovable once said it's competing with large AI labs, not other vibe-coding companies. Mark Cuban thinks it has a leg up on those guys, too. At a conference in Paris on Wednesday, the "Shark Tank" investor said that AI coding services like Lovable have add-on services that could prevent them from being replaced by AI labs. "When we question a lot of the front ends like Lovable, Replit, etc, we wonder if the foundational models, the big guys are going to just replace them," Cuban said at a talk with Lovable CEO Anton Osika. "But what you're saying is you have a base of data that is localized that applies not only to startups and entrepreneurs, but to any business that's looking to grow." Cuban, a Lovable investor, was referring to Osika's comments. The founder said that Lovable has moved from being a tool for writing code to a more all-inclusive platform — founders can incorporate their businesses and set up their payment systems, for example. "People saw us as an AI software engineer, the product. Now what we're increasingly seeing is that people see Lovable as their partner, as literally their AI cofounder," Osika said. Cuban's comments follow a period of brutal comparisons between products made by vibe-coding startups and Anthropic's Claude Code. After Anthropic released Opus 4.6 earlier this year, founders and developers said on X that they were ditching their Cursor and Lovable subscriptions for Claude Code. Founders and investors often compare AI applications, such as vibe-coding tools, to the capabilities of large language models. They are concerned that they may build or invest in a product that cannot defend itself against a new feature from Anthropic or OpenAI. Lovable, too, has acknowledged this threat. In a March podcast, Lovable's head of growth, Elena Verna, said that other vibe-coding players are not her biggest concern. "I always worry about the big boys and girls in the world," Verna said. "So, OpenAIs, Anthropics, Googles, Apples, more so than our competitors that spring up from the bottom or from sideways." This is because the distribution power of these tech giants and frontier labs in the market is unparalleled, she said. Read the original article on Business Insider

Techmeme
techtechmeme

Foo Yun Chee / Reuters: The EU General Court dismisses Apple's legal challenges against its designation as a DMA “gatekeeper” for its App Store and iOS; Apple filed the case in 2024  —  Apple (AAPL.O) lost its challenge on Wednesday against landmark EU rules that designate its app stores and operating system iOS …

Business Insider
newsbusiness insiderragpolicy

Getty Images; Alyssa Powell/BI The kids aren't all right. Gen Zers are graduating into a daunting job market. The housing market is a mess. Typical milestones — getting married, having children, buying a home — are drifting further into the future. If there's one stat that neatly sums up the current case of arrested development, it's that nearly a fifth of 25- to 34-year-olds live with their parents or grandparents, a record high, per a recent analysis of census data by John Burns Research and Consulting. Rather than renting their first apartments or grabbing the keys to a starter home, today's young adults are shacking up with their parents in droves. By and large, young adults tell pollsters that they still harbor dreams of homeownership. But the affordability hurdles are steep, and social norms have changed, too — living with parents well into adulthood may no longer be the calamity it was for, say, Seinfeld's George Costanza, whose move back home in a 1993 episode was portrayed as nothing less than rock bottom. If all of this sounds familiar, it's because millennials traveled a similar path. Overeducated and underemployed in the wake of the Great Recession, they spent their early careers scrounging for jobs and crashing with their parents to save cash. For many, the prospect of homeownership seemed distant. One 2015 headline from this very website summed up the gloomy outlook: "America's 'children in the basement' won't be moving out anytime soon." The millennial comparison, however, may actually offer some hope for Zoomers. Now in their early 40s, the oldest millennials own homes at roughly the same rate as their Gen X predecessors. The goalposts may have shifted back, but they didn't disappear. That's no guarantee, however, that this game of catch-up will turn out the same way for Gen Zers, the oldest of whom are closing in on 30. Every generation encounters its own speed bumps en route to the American dream. Gen Zers, however, are caught in the middle of several unique forces, from the country's slowing population growth to the death of the traditional starter home. Their housing choices today could reverberate for decades to come. Millennials were once the poster children of stunted adulthood. The slow economic recovery after the financial crisis left the cohort struggling to find its footing in the housing and job markets. In 2016, roughly 16% of 25 to 34-year-olds lived with relatives, compared with 11% a decade earlier. The prevailing narrative in those days was, "Millennials will never become homeowners," recalls Mark Fleming, the chief economist at the title company First American. Indeed, while nearly half of baby boomers owned their homes at 30, millennials' share at the same age was a paltry 33%. Since then, however, elder millennials have "mostly caught up," says Eric Finnigan, a demographer at John Burns and a member of the cohort himself. A 42-year-old whose career might have appeared dead on arrival in 2008 has roughly the same odds of owning their home as a Gen Xer did at the same age. (Boomers, notably, were well ahead of both groups.) "It's just a delayed life stage, not a lost life stage," Finnigan tells me. Younger adults have yet to reach that point. The share living with parents jumped in 2020 during the height of the pandemic, but trended downward in the next couple of years as borrowing rates plummeted and companies hired with gusto. "The job market for young people was white-hot," Finnigan tells me. "Anybody who wanted to find a job could probably find a job, and that wasn't true 10 years before." Then came the spring of 2022, when the era of ultra-cheap home loans came to an abrupt end. Interest rates spiked, making all kinds of borrowing more expensive, and the job market hit the skids. The homeownership rate for young adults fell for two consecutive years, erasing the steady gains made from 2017 to 2022. A little less than 30% of 25-to-24-year-olds owned their home as of 2025, down from about 40% in 1990, an Apartment List analysis of census data found. The sudden shift in the market outlook was compounded by longer-term trends, like the shrinking pool of "starter homes" — cheaper, entry-level places that allow younger homeowners to start amassing equity. Given the rising costs of land, materials, and cutting through red tape, homebuilders are pivoting to more lucrative projects that favor older buyers with deeper pockets: A Business Insider analysis of census data found that by late 2022, hardly any newly built homes in the US were selling for less than $200,000. Meanwhile, more than 60% of new homes were priced above $400,000, double the share from a couple of years earlier. "It just became a lot more difficult to build entry-level homes," Rob Dietz, the chief economist for the National Association of Home Builders, tells me. "There's a lot fewer of those available." For young people, the hurdles to getting keys to their own place go beyond the dearth of cheap new builds. Renting a non-mom-and-dad-owned place is more expensive: The typical rent nationwide jumped by nearly 30% between 2020 and 2024, Zillow data shows. And a study a few years ago by the Federal Reserve Bank of St. Louis found that 36% of older Gen Zers were sitting on student debt, compared to 31% of millennials at the same life stage. Chiamaka Onyemelukwe, 35, has lived with her mother in Alameda, California, since finishing her master's program in 2018. She dreams of working as a storyboard artist for a TV or film company, but has spent the past couple of years job-hunting after being laid off in late 2023. "It's just hard to find something stable right now," Onyemelukwe tells me. She recognizes that she's lucky to have a place to stay in her mother's two-bedroom rental, but she hopes the arrangement isn't permanent. "I definitely want to get a house one day," she says. For now, though, she's dipping into savings and relying on her mother for support. "When you're in your head a lot, and you're thinking about the situation you're in, like not having work and not having a place of your own," Onyemelukwe tells me, "it's going to make it difficult for you to focus on what really matters outside of that." While some younger adults may be prioritizing careers or simply putting off the whole white-picket-fence thing, the stark rise in young adults living with their folks, combined with the mounting barriers to entry for wannabe homeowners, sends a clear message to economists like Dietz. "I actually think there is reason to be concerned," Dietz tells me. One person who may see fewer reasons for consternation is Elizabeth Gomes, a 29-year-old in San Diego who works in public relations. Gomes lives the kind of rich, full life enjoyed by young professionals in big cities around the country. She travels, dines out weekly with friends, and enjoys going to concerts and Disneyland. She also lives with her parents. "I have a ton of friends that are in very similar situations," Gomes tells me. Some have settled back into their childhood bedrooms, while others are living on their parents' properties. All have jobs. "It's just the situation that's happened, or that we've chosen because of how we want to live," she says. Of the roughly 7.5 million 25 to 34-year-olds living at home, seven in 10 are employed, according to a Realtor.com analysis of census data. The prevalence of those working but still living at home suggests "this is a housing cost story, not a jobs story," writes Hannah Jones, a senior economist at Realtor.com. Gomes has scanned the nearby rental market and decided that moving out wouldn't be worth the tradeoffs — less travel, more meals in, fewer concert tickets purchased. The $700 her parents charge for monthly rent is a fraction of what she'd pay if she lived alone, given that the median rent for a one-bedroom in San Diego stretches to nearly $2,000, per Apartment List. Plus, she likes hanging out with her parents. They've been roommates since she was a sophomore in college, when she decided to give up dorm life in favor of a cheaper (and calmer) alternative. "I don't really have any shame in living at home," Gomes tells me. Finnigan, the demographer at John Burns, spends a lot of time talking to young people about their living situations, and he's come away surprised by how much the Seinfeld-era caricature of the stay-at-home adult has faded. "It's way more of a normal thing," he tells me. "There's not a hit to status, or someone's self-esteem." "I don't take it as an embarrassing kind of thing anymore, really. Because I just feel like it's the new reality.Nick Chapman, 30-year-old in Simi Valley, California Nick Chapman, 30, works in film marketing and recently moved home to the Southern California suburb of Simi Valley. Though he could afford the $1,550 monthly rent at his old apartment in Los Angeles, the cost pressures of living on his own left him feeling "stagnant," he tells me, with little room to save or splurge. Chapman grew up on movies like "Stepbrothers," which played the idea of grown men living at home for laughs: "It was kind of a pathetic thing, and I'm obviously like, 'I hope that's not me,'" Chapman says. But his parents had always made clear that he'd have a place to stay if needed, and his older brother had also spent some time at home after school. So far, he's enjoying the comforts (and savings) of home. Like the other young adults I spoke with, Chapman still hopes to buy a home one day. "That's the ultimate goal," Chapman tells me. But maturing, he says, has also meant accepting that "everyone is on their own clock." "I don't take it as an embarrassing kind of thing anymore, really," he says. "Because I just feel like it's the new reality." Odds are that you know others in this boat. Maybe you find yourself back in your childhood bedroom, or perhaps you're on the flip side, pulling double duty as a parent-landlord. My own 25-year-old brother has spent the past few years living at home. He has a good job, but making it in a costly city like Austin is a stretch for workers of all stripes. I'm glad he has a place to stay, and I often envy the moments he gets to spend with my parents while I settle for FaceTime calls. I wouldn't trade it for my experiences living on my own, though, and I'll admit that I hope he soon gets the same chance. Mark Fleming, the economist at First American, offers a counterpoint to all the handwringing. "I think a lot of times the data is interpreted as, 'Well, they will never become homeowners,'" Fleming tells me. "That's not true." Fleming counseled similar patience when elder millennials were in dire straits, and he points to their rising homeownership rate as evidence that a delay doesn't equate to total discouragement. Extend some of these chart lines into the future, and the group may surpass Gen X in terms of homeownership. "It just took time," Fleming says. This story has been playing out for decades — for the vast majority of their adult lives, even baby boomers appeared to lag behind the silent generation before them on the homeownership front. The economy shifts, and priorities change. If there's one constant, it's the younger generations' knack for stirring concern among the silver-haired crowd. All signs point to the idea that once Gen Zers find their financial footing, they're likely to jump into the housing market. Finnigan views the delay as pent-up demand — people waiting in the wings to live on their own once conditions improve or the timing feels right. "They'll catch up at some point," Finnigan tells me. "It's just a matter of when." James Rodriguez is a correspondent on Business Insider's Discourse team. Read the original article on Business Insider

Business Insider
newsbusiness insideranthropicmeta

Cursor's David Pan wrote that two pizzas — the amount advised by Jeff Bezos — may be too much food to fill a team in the AI era. Alexander Spatari/Getty Images Cursor field CTO David Pan wrote a eulogy to Jeff Bezos' "2-pizza rule" on X. Bezos initially proposed that a meeting should always be small enough to feed with two pizza pies. "In the AI era, two pizzas is too much pizza," Pan wrote. In the "tiny team" era of AI, your meeting doesn't need as many pepperonis. Jeff Bezos popularized the "two pizza rule," the idea that any Amazon meeting should remain small enough to be fed with two pies. The rule also influenced team sizes These days, many workplace teams have gotten even smaller. David Pan, Cursor's field CTO, put it simply: "RIP to the two pizza team." Pan wrote on X that Bezos created an "all-time great metaphor," one that shaped engineering for over twenty years. "He was right about small teams," he wrote. "But in the AI era, two pizzas is too much pizza." RIP to the two pizza team. Jeff Bezos gave us an all-time great metaphor: keep the team small enough to feed with two pizzas. It shaped how engineering orgs got built for 20+ years. He was right about small teams. Still is. But in the AI era, two pizzas is too much pizza. — David Pan (@davep) July 6, 2026 In the comments, users debated what the new metaphor should be. One proposed the quarter-pizza team; another said it was the three-slice team. Amazon writes that "ideally, this is a team of less than 10 people: smaller teams minimize lines of communication and decrease overhead of bureaucracy and decision-making." "This allows two-pizza teams to spend more time focusing on their customers, and constantly experimenting and innovating on their behalf — the biggest priority of high-performing teams at Amazon," the company's website says. Some still follow the rule. Spencer Rascoff, the CEO of dating app company Match Group, told Business Insider that he was a fan in March. Pan knows the rule firsthand. Before Cursor, he worked at Amazon from 2011 to 2014 as an engineering manager, according to his LinkedIn. There are also competing pressures on team sizes. Some managers want smaller teams, with fewer people doing more tasks. Others want bigger, flatter teams with fewer layers of management. Those wide teams may need more pizzas. Of course, many users commented that they themselves could devour the pies — even on a shrunken team. "I've learned that you all can eat a lot of pizza," Pan commented in the wake of his initial post. Read the original article on Business Insider

Business Insider
newsbusiness insidervision

Ukrainian engineers repair a Vampire hexacopter, one of the most commonly used drones for bombing missions. Viacheslav Madiievskyi/Ukrinform/NurPhoto A Ukrainian drone unit said it attacked over 230 pieces of Russian artillery in two days. It used a special munition designed to target the barrels of howitzers and other field guns. Fedorov said it was part of a two-stage operation involving intense flurries of drone strikes. An elite Ukrainian drone unit said on Tuesday that it had used a new anti-howitzer munition to destroy over 170 pieces of Russian artillery within two days. The Lasar Group, part of the National Guard of Ukraine, said in a statement that the special operation in early June involved over 800 drones deployed along the war's eastern and southern fronts. "As a result, 231 enemy artillery pieces were hit, 171 of them destroyed. This is equivalent to roughly 10 to 14 artillery battalions, or about three artillery brigades," the group wrote. It published multiple clips of overhead drone footage that appear to show fortified, camouflaged howitzer and gun positions coming under attack. Several of the clips show quadcopters that appear to act as bombers flying above the gun barrels before the positions are engulfed in explosions. Business Insider could not independently verify when and where the footage was shot. The Lasar Group said the missions were planned by its analysts who worked with several other units to sift through large tranches of intelligence collected on the gun locations. A "special munition" was also created for these strikes, the group said. "The group's engineers developed this warhead specifically to destroy howitzers by targeting their gun barrels, making the weapons inoperable," its statement read. Ukraine's defense minister, Mykhailo Fedorov, said on Tuesday that the flurry of drone strikes was part of a multi-phase campaign called Operation Auchan. The first stage had seen Ukrainian drone units attack 980 Russian targets over three days in 2025, in an effort to force the Kremlin to withdraw its heavy armor from a mechanized offensive that year. Across both phases, the operation has resulted in 1,180 targets being hit in five days, he said. Fedorov also highlighted the special payload designed for the second round of drone missions, mentioning a "newly developed munition specifically targeting enemy artillery." Neither the Lasar Group nor Fedorov provided more details on the warhead. The National Guard of Ukraine did not immediately respond to a request for comment sent outside regular business hours by Business Insider. The Ukrainian defense minister said that that 10% to 20% of Ukraine's battlefield casualties are caused by artillery fire. Destroying 171 pieces of artillery with drones in two days is significant at the war's current pace. Ukraine's Unmanned Systems Forces, for example, recorded 417 howitzers and 57 other pieces of artillery struck by 16 of its top units in the entire month of June. A total of 61 of these weapons were confirmed as destroyed — a claim that Ukrainian troops typically only make after having video footage of the attack verified by analysts. Founded by television producer Pavlo Yelizarov, the Lasar Group is one of Ukraine's most prominent drone units and has been credited by the Kyiv officials with destroying $15 billion worth of Russian equipment. Read the original article on Business Insider

Datafloq
cybersecurityagentic aiai agentsai securitycloud securitycve-2025-3248cyberattackdata extortiondatabase securityincident responsejadepufferlangflowllm securityminionacosransomwaresysdigthreat researchvulnerability managementdatadatafloqopenaianthropicgeminillmragagentvision

An AI agent broke into a production database, corrected a failed login attempt, and wrote a … JADEPUFFER Automates the Ransomware Kill Chain: What the First Documented Agentic Extortion Attack Means for DefendersRead more The post JADEPUFFER Automates the Ransomware Kill Chain: What the First Documented Agentic Extortion Attack Means for Defenders appeared first on .

Product Hunt
startupproduct huntgoogle

Orbit for Mac

By Andrew Kwak

Discussion | Link

Product Hunt
startupproduct huntagentagents

Prefactor

By Rohan Chaubey

Discussion | Link

Business Insider
newsbusiness insiderrag

Hyebin led a once-top girl group. The Chosunilbo JNS/Imazins via Getty Images Hyebin is the leader of Momoland. At its prime in 2018, it was a top-tier K-pop girl group. In a new video, the singer broke down how much idols like her earn per event. The math, according to Hyebin, means most idols may be far from rolling in cash. K-pop idols are often decked out in luxury clothes from brand deals, but the reality of their careers may look pretty grim. Hyebin, the leader of the six-member girl group Momoland, took to Instagram this week to dispel the illusion that every K-pop idol is rolling in cash. "It has already been more than 10 years since I became an idol," she said in a video posted on Monday. "Many people probably wonder how much money idols actually make. It may seem like they earn a crazy amount of money, but they don't." Hyebin started the video with an explanation about how many idols — especially those who debut in midsize or small agencies — start their careers with hundreds of thousands of dollars in debt. Costs mount during the trainee period that all K-pop stars must go through before their debut, including lessons, food, and housing to practice room rentals. It's all "billed straight to you after debut" like a "post-paid system," she said. Hyebin added that reaching the top of the charts or being part of a popular group could mean an idol is "still poor." "In general, idols split all expenses 50-50 with the agency," she said. Releasing a song means buying a track worth thousands from a composer. Music videos cost money to make, too, and the idols shoulder the cost out of pocket. "Every time we shoot a music video, tens of thousands of my own money vanishes," she said. Even active idols who are constantly doing events and activities may not be raking in cash. The average booking fee for an idol event is around $38,000, she said, but the company takes a 50% cut. Then there's manager salaries, gas, and the necessary event prep, like hair and makeup that the idols need to pay part of the cost for, too. And after all that, the profit is split among the band members. Momoland had nine members at one point. Some bands, like Seventeen, have 13 people. "Every time I do an event, I walk away with around 2 million Korean won," she said. That's around $1,300. "The money you worked your butt off to make, it sits in a pile and then goes straight into investing in the next music video," she said. "Then, boom, you're back in the negatives." "It's a literal infinite Dormammu loop," she joked, referring to the "Marvel" villain. "I'm telling you, the money made a U-turn before it even touched my bank account." Hyebin acknowledged that some idols get rich, but the chances are minuscule. Only 1% of people who try to become K-pop idols make it to debut, and only the top 1% of those who debuted make "any real money," she said. K-pop is a multibillion-dollar global industry. In April, Hybe, the mega K-pop label behind BTS and their peers, Seventeen, reported its highest-ever first-quarter revenue of 698.3 billion Korean won, or around $473 million. Hyebin's Momoland was a top girl group in 2018. It disbanded in February 2023, when the members' contracts with their previous agency, MLD Entertainment, expired. In 2025, six members of the group, including Hyebin, reunited to work together under a new label, but with the same band name. Read the original article on Business Insider

ArXiv Machine Learning
cs.lgcs.aics.roresearcharxiv machine learning

Geometry-Aware Infrastructure-Anchored Denoiser for UWB Sensing and Work-Zone Reconstruction

By Weizhe Tang, Jiaxi Liu, Junwei you, Steven T. Parker, Pei Li, Sikai Chen, Meng Ran, Bin Ran

arXiv:2607.05449v1 Announce Type: new Abstract: Accurate work-zone geometry perception is critical for intelligent transportation systems, and ultra-wideband sensing offers a low-cost approach for infrastructure-aided reconstruction. However, outdoor UWB ranging is often degraded by non-line-of-sight propagation, burst noise, and long-tail errors, which can distort downstream spatial reconstruction. We present GAIA, a geometry-aware, infrastructure-anchored learning framework that couples temporal range modeling with latent anchor-layout estimation and deterministic distance projection. GAIA preserves range denoising as the supervised task while orienting the learned distances toward boundary-consistent reconstruction. We evaluate GAIA on a real-world outdoor UWB dataset with synchronized UWB, GNSS, and IMU measurements, and further test robustness using a real-data-calibrated stress-test simulator. GAIA achieves the lowest overall range MSE and highest polygon IoU among evaluated filtering-based and learning-based baselines, reducing MSE by 18.4% and improving polygon IoU by 15.5% over PoseMLP. These results show that geometry-aware range denoising provides an effective path toward spatially coherent work-zone reconstruction.

ArXiv Computer Vision
cs.cvcs.lgeess.ivresearcharxiv computer vision

A Task-Driven Evaluation of UAV Detection and Tracking under Synthetic Fog

By Amir Pouladi, Vesal Ahsani, Haijun Li, Homayoun Najjaran, Afzal Suleman

arXiv:2607.05467v1 Announce Type: new Abstract: Fog severely degrades the visibility of small unmanned aerial vehicles (UAVs) in skydominant, long-range imagery, reducing the reliability of downstream detection and tracking. This paper presents a task-driven evaluation framework that links depth-aware synthetic fog generation, image restoration, object detection, and tracking within a unified pipeline. Given the practical difficulty of collecting and annotating foggy UAV scenes, synthetic fog is generated from real clear-weather outdoor images containing UAV targets using monocular depth estimation and the atmospheric scattering model. Representative restoration methods from classical, convolutional neural network (CNN)-based, and transformer-based families are first compared, after which the selected restoration model is integrated into the downstream perception pipeline. Detection is evaluated under both clean-only and fog-inclusive training regimes using multiple detector variants, while tracking-by-detection is assessed on clean, foggy, and restored video sequences. Beyond image-level restoration metrics, the study evaluates how fog and restoration affect detection robustness and tracking performance. The results show that fog substantially degrades both detection and tracking, primarily through increased missed detections. Fog-inclusive training provides the most consistent improvement in robustness, whereas test-time restoration is most beneficial when the detector has been trained only on clean imagery. These findings show that restoration quality does not necessarily translate into proportional gains in downstream perception and therefore should be evaluated jointly with detection and tracking performance.

ArXiv Computer Vision
cs.cvcs.airesearcharxiv computer visionraginference

Rendering-Aware Bayesian 3D Gaussian Splatting with Native Uncertainty and Adaptive Complexity Control

By Gaoxiang Jia, Vikram Appia, Junzhou Huang, Xinlei Wang

arXiv:2607.05522v1 Announce Type: new Abstract: 3D Gaussian splatting (3DGS) is a strong representation for real-time novel-view synthesis, but its standard training pipeline relies on point estimates and hand-tuned heuristics, providing no native uncertainty or principled complexity control. This is most limiting under sparse views or fixed acquisition budgets, where a model must identify weakly supported geometry and select informative views. We introduce a rendering-aware Bayesian 3DGS framework that tracks Gaussian geometry with a Normal-Inverse-Wishart posterior over means and covariances using renderer-derived surrogate summaries. An optional Dirichlet-process extension adds a probabilistic component-usage signal, and the training schedule makes the closed-form versus approximate inference boundary explicit. Re-rendering posterior geometry samples yields native predictive uncertainty for interval calibration and active view selection. In a fixed-budget 16-to-32 active-view task, native NIW acquisition improves PSNR by +0.453 dB and LPIPS by -0.0146 over a scoring-only 3-member standard-ensemble baseline, winning 29/39 scene-seed pairs and 10/13 scene means; it also improves over PPU-style (+0.355 dB) and NIW-proxy (+0.401 dB) acquisition. NIW native intervals reduce 95% coverage error by about 17x relative to a shared proxy (0.046 vs. 0.796) and are about 10x closer to nominal coverage than a 3-member deep ensemble (0.047 vs. 0.454) at roughly one-third the training cost. As a reconstruction compatibility check, paired NIW-vs-standard analysis over 39 scene-seed runs yields +0.030 dB PSNR with 1.6% additional training time. These results position Bayesian 3DGS as a practical probabilistic scene representation for decision-facing tasks such as active view selection.

ArXiv Computer Vision
cs.cvcs.aics.crcs.lgresearcharxiv computer visionvision

Statistical Adversaries: Natural Backdoor-like Features in Vision Datasets

By Paul K. Mandal, Pavan Reddy, Tristan Malatynski

arXiv:2607.05516v1 Announce Type: new Abstract: Model-specific adversarial attacks have been extensively studied. We study a different failure mode: naturally occurring statistical signals in vision data that can behave like backdoor-like triggers without being maliciously inserted. We call these signals statistical adversaries. We analyse Imagenet to find patterns that are strongly linked to certain labels. We then use statistical controls to remove random correlations from our candidate signals. Finally, we demonstrate that these signals directly and predictably alter model predictions. These statistical adversaries are more targeted than generic corruptions and transfer across different model architectures. This suggests that some vulnerabilities are driven by dataset structure and distribution rather than a single model's idiosyncrasies. We conclude that ordinary datasets can contain exploitable adversarial surfaces even in the absence of poisoning, and suggest that dataset audits should treat spurious structure not only as a source of bias or interpretability failure, but also as a latent attack surface for vision models.

ArXiv Computer Vision
cs.cvresearcharxiv computer visionmultimodal

Ground3D-LMM: Fine-Grained 3D Point Grounding and Spatial Reasoning with LMM

By Amol Harsh, Zongyan Han, Jean Lahoud, Ye Liu, Rao Muhammad Anwer, Hisham Cholakkal, Salman Khan, Fahad Khan

arXiv:2607.05493v1 Announce Type: new Abstract: Natural-language queries about 3D environments become actionable when responses are verifiable and metric. Verifiability requires explicit grounding to the referred 3D region, while metric answers report physical measurements in real-world units (e.g., size, thickness, clearance, and distance). Existing 3D large multimodal models (LMMs) approaches remain limited: conversational systems typically respond without explicit 3D grounding, while 3D grounding models are not designed for interactive, metric-aware dialogue. In this paper, we present Ground3D-LMM, a unified model that takes a point cloud and an optional RGB image as input and supports 3D spatial conversation with (i) point-grounded responses and (ii) metric numeric outputs at both object and part granularity, including multi-object queries. To evaluate this intersection of grounding and measurement, we define the 3D Grounded Measurement task, which requires predicting the referred 3D region and the corresponding metric quantities in real-world units. We introduce a large-scale dataset built on ScanNet and ScanNet++ datasets with dense object and part annotations and roughly 2.5M question-answer pairs spanning eight tasks, along with a manually verified test set. Extensive experiments on multiple datasets and tasks show that our proposed Ground3D-LMM model provides a strong baseline for grounded, metric-aware 3D conversational understanding. Our dataset and model are publicly available.

ArXiv Computer Vision
cs.cvcs.airesearcharxiv computer vision

Binocular Gaze Estimation with Single Camera and Single Light Source

By Tongbing Huang, Yang Fu, Yunfei Wang, Zhaocan Wang

arXiv:2607.05473v1 Announce Type: new Abstract: According to commonly consented theories, the minimum hardware requirement for gaze tracker is one camera and two light sources to realize gaze estimation with free head movements. However, in some scenarios such as eye tracking on mobile devices, it is preferable to use less components, especially light sources. We propose a gaze estimation method with one camera and one light source. A "virtual light source" is introduced, which is geometrically placed symmetrically to the real light source with respect to the camera, and generates a "virtual glint" in the acquired image. We estimate the "virtual glint" by exploiting the relationship between the distance between two pupils and two glints in the captured image, and estimate the gaze with polynomial regression assuming two light sources are available. A new normalization factor for regression method is verified, which turns out to be practical for one-glint system. The performance is proved to be acceptable, while degradation is noticed compared to system with two actual light sources.