arXiv:2604.13153v1 Announce Type: new Abstract: 3D Gaussian Splatting (3DGS) has recently enabled highly photorealistic 3D reconstruction from casually captured multi-view images. However, this accessibility raises a privacy concern: publicly available images or videos can be exploited to reconstruct detailed 3D models of scenes or objects without the owner's consent. We present PatchPoison, a lightweight dataset-poisoning method that prevents unauthorized 3D reconstruction. Unlike global perturbations, PatchPoison injects a small high-frequency adversarial patch, a structured checkerboard, into the periphery of each image in a multi-view dataset. The patch is designed to corrupt the feature-matching stage of Structure-from-Motion (SfM) pipelines such as COLMAP by introducing spurious correspondences that systematically misalign estimated camera poses. Consequently, downstream 3DGS optimization diverges from the correct scene geometry. On the NeRF-Synthetic benchmark, inserting a 12 X 12 pixel patch increases reconstruction error by 6.8x in LPIPS, while the poisoned images remain unobtrusive to human viewers. PatchPoison requires no pipeline modifications, offering a practical, "drop-in" preprocessing step for content creators to protect their multi-view data.

arXiv:2604.13127v1 Announce Type: new Abstract: The need to selectively and efficiently erase learned information from deep neural networks is becoming increasingly important for privacy, regulatory compliance, and adaptive system design. We introduce Graph-Propagated Projection Unlearning (GPPU), a unified and scalable algorithm for class-level unlearning that operates across both vision and audio models. GPPU employs graph-based propagation to identify class-specific directions in the feature space and projects representations onto the orthogonal subspace, followed by targeted fine-tuning, to ensure that target class information is effectively and irreversibly removed. Through comprehensive evaluations on six vision datasets and two large-scale audio benchmarks spanning a variety of architectures including CNNs, Vision Transformers, and Audio Transformers, we demonstrate that GPPU achieves highly efficient unlearning, realizing 10-20x speedups over prior methodologies while preserving model utility on retained classes. Our framework provides a principled and modality-agnostic approach to machine unlearning, evaluated at a scale that has received limited attention in prior work, contributing toward more efficient and responsible deep learning.

arXiv:2604.13112v1 Announce Type: new Abstract: Reliable image quality assessment is essential in applications where large volumes of images are acquired automatically and must be filtered before further analysis. In many practical scenarios, a pristine reference image is unavailable, making no reference image quality assessment (NR-IQA) particularly important. This paper introduces Multi-Metric Image Quality Assessment (MM-IQA), a lightweight multi-metric framework for NR-IQA. It combines interpretable cues related to blur, edge structure, low resolution artifacts, exposure imbalance, noise, haze, and frequency content to produce a single quality score in the range [0,100].MM-IQA was evaluated on five benchmark datasets (KonIQ-10k, LIVE Challenge, KADID-10k, TID2013, and BIQ2021) and achieved SRCC values ranging from 0.647 to 0.830. Additional experiments on a synthetic agricultural dataset showed consistent behavior of the designed cues. The Python/OpenCV implementation required about 1.97 s per image. This method also has modest memory requirements because it stores only a limited number of intermediate grayscale, filtered, and frequency-domain representations, resulting in memory usage that scales linearly with image size. The results show that MM-IQA can be used for fast image quality screening with explicit distortion aware cues and modest computational cost.

arXiv:2604.13060v1 Announce Type: new Abstract: Dental triage is a safety-critical clinical routing task that requires integrating multimodal clinical information (e.g., patient complaints and radiographic evidence) to determine complete referral plans. We present Dental-TriageBench, the first expert-annotated benchmark for reasoning-driven multimodal dental triage. Built from authentic outpatient workflows, it contains 246 de-identified cases annotated with expert-authored golden reasoning trajectories, together with hierarchical triage labels. We benchmark 19 proprietary, open-source, and medical-domain MLLMs against three junior dentists serving as the human baseline, and find a substantial human--model gap, on fine-grained treatment-level triage. Further analyses show that accurate triage requires both complaint and OPG information, and that model errors concentrate on cases with multiple referral domains, where MLLMs tend to produce overly narrow referral sets and omission-heavy errors. Dental-TriageBench provides a realistic testbed for developing multimodal clinical AI systems that are more clinically grounded, coverage-aware, and safer for downstream care.

arXiv:2604.13059v1 Announce Type: new Abstract: Most dialogue-based electronic medical record (EMR) systems still behave as passive pipelines: transcribe speech, extract information, and generate the final note after the consultation. That design improves documentation efficiency, but it is insufficient for proactive consultation support because it does not explicitly address streaming speech noise, missing punctuation, unstable diagnostic belief, objectification quality, or measurable next-action gains. We present an end-to-end proactive EMR assistant built around streaming speech recognition, punctuation restoration, stateful extraction, belief stabilization, objectified retrieval, action planning, and replayable report generation. The system is evaluated in a preliminary controlled setting using ten streamed doctor-patient dialogues and a 300-query retrieval benchmark aggregated across dialogues. The full system reaches state-event F1 of 0.84, retrieval Recall@5 of 0.87, and end-to-end pilot scores of 83.3% coverage, 81.4% structural completeness, and 80.0% risk recall. Ablations further suggest that punctuation restoration and belief stabilization may improve downstream extraction, retrieval, and action selection within this pilot. These results were obtained under a controlled simulated pilot setting rather than broad deployment claims, and they should not be read as evidence of clinical deployment readiness, clinical safety, or real-world clinical utility. Instead, they suggest that the proposed online architecture may be technically coherent and directionally supportive under tightly controlled pilot conditions. The present study should be read as a pilot concept demonstration under tightly controlled pilot conditions rather than as evidence of clinical deployment readiness or clinical generalizability.

arXiv:2604.13058v1 Announce Type: new Abstract: We introduce KMMMU, a native Korean benchmark for evaluating multimodal understanding in Korean cultural and institutional settings. KMMMU contains 3,466 questions from exams natively written in Korean, covering nine disciplines and nine visual modality categories, along with a 300-item Korean-specific subset and a hard subset of 627 questions. Unlike translated or English-centric benchmarks, KMMMU targets information-dense problems shaped by local conventions, official standards, and discipline-specific visual formats. Experiments show that the strongest open-source model reaches only 42.05% accuracy on the full set, while the best proprietary model achieves 52.42% on the hard subset. Performance varies across disciplines, with some disciplines emerging as bottlenecks, and Korean-specific questions showing gaps of up to 13.43%. Error analysis suggests that these failures stem less from insufficient reasoning depth than from weak convention-to-label mapping, few-shot symbolic induction, localized knowledge recall, and domain-specific standards understanding. KMMMU provides a testbed for multimodal evaluation beyond English-centric benchmarks and for developing more reliable systems for expert real-world tasks.

arXiv:2604.13056v1 Announce Type: new Abstract: This paper presents a practical pipeline for turning text corpora into quantitative semantic signals. Each news item is represented as a full-document embedding, scored through logprob-based evaluation over a configurable positional dictionary, and projected onto a noise-reduced low-dimensional manifold for structural interpretation. In the present case study, the dictionary is instantiated as six semantic dimensions and applied to a corpus of 11,922 Portuguese news articles about Artificial Intelligence. The resulting identity space supports both document-level semantic positioning and corpus-level characterization through aggregated profiles. We show how Qwen embeddings, UMAP, semantic indicators derived directly from the model output space, and a three-stage anomaly-detection procedure combine into an operational text-as-signal workflow for AI engineering tasks such as corpus inspection, monitoring, and downstream analytical support. Because the identity layer is configurable, the same framework can be adapted to the requirements of different analytical streams rather than fixed to a universal schema.

arXiv:2604.13081v1 Announce Type: new Abstract: The Forward-Forward (FF) algorithm is a biologically plausible alternative to backpropagation that trains neural networks layer by layer using a local goodness function to distinguish positive from negative data. Since its introduction, sum-of-squares (SoS) has served as the default goodness function. In this work, we systematically study the design space of goodness functions, investigating both which activations to measure and how to aggregate them. We introduce top-k goodness, which evaluates only the k most active neurons, and show that it substantially outperforms SoS, improving Fashion-MNIST accuracy by 22.6 percentage points. We further introduce entmax-weighted energy, which replaces hard top-k selection with a learnable sparse weighting based on the alpha-entmax transformation, yielding additional gains. Orthogonally, we adopt separate label feature forwarding (FFCL), in which class hypotheses are injected at every layer through a dedicated projection rather than concatenated only at the input. Combining these ideas, we achieve 87.1 percent accuracy on Fashion-MNIST with a 4x2000 architecture, representing a 30.7 percentage point improvement over the SoS baseline while changing only the goodness function and the label pathway. Across controlled experiments covering 11 goodness functions, two architectures, and a sparsity spectrum analysis over both k and alpha, we identify a consistent principle: sparsity in the goodness function is the most important design choice in FF networks. In particular, adaptive sparsity with alpha approximately 1.5 outperforms both fully dense and fully sparse alternatives.

arXiv:2604.13082v1 Announce Type: new Abstract: Grokking in transformers trained on algorithmic tasks is characterized by a long delay between training-set fit and abrupt generalization, but the source of that delay remains poorly understood. In encoder-decoder arithmetic models, we argue that this delay reflects limited access to already learned structure rather than failure to acquire that structure in the first place. We study one-step Collatz prediction and find that the encoder organizes parity and residue structure within the first few thousand training steps, while output accuracy remains near chance for tens of thousands more. Causal interventions support the decoder bottleneck hypothesis. Transplanting a trained encoder into a fresh model accelerates grokking by 2.75 times, while transplanting a trained decoder actively hurts. Freezing a converged encoder and retraining only the decoder eliminates the plateau entirely and yields 97.6% accuracy, compared to 86.1% for joint training. What makes the decoder's job harder or easier depends on numeral representation. Across 15 bases, those whose factorization aligns with the Collatz map's arithmetic (e.g., base 24) reach 99.8% accuracy, while binary fails completely because its representations collapse and never recover. The choice of base acts as an inductive bias that controls how much local digit structure the decoder can exploit, producing large differences in learnability from the same underlying task.

arXiv:2604.13054v1 Announce Type: new Abstract: Multimodal large language models (MLLMs) have achieved rapid progress, yet their scaling behavior remains less clearly characterized and often less predictable than that of text-only LLMs. Increasing model size and task diversity often yields diminishing returns. In this work, we argue that the primary bottleneck in multimodal scaling is not task format, but knowledge density in training data. We first show that task-specific supervision such as Visual Question Answering (VQA) contributes little incremental semantic information beyond image captions: VQA signals can be reconstructed from captions with negligible performance loss. We then demonstrate that increasing knowledge density -- through structured caption enrichment and cross-modal knowledge injection -- leads to consistent performance improvements across multimodal and downstream benchmarks. Across controlled experiments, performance correlates more strongly with semantic coverage than with task diversity. These findings suggest that current MLLMs fail to scale primarily because training data lacks sufficient knowledge coverage. We advocate for knowledge-centric multimodal training as a principled foundation for scalable multimodal models.

arXiv:2604.13131v1 Announce Type: new Abstract: Satellite sea surface temperature (SST) products underpin global coral bleaching monitoring, yet they measure only the ocean skin. Corals inhabit depths from the shallows to beyond 20 metres, where temperatures can be 1-3{\deg}C cooler than the surface; applying satellite SST uniformly to all depths therefore overestimates subsurface thermal stress. We present a physics-informed neural network (PINN) that fuses NOAA Coral Reef Watch SST with sparse in-situ temperature loggers within the one-dimensional vertical heat equation, enforcing SST as a hard surface boundary condition and jointly learning effective thermal diffusivity (\k{appa}) and light attenuation (Kd). Validated across four Great Barrier Reef sites (30 holdout experiments), the PINN achieves 0.25-1.38{\deg}C RMSE at unseen depths. Under extreme sparsity (three training depths), the PINN maintains 0.27{\deg}C RMSE at the 5 metre holdout and 0.32{\deg}C at the 9.1 metre holdout, where statistical baselines collapse to >1.8{\deg}C; it outperforms a physics-only finite-difference baseline in 90% of experiments. Depth-resolved Degree Heating Day (DHD) profiles show that thermal stress attenuates with depth: at Davies Reef, DHD drops from 0.29 at the surface to zero by 10.7 metres, consistent with logger observations, while satellite DHD remains constant at 0.31 across all depths. However, the PINN underestimates absolute DHD at shallow depths because its smooth predictions attenuate the short-duration peaks that drive threshold exceedances; PINN DHD values should be interpreted as conservative lower bounds on depth-resolved stress. These results demonstrate that physics-constrained fusion of satellite SST with sparse loggers can extend bleaching assessment to the depth dimension using existing observational infrastructure.

arXiv:2604.13130v1 Announce Type: new Abstract: We study learning to learn for regression problems through the lens of hyperparameter tuning. We propose the Langevin Gradient Descent Algorithm (LGD), which approximates the mean of the posterior distribution defined by the loss function and regularizer of a convex regression task. We prove the existence of an optimal hyperparameter configuration for which the LGD algorithm achieves the Bayes' optimal solution for squared loss. Subsequently, we study generalization guarantees on meta-learning optimal hyperparameters for the LGD algorithm from a given set of tasks in the data-driven setting. For a number of parameters $d$ and hyperparameter dimension $h$, we show a pseudo-dimension bound of $O(dh)$, upto logarithmic terms under mild assumptions on LGD. This matches the dimensional dependence of the bounds obtained in prior work for the elastic net, which only allows for $h=2$ hyperparameters, and extends their bounds to regression on convex loss. Finally, we show empirical evidence of the success of LGD and the meta-learning procedure for few-shot learning on linear regression using a few synthetically created datasets.

arXiv:2604.13085v1 Announce Type: new Abstract: Autonomous AI agents operating in dynamic environments face a persistent challenge: acquiring new capabilities without erasing prior knowledge. We present Adaptive Memory Crystallization (AMC), a memory architecture for progressive experience consolidation in continual reinforcement learning. AMC is conceptually inspired by the qualitative structure of synaptic tagging and capture (STC) theory, the idea that memories transition through discrete stability phases, but makes no claim to model the underlying molecular or synaptic mechanisms. AMC models memory as a continuous crystallization process in which experiences migrate from plastic to stable states according to a multi-objective utility signal. The framework introduces a three-phase memory hierarchy (Liquid--Glass--Crystal) governed by an It\^o stochastic differential equation (SDE) whose population-level behavior is captured by an explicit Fokker--Planck equation admitting a closed-form Beta stationary distribution. We provide proofs of: (i) well-posedness and global convergence of the crystallization SDE to a unique Beta stationary distribution; (ii) exponential convergence of individual crystallization states to their fixed points, with explicit rates and variance bounds; and (iii) end-to-end Q-learning error bounds and matching memory-capacity lower bounds that link SDE parameters directly to agent performance. Empirical evaluation on Meta-World MT50, Atari 20-game sequential learning, and MuJoCo continual locomotion consistently shows improvements in forward transfer (+34--43\% over the strongest baseline), reductions in catastrophic forgetting (67--80\%), and a 62\% decrease in memory footprint.

arXiv:2604.13123v1 Announce Type: new Abstract: Grokking -- delayed generalisation long after memorisation -- lacks a predictive mechanistic explanation. We identify the normalised spectral entropy $\tilde{H}(t)$ of the representation covariance as a scalar order parameter for this transition, validated on 1-layer Transformers on group-theoretic tasks. Five contributions: (i) Grokking follows a two-phase pattern: norm expansion then entropy collapse. (ii) $\tilde{H}$ crosses a stable threshold $\tilde{H}^* \approx 0.61$ before generalisation in 100% of runs (mean lead: 1,020 steps). (iii) A causal intervention preventing collapse delays grokking by +5,020 steps ($p=0.044$); a norm-matched control ($n=30$, $p=5\times10^{-5}$) confirms entropy -- not norm -- drives the transition. (iv) A power-law $\Delta T = C_1(\tilde{H}-\tilde{H}^*)^\gamma+C_2$ ($R^2=0.543$) predicts grokking onset with 4.1% error. (v) The mechanism holds across abelian ($\mathbb{Z}/97\mathbb{Z}$) and non-abelian ($S_5$) groups. Crucially, MLPs show entropy collapse without grokking, proving collapse is necessary but not sufficient -- architecture matters. Code: https://anonymous.4open.science/r/grokking-entropy

Business Insider's The Long Play event addressed how professionals and companies are thriving in the age of AI. Tammy Horton for Nikki Ritcher Business Insider brought together some of the buzziest names in AI, media, health, and longevity. Speakers, including Axiom Math's Carina Hong and Blumhouse Productions' Jason Blum, talked about AI. Here's a glimpse of what it was like at Business Insider's The Long Play. AI is shaking up every industry, and some of the top players in superintelligence, health, media, and longevity know it. At San Francisco's Exploratorium on Tuesday evening, Business Insider hosted The Long Play, gathering 150-some industry insiders and executives, including Asana CEO Dan Rogers and the NFL tight end-turned-executive coach Damian Vaughn, to get the inside scoop on how career climbers and companies can thrive in the age of AI. Business Insider hosted The Long Play inside San Francisco's Exploratorium. Lloyd Lee/BI Will AI replace Hollywood creatives and healthcare clinicians? Who will AI make obsolete first: coders or mathematicians? And what are the broad societal implications of a world where everybody lives longer? Colleen Kuhn (left) and Courtney Laub (right) from Workday with Business Insider's Rosalie Chan. Tammy Horton for Nikki Ritcher Business Insider spoke to Axiom's Carina Hong, Jason Blum of Blumhouse Productions, Midi Health founder Joanna Strober, and entrepreneur Bryan Johnson, the "most-measured man on planet Earth," to get their insights. A running theme: The tech world — and modern culture overall — rewards speed, scale, and certainty. The speakers showed there's still room for limits, skepticism, and human judgment in the long run. Business Insider CEO Barbara Peng speaks at The Long Play. Tammy Horton for Nikki Ritcher The night was paired with health-conscious poke bowls, booze (a Johnson no-no), and a quick talk on sex and night-time erections (a Johnson yes-yes). Here are some of the highlights from The Long Play. Asana CEO Dan Rogers shared his best career advice and how he uses AI Asana CEO Dan Rogers (right) chatting with Business Insider's Anthony DeMaio and Andrew Hammer. Nina Menconi for Nikki Ritcher Asana CEO Dan Rogers said he relies on AI as a "day-to-day companion," using the company's AI teammates to summarize information, consolidate insights, and conduct competitive research. The tools, he said, are embedded directly into his workflow and used daily. On career advice, Rogers pointed to a familiar Silicon Valley mantra that he believes still holds up: join a "rocket ship." In a fast-moving tech landscape, he said, workers should focus less on roles and more on companies with breakout potential. Outside work, Rogers emphasized consistency, calling exercise a non-negotiable. He prioritizes a mix of cardio, strength training, and yoga. Sofia Gomes Coelho (left) and Taina Martinez (right) from Apple. Tammy Horton for Nikki Ritcher Axiom Math CEO Carina Hong believes that a powerful AI checks its own work LLMs have been revolutionary for the front-end software engineer, but we're not yet seeing AI taking over systems where safety is a non-negotiable without a human in the loop, Hong, a Rhodes scholar and founder of Axiom Math, said. Carina Hong, founder of Axiom Math, is building an AI mathematician. Lloyd Lee/BI The mathematician said the most powerful AI will be the one that can check its own output. Because what good is an LLM that gives us five different answers to the mysteries of our universe, Hong proposed. How could a human check that answer? "I would say that superintelligence is at its sort of max power when it's verified," she said. "When you know that you can just execute the output like a computer program, and then you get sort of the verifiable signal." Dave Bottoms (right), SVP and GM, Marketplace at Upwork. Nina Menconi for Nikki Ritcher Founders weigh in on the most AI-proof skills Some of the founders who spoke with Business Insider talked about how human judgment has become all the more valuable in the AI era. Talha Khan (center), CEO of Seda, said that "taste" is a skill that AI has made more important. Nina Menconi for Nikki Ritcher "The simple answer is taste," Talha Khan, CEO of Seda, said, when asked what skill AI has made more important. Khan said the more context and insight you can provide an AI, the better and more unique the output. Carmen Li, CEO of Silicon Data, said she lets AI know she's never satisfied with the answer. Lloyd Lee/BI Carmen Li, CEO of Silicon Data, said she uses her judgment to discern between a flattering AI and one that's providing an accurate answer. "Don't make me feel good," she said. "I couldn't care less about that." Gary Yasuda (left) said that different AI tools act like "a board of directors" that needs the correct human direction. Nina Menconi for Nikki Ritcher Gary Yasuda, the president of Milan Institute, said that various AI tools are like "a board of directors" that needs the correct human direction. "Everybody has a favorite — different technology, ChatGPT, Claude, Perplexity — but they all ask their own style of questions because if you don't ask the right probing questions, they're never going to really help you," said Yasuda. Off-the-shelf AI won't cut it for healthcare Midi Health CEO Joanna Strober set out to launch a menopause startup, but ended up building an AI company. Tammy Horton/Nikki Ritcher Strober, founder of Midi Health, a virtual health clinic for women, said generic, popular chatbots still face inaccuracies when it comes to questions around patient health. "We spend a lot of time debunking the things they saw in AI because it doesn't include only the most updated women's health research," she said. The founder envisions a future when patients can rely on a specialized AI to answer some of their pressing health questions, but Strober's bet is that it won't replace a human anytime soon. "Honestly, I'm making a bet on the fact that people still want to talk to someone for their care," Strober said. "Other companies are probably going to make a different bet, but we have found that what we are going to be really good at is empathy." Hollywood isn't competing with AI right now Blum, whose production company is behind some of the most popular horror franchises and films, doesn't think AI will make movies better. Jason Blum shares how Blumhouse's original model has had to evolve with Hollywood. Tammy Horton/Nikki Ritcher The producer has experimented with the technology. In 2024, Blumhouse Productions announced a partnership with Meta, producing AI-generated shorts and VR experiences. He was roasted online. That experience, Blum said, made him confident that AI can't make movies better and that Hollywood isn't competing with it. Instead, AI is competing with user-generated content. "What it is competing with is scrolling," he said. "On Instagram and doomscrolling, I think there'll be a ton of AI there and you will see that. So I think for once creators have more to worry about than directors and writers." Grind culture is not a badge of honor Bryan Johnson at Business Insider's The Long Play. Nina Menconi Founders might be skipping booze, but the grind culture persists. On top of that, some young professionals are swearing off sex and dating, all in the name of "locking in." Johnson, the popular longevity guru, wants that to stop. "We have not created a culture of health, and it's hurting everybody," he said. His message, in short: Have sex, get off your phone, and go to sleep. No one has a crystal ball Bryan Johnson and Business Insider's Zak Jason talk about the future. Tammy Horton for Nikki Ritcher CEOs have long been expected to make institutional forecasts, but it seems these days they're being turned to as an oracle of the future — especially amid the AI boom cycle. When will AI wipe out the white-collar workforce? What value will money have when humanoid robots take over the world? Johnson makes his case: Technology has enabled us to advance in our work and our health, but it doesn't make us any better at predicting the future, he said, when asked what it means for the world to have a population that can live longer. "My personal opinion is nobody has anything intelligent to say about the future," he said. "In fact, when they try to say things, they're revealing their ignorance, not their knowledge." Read the original article on Business Insider

Allbirds started as a sustainable footwear brand. Now it's pivoting to AI and plans to shift its focus away from the environment. Bloomberg/Getty Images Allbirds is pivoting from sustainable footwear to AI compute infrastructure. The company plans to be less focused on environmental conservation, according to an SEC filing. Allbirds' stock price surged 582% after the company announced it was transitioning to AI. Allbirds isn't only pivoting to AI — it's pivoting away from its core environmental principles. Once dubbed the favorite shoemaker of Silicon Valley, Allbirds announced Wednesday it was transitioning from footwear to AI compute infrastructure and becoming NewBird AI. The company said it planned to buy GPUs, or powerful chips, and become a GPU-as-a-service company. It's also selling off its footwear assets and its original name, meaning Allbirds-branded shoes could continue to be made, under new ownership. The company partially answered questions about what that might mean for Allbirds' status as a company committed to ESG — environmental, social, and governance principles — in a filing with the Securities and Exchange Commission. The filing said stockholders would vote next month on proposed changes to the company, including removing environmental commitments written into its charter and revoking its status as a public benefit corporation. PBCs are for-profit corporations intended to produce a public benefit and operate responsibly and sustainably. "The anticipated Electronics Infrastructure Business would be less focused on the public benefit of environmental conservation," the company said. Allbirds said in the filing that there were risks associated with some of the proposed changes to its charter, including ceasing to operate as a Delaware PBC, which requires the company to "produce a public benefit and operate in a responsible and sustainable manner." "Eliminating that status could damage our reputation, adversely affect our ability to attract and retain customers, employees and partners, and negatively impact our business, financial condition and results of operations," the filing said. If the amended charter is accepted, the company said, it would operate "in the best interests of our stockholders, without balancing the best interests of our stockholders against the public benefit of environmental conservation." Allbirds did not respond to a request for comment from Business Insider. The transition to AI is an especially striking move for Allbirds, given concerns about the environmental impacts and resource consumption of AI. An investigation by Business Insider found the AI and data center boom is straining power grids and water supplies across the US. Allbirds was founded as a sustainable footwear company When Allbirds started out in 2015, sustainability was central. The company touted its shoes, made from what it said were natural, sustainable materials, as better for people and the environment. In 2016, the company earned its B Corp certification, given to companies that meet specific standards for social and environmental impact. It also became a PBC. When Allbirds went public in 2021, the company said in an SEC filing that its mission was to "make better things in a better way, through nature — products that people feel good in and feel good about." "We aim to reverse climate change through better business by empowering people to make better, more conscious decisions for themselves as well as the planet," the company said, adding, "We are proud of the alignment of financial and environmental benefits from our work, and that we are able to serve as a driving force in a new age of sustainable enterprise." Last year, the company said it had made the "world's first net zero carbon shoe" with its limited edition "M0.0NSHOT Zero," made from "carbon-negative, regenerative wool" sourced from New Zealand. The company's critics have accused it of greenwashing — that its claims about its positive environmental impact were more marketing than reality. While Allbirds said in its latest SEC filing that it will be less focused on environmental conservation, it's unclear whether environmental principles will remain part of the company in some capacity. For now, it looks like the transition might be good for shareholders, if not for business. Allbirds' stock price was up 582% from the day prior at market close on Wednesday, reversing a yearslong slump. Read the original article on Business Insider

Lucas Ropek / TechCrunch: OpenAI updates Agents SDK with native sandboxing and an in-distribution harness for deploying and testing agents on long-horizon tasks  —  Agentic AI is the tech industry's newest success story, and companies like OpenAI and Anthropic are racing to give enterprises the tools they need to create these automated little helpers.

Discussion | Link

OpenAI introduces GPT-Rosalind, a frontier reasoning model built to accelerate drug discovery, genomics analysis, protein reasoning, and scientific research workflows.

Kalshi has been fined $5 million in Ohio. Illustration by Davide Bonaldo/SOPA Images/LightRocket via Getty Images Kalshi is facing a fine in Ohio for unlicensed sports betting activities. The prediction market operator has been in an Ohio court battle to allow it to operate in the state. Kalshi suffered a loss in March, when an Ohio judge ruled its offerings should be regulated. It's more money, more problems for the online prediction market Kalshi. The Ohio Casino Control Commission said on Tuesday that it was fining Kalshi $5 million for unlicensed sports betting in the state. "The Commission takes its regulatory responsibilities to ensure compliance with the law and the integrity of sports gaming in Ohio seriously," the regulators said in a statement. "Kalshi's refusal to stop offering sports gaming in Ohio necessitated the Commission to take action to uphold the requirements of Ohio law. " This fine comes on the heels of a court loss Kalshi suffered in March. A judge in Ohio's federal court ruled that Kalshi's sports-related prediction market bets should be considered gambling and be regulated by the state. In April 2025, the commission issued three cease and desist notices — to Kalshi, Robinhood, and Crypto.com — directing them to stop offering sports event contracts in the state by April 14, 2025. Kalshi sued Ohio and asked the court to grant an injunction against the casino regulators' attempt to stop it from taking market bets. Kalshi had asked the court to treat its sports-related market options as swaps, which would designate picks on its platform as a kind of investment. The presiding judge, District Judge Sarah Morrison, ruled against Kalshi, saying she had an obligation to "avoid absurdity." Swaps, she said, are "understood as a transaction involving financial instruments and measures that traditionally and directly affect commodity prices." "Currency exchange rates, the weather and energy costs all do that: the number of points scored in the Huskies-Bobcats game does not," the judge said. Kalshi said in a statement to Business Insider that it's reviewing the gaming commission's fine. "We are disappointed in this latest development, especially considering our ongoing litigation with Ohio and recent rulings in other courts confirming our right to operate as a federally licensed exchange," a Kalshi spokesperson said. The company raised $1 billion last month at a $22 billion valuation, according to data provider PitchBook. Kalshi's expansion across the US is running into a range of state-level resistance. Regulators from Arizona to Nevada say its offerings may amount to unlicensed gambling. The platform is also set to see more federal oversight from the Commodity Futures Trading Commission. Its enforcement chief, David Miller, said in March that he's hiring more staff to crack down on insider trading on prediction markets. Read the original article on Business Insider