arXiv:2605.15309v1 Announce Type: new Abstract: Despite remarkable progress, image generation is far from solved. The dominant metric, FID, conflates sample fidelity with mode coverage and is close to being saturated. Yet a model can still exhibit mode collapse while achieving a low FID, since a handful of sharp, near-duplicate images can outscore a model that faithfully covers the full data distribution. We argue that precision and recall are essential complements to FID, and that because FID is already saturated, the more meaningful goal is to improve diversity and coverage. Achieving high recall requires a model that explicitly prioritizes mode coverage, unlike most generative models, which optimize sample fidelity. We introduce RTM, which replaces the single-pass latent mapping in style-based generators with an iterative refinement process, and show that this consistently improves both quality and diversity. Integrated with Implicit Maximum Likelihood Estimation (IMLE), which optimizes mode coverage by design, RTM achieves the highest precision and recall among current state-of-the-art approaches while maintaining competitive FID, with improvements across CIFAR-10, CelebA-HQ at 256x256, and nine few-shot benchmarks. RTM also improves StyleGAN2 and StyleGAN2-ADA on CIFAR-10 and AFHQ-v1 at 512x512, demonstrating that the benefit is not specific to IMLE. Unlike flow-matching baselines that achieve competitive FID at the expense of coverage, recursive refinement improves both quality and diversity simultaneously.

arXiv:2605.15325v1 Announce Type: new Abstract: Vision-language models (VLMs) have shown strong performance in video anomaly detection (VAD) while providing interpretable predictions. However, existing VLM-based VAD methods suffer from a fundamental mismatch between training and inference in both data distribution and model configuration. First, most approaches rely on static post-training adaptation, limiting generalization under distribution shifts such as unseen environments or anomaly types. Second, they train VLMs on sparse frames from long videos, but perform inference on densely sampled short segments, creating inconsistencies between training and testing. To address these limitations, we propose COPRA, a conditional parameter adaptation framework for VLM-based VAD. Instead of fixed prompts or shared parameter updates, COPRA generates input-specific parameter updates to dynamically adapt a frozen VLM for each video segment during both training and inference. Experiments show strong performance on standard VAD benchmarks, consistently outperforming static baselines in both in-domain and cross-domain settings. Moreover, COPRA generalizes beyond VAD to unseen tasks such as multiple-choice Video Question Answering and Dense Captioning. These results highlight COPRA as an effective weight-space generation framework for scalable, adaptive, and context-aware video understanding. The code will be released at https://github.com/THE-MALT-LAB/COPRA

arXiv:2605.15326v1 Announce Type: new Abstract: Reliable detection of humans beneath forest canopy remains a difficult remote-sensing challenge due to sparse, structured, and viewpoint-dependent occlusion. This paper presents a multimodal proof-of-concept pipeline that integrates three complementary approaches: (i) experimental evaluation of LiDAR returns through vegetation to assess the feasibility of active sensing, (ii) visible--thermal image fusion using a multi-scale transform and sparse-representation framework to enhance human saliency, and (iii) synthetic-aperture image formation via Airborne Optical Sectioning (AOS) to suppress canopy clutter. A YOLOv5 detector is fine-tuned on the Teledyne FLIR thermal dataset and evaluated on thermal and fused imagery. Results show that the tested terrestrial LiDAR configuration provides limited penetration for object-level detection, while visible--thermal fusion improves target visibility in low-contrast scenes and AOS enhances ground-plane detection in synthetic forest imagery. The fine-tuned YOLOv5 achieves a mean average precision of $\sim$0.83 on the top three FLIR classes. These findings establish an initial baseline for UAV-deployable search-and-rescue and surveillance systems operating in forested environments, and motivate future work on dedicated forest datasets and real-time multimodal integration.

arXiv:2605.15342v1 Announce Type: new Abstract: Video reasoning models are a core component of egocentric and embodied agents. However, standard benchmarks for assessing models provide only evaluation of the output (e.g. the answer to a question), without evaluation of intermediate reasoning steps, and most provide answers only in the text domain. We introduce Minerva-Ego, a benchmark for evaluating complex egocentric visual reasoning. We extend recent high-quality video data sources recorded from egocentric / embodied settings with a set of challenging, multi-step multimodal questions and spatiotemporally-dense human-annotated reasoning traces. Benchmarking experiments show that state-of-the-art models still have a large gap to human performance. To investigate this gap in detail, we annotate each reasoning trace in the dataset with the objects of interest required to solve the question, as spatiotemporal mask annotations. Through extensive evaluations, we identify that prompting frontier models with hints of 'where' and 'when' to look yields substantial improvements in performance. Minerva-Ego can be downloaded at https://github.com/google-deepmind/neptune.

arXiv:2605.15375v1 Announce Type: new Abstract: Remote sensing change detection (RSCD) aims to localise changes between two images of the same geographic region. In practice, change masks often follow region-level annotation conventions rather than purely local appearance differences, making them context-dependent and occasionally ambiguous. Most state-of-the-art methods utilise per-pixel discriminative classification, which produces a single prediction per input and fails to explicitly model the changed region as a coherent whole. A natural alternative is generative formulation, which can model a distribution of plausible masks, enabling sampling to capture ambiguity and encourage global consistency. However, existing generative RSCD approaches typically lag behind strong discriminative baselines due to the high computational cost of pixel-space generation and the complexity of their conditioning mechanisms. To address the limitations of prior discriminative and generative methods, we propose ChangeFlow, a generative framework that reformulates change detection as the synthesis of a change mask in latent space via rectified flow. ChangeFlow is guided by a structured yet lightweight conditioning signal, and its stochastic design naturally supports sampling-based prediction ensembling. Namely, aggregating multiple predicted change masks improves robustness, while sample agreement provides a practical confidence estimation that highlights ambiguous regions. Across four benchmarks, ChangeFlow achieves an average F1 of 80.4\%, improving by 1.3 points on average over the previous best method, while maintaining inference speed comparable to recent strong baselines. Project page: https://blaz-r.github.io/changeflow_cd

arXiv:2605.15235v1 Announce Type: new Abstract: Multimodal physiological data powers clinical AI systems from intensive care units to wearable devices, but sensors routinely fail in practice. Two failure modes are common: modality missing, where an entire channel is absent, and within-modality missing, where a contiguous time segment is lost. No existing benchmark evaluates multiple fusion architectures under both failure modes at controlled severity levels across diverse clinical datasets. We present MuteBench, a benchmark covering 9 datasets from 7 clinical domains, 6 fusion architectures, and 2 missing-data modes over 125,000 samples. Through this benchmark, we find that architecture family is the strongest predictor of robustness, outweighing parameter count. Channel-independent models tolerate modality missing well but can be sensitive to within-modality missing, especially on short sequences. Curriculum modality dropout protects reliably only up to the maximum dropout rate used in training. We also find that channel count, sequence length, and modality alignment jointly determine which failure mode poses the greater threat. Finally, a PTB-XL case study suggests that diffusion-based imputation can improve downstream classification under within-modality missing, with the largest gains for models whose expert routing is most sensitive to corrupted inputs, though broader validation across datasets remains an open direction. MuteBench provides practitioners with concrete guidance for both selecting existing architectures and informing the design of future robust multimodal fusion methods.

arXiv:2605.15362v1 Announce Type: new Abstract: Half a billion citation edges extracted from 100.7 million Ukrainian court decisions reveal that judicial citation structure encodes legal domain boundaries without supervision and predicts future legislative importance with near-perfect accuracy. We construct the first large-scale citation graph from the complete EDRSR registry (99.5 million full texts, 1.1 TB), extracting 502 million citation links across six types via regex on commodity hardware in approximately 5 hours, with precision of 1.00 on a 200-decision validation sample (95% Wilson CI: [0.982, 1.000]). Three principal findings emerge. (1) The degree distribution follows a power law (alpha = 1.57 +/- 0.008), placing the Ukrainian court network near the EU Court of Justice and below the US Supreme Court, with hub articles cited by millions of decisions. (2) Louvain community detection on the co-citation projection recovers legal domain boundaries (civil, criminal, administrative, commercial) with modularity Q = 0.44-0.55 and temporal stability (NMI = 0.83-0.86 across periods), constituting an automatically constructed legal ontology grounded in judicial practice. (3) Citation features predict top-1000 articles with AUC = 0.9984, substantially outperforming a naive frequency baseline (P@1000 = 0.655); temporal dynamics detect legislative regime changes as phase transitions and the 2022 invasion as a citation entropy spike (H: 11.02 -> 13.49) with emergent wartime legislation nodes. The citation-derived ontology is operationalized as the domain layer of a workflow memory system for LLM-assisted legal analysis, connecting to the ontology-controlled paradigm. The extraction pipeline, analysis code, and aggregated statistics are released as open data.

Discussion | Link

Wall Street Journal: Sources: US authorities in recent months have issued multiple information requests to Kalshi and Polymarket, many tied to wagers on events in Iran and Venezuela  —  Regulators are seeking information from Kalshi and Polymarket over wagers tied to political events and military operations

Madison McLauchlan / BetaKit: Quebec-based telecom software startup Gaiia raised a CA$54.8M Series B led by JMI Equity, bringing its total funding to ~CA$90M from backers including YC  —  Québec City-based telecom software startup Gaiia has raised a $40-million USD ($54.8-million CAD) Series B round as it hopes to invest an …

South Korea's local elections next month will be a test bed for how effective regulations might be to stymie the flow of deepfakes.

In this article, we create a real-time audio transcription application on of the RealtimeSTT library. The post Creating an Audio Transcription Application with RealtimeSTT appeared first on DebuggerCafe.

Tse Hao Guang started a literary publishing business while working as a researcher. Tse Hao Guang Tse Hao Guang is a researcher and writer in Singapore. He started a side project publishing literary pamphlets that spotlight emerging writers' works. He wanted to show that a creative venture does not have to bleed money. This as-told-to essay is based on a conversation with Tse Hao Guang, a 38-year-old researcher and writer based in Singapore. It has been edited for length and clarity. I started a creative publishing project to spotlight emerging writers, with the goal of not losing money doing it. My journey in literature began when I was young, traveling to meet my dad's side of the family in suburban Canada during the holidays. There was nothing to do, and I'd beg my uncle to drive all the way to the library, 20 minutes away, and borrow huge stacks of books. Since 2015, I've published two books of my own and edited several anthologies. It's very, very rare that you can survive just by writing books. When I was a freelance writer, I taught writing workshops at schools and universities and did copywriting gigs. It tired me out, because client writing and creative writing used similar parts of my brain. Now, I work full time as a strategic foresight researcher. Earning a stable salary has removed the pressure to make money from my writing, allowing me to pursue many creative projects. One of those projects is Paper Jam, an annual publication that spotlights Singaporean writers. Starting Paper Jam The big gap I saw for emerging writers was that when they want to showcase their work, there are only a few avenues. They can either publish in a journal alongside many other writers, so it's not their solo work, or amass enough stories to create their own manuscript, which has a much higher bar to entry. They would also need to print up to 1,000 copies for their first run, which is tricky to manage financially and inventory-wise. I noticed that Singapore didn't have a flourishing independent zine culture like other countries do. So I came up with the idea for Paper Jam, a literary pamphlets publishing venture. I received funding from Sing Lit Station, a Singapore-based nonprofit that supports local writers, and launched in 2024. Limited copies for exclusivity Paper Jam only produces 100 copies of each pamphlet to maintain exclusivity. Aditi Bharade The idea was that I'd pick three writers each year via an open call on Instagram, and each would write a 28-page pamphlet. For the last two years, I've put out the open call in November. While editing, I tend to over-suggest, giving them many edits. I expect them to push back and encourage them to disagree with me until we find a middle ground. I decided to print only 100 copies each of the three A5-sized pamphlets and not reprint if they went out of stock. The idea is that a pamphlet isn't meant to stay forever. The appeal of it is that it's a limited edition, something they might not be able to buy again. Then in September, I sell the pamphlets at bookstores in Singapore and at literary events. Won't lose you money I wanted to prove that this venture would not cost me money. Aditi Bharade From the start, I set a constraint on myself — the business may not make me a lot of money, but it shouldn't cost me any. And so far, we have been successful in that. In 2024, I priced the 300 pamphlets at SG$16 each, and in 2025, I raised the price to SG$18. The cost of producing, without design and labor costs, is roughly SG$2,000 to SG$3,000. The past two years, we've about broken even. Last year, we earned a bit more, and I've used that money to hire an intern to help me with this year's run. If you think about this project as a way to earn extra money, it doesn't work. You can earn a lot more easily by doing other things. But I see it as a success that we don't lose money. It's not a vanity project where you spend too much, and the readership is too low. Paper Jam's intent has always been to publish works that would not have seen the light of day otherwise. That's been the value of it to me, so I see it as a success. Read the original article on Business Insider

After living abroad for eight years, Kevin Elliott and his wife, Camille, decided to retire in Thailand rather than return to the US. Provided by Kevin Elliott. Kevin Elliott and his wife, Camille, moved from the US to Qatar for his job. Eight years later, when it was time for him to retire, they chose not to return to the US. Instead, they moved to Hua Hin, a beach town in Thailand, where they live with their four dogs. When Kevin Elliott's job in Qatar ended, he and his wife, Camille, had to decide quickly where to retire. The couple had spent nearly eight years in Doha after moving from South Carolina in 2018 for Kevin's job. Life in the Middle East took some getting used to, especially the heat and the language barrier when dealing with bureaucracy. But once they settled in, they built a social life and traveled extensively. "It was fun," Kevin, now 63, told Business Insider. Staying long-term in Doha was never their plan, nor did they want to return to the US, given the political divide and the high cost of living. They had traveled to Malaysia and Thailand over the years and later visited Costa Rica and Panama to explore them as retirement options. "One of the things we always tried to look at was, 'Where could we live off of our Social Security alone?'" Kevin said. In the end, Thailand won, with its lower cost of living and relatively accessible retirement visas. They knew they didn't want to return to the US, given the political divide and the rising cost of living. Provided by Kevin Elliott. After Kevin's contract ended, the couple traveled to Hua Hin, a coastal town about 90 miles southwest of Bangkok, in November to see if it could check all the boxes in their long-term plan. With about a month before their visas in Qatar expired, they were on a tight timeline. "It was a compressed schedule because my goal was, at least, by the time we flew back to Doha, we had to sign a lease," Kevin said. "And we did," he added. Thailand, their next chapter Back in Qatar, they teed up their Thailand retirement visas and pet relocation plans. Bringing their two dogs — a Pomeranian and a Mauzer — to Thailand cost about $750 between paperwork and airline fees. The couple moved to Hua Hin on January 1, settling into their two-story home in a gated community. Rent is 28,000 Thai baht, or about $870, each month. They live in a house in a gated community in Hua Hin. Provided by Kevin Elliott. After paying 8,000 Qatari riyal, or about $2,100, for a one-bedroom apartment in Doha, their new three-bedroom house with a yard felt like a major upgrade. The neighborhood's greenery and space also reminded Camille of Jamaica, where she grew up. "We knew it needed work, but I loved what I saw," Camille, 62, told Business Insider. Their neighborhood has its quirks, too, including monkeys that occasionally show up in their yard. "Right now, it seems like a bonus, but I don't know. At some point, it might not. But we kind of like watching the monkeys because we're not used to them," Kevin said. For a while, the couple relied on a ride-hailing app for food, groceries, and transportation. After they borrowed a friend's car, they plan to buy their own. While prices are generally lower than in the US, they said expenses can still creep up if they're not careful. Provided by Kevin Elliott. Their goal is to keep their monthly expenses under $2,500, though they're still getting a clear sense of their spending. "We're just starting to get out of that phase of buying stuff for the house," Kevin said. They're still waiting to file their Social Security, but estimate they'll get about $4,000 a month. While most things in Thailand are inexpensive by US standards, they say costs can add up quickly if they don't pay attention. "If you're always comparing to Doha or the US, you forget that at some point it needs to be Thai affordable. I'm in Thailand. Is this affordable? Not compared to the US and Doha, because you'll overspend," he said. Even with that, they said the move has given them a quality of life they don't think they could replicate in the US. "It's not possible," Camille said. They can eat better for less, Kevin said. "We're eating a lot of seafood, fresh vegetables, and fruits here. If we tried to do that in the US on a regular basis, it would probably be cost-prohibitive, especially in retirement." Thailand also has a robust healthcare system. More importantly, they said they've never felt unsafe living there, unlike in the US, where gun violence is "a concern," he said. A new country, a new rhythm Though it's difficult to determine exactly how many American retirees live in Thailand, at least 7,178 retirees there received Social Security benefits as of December 2024, according to the most recent US government data. The couple said they're still settling into their new routine, so they haven't ventured much into the social scene yet. Provided by Kevin Elliott. Unlike in Doha, where social life often revolved around surface-level interactions, Kevin says that it's easier to form stronger social connections in Thailand among the established expat scene. "If we're out somewhere to eat, it's nothing for us to say, 'What is your WhatsApp number? Let's stay in touch,'" he said. Even so, they said they haven't fully tapped into that social scene yet, as they've been focused on settling in — and looking after two more puppies they recently adopted. Alongside the two dogs they brought from Qatar, they also adopted two more. Provided by Kevin Elliott. "I'm probably going to plug into some of the golfing groups because golf is big here. I hadn't played in years, but I enjoy the game," Kevin added. Life in their new home has taken some getting used to. The couple said adapting to Thailand's heat and humidity has been a challenge, especially compared to the more air-conditioned lifestyle they had in Doha. "We are still trying to adjust to the language barrier," Kevin said. Even though Google Translate has been a great starting point, it's not a long-term solution. "Having to pull out your phone, open the app, and go back and forth sometimes wears on you," he said. The couple said they sometimes miss the efficiency of getting things done in Qatar and the US, but have learned to adjust to a slower pace. If anything, the cultural differences have been a positive, especially the friendliness and openness of the people they've met. Kevin says he tries to live by two pieces of advice he once received. "One, it's their culture. You don't have to forget your culture, but plug into theirs. And then two, try to make most of your life an adventure," he said. "And if you do those two things, you should be OK." Do you have a story to share about relocating to a new city or country? Contact this reporter at agoh@businessinsider.com. Read the original article on Business Insider

Discussion | Link

Jack Kubinec / Fortune: Ciridae, which is developing AI to streamline back-office operations for real-economy businesses, raised a $20M seed led by Accel, with a16z participating  —  AI is swiftly rewiring the way people work, but an outsize amount of this workplace revolution has so far been confined to Fortune 500-style companies.

Once one of the most prominent AI skeptics, Citadel CEO Ken Griffin has had a change of heart. Patrick T. Fallon / AFP via Getty Images Citadel CEO Ken Griffin has long been one of the most prominent skeptics of AI. In January, he called the technology "garbage." Now, he says AI is "profoundly more powerful" and will reshape society. Citadel CEO Ken Griffin has finally joined the AI hype train. While many CEOs have been saying for years that AI can do the work of many white-collar professionals, and remaking their companies accordingly, Griffin had been a notable holdout. At the beginning of the year, during a panel discussion at the World Economic Forum in Davos, the hedge fund billionaire said AI was impressive on the surface, but as soon as you dug deeper, "it's all garbage." Griffin has long been seen as one of AI's most prominent skeptics. Earlier this month, however, during a conversation with professors at Stanford Business School, Griffin took a starkly different view. "I got to tell you, I went home one Friday, actually fairly depressed," Griffin said. "You could just see how this was going to have such a dramatic impact on society." Griffin said AI had become "profoundly more powerful" than it was nine months ago, and had allowed the hedge fund to "unleash" a wider range of use cases for the technology. "For the first time, AI is real," he said. While AI has impacted all kinds of jobs, it's been most visible in the tech industry, where bots like OpenAI's Codex and Anthropic's Claude Code are doing the work of software engineers in vastly compressed timelines. Tech companies like Cloudflare have laid off thousands of employees, citing AI's ability to do their work instead. Now, Griffin says the technology is at a place where it can do work once reserved for highly trained finance professionals. "To be blunt, work that we would usually do with people with master's and PhDs in finance over the course of weeks or months is being done by AI agents over the course of hours or days," Griffin said. He emphasized that this goes beyond what he called "mid-tier white-collar jobs" that are now being automated with agentic AI. While AI-assisted software engineering has delivered productivity gains — "you get a 15 or 20% boost or 25% boost," he said — he described the impact on research as more jarring. "When you're seeing really high-level research being done by AI engines, it's, it's quite eye-opening," Griffin said. Work that used to take humans years is now being done in days or weeks, he added. The broader implication, he said, is that both companies and workers will need to become more adaptable. "The success in your career will be defined as to whether or not you will be a lifelong learner or not, and AI will just make this all the more important," Griffin said. Read the original article on Business Insider

Lauren Sherman / Puck: Source: Shein is acquiring DTC clothing retailer Everlane, which focuses on sustainability and “radical transparency”, in a $100M deal  —  The Millennial D.T.C. company built a global brand based on sustainability and radical transparency.  Now, after a years-long slump …

The Information: Analysis: 34 leading AI startups are generating ~$80B in annualized revenue, up 112% from six months ago, with Anthropic and OpenAI capturing 89% of the revenue  —  Anthropic and OpenAI are widening the revenue gap between themselves and the rest of the AI startup field.

Privacy will be a major theme when Apple unveils a new version of Siri.