Brief

Researchers have developed ObjSplat, a novel framework for active object reconstruction that utilizes Gaussian surfels for high-fidelity digital asset creation. This system enhances reconstruction accuracy by modeling back-face visibility and occlusion-aware covisibility, enabling it to identify and reconstruct complex geometries. ObjSplat also employs an advanced planning strategy that optimizes for information gain and movement cost to generate efficient reconstruction trajectories, significantly reducing scan time and path length compared to existing methods. AI

Researchers have introduced a new method called the hyper-scaled NLP bound (hNLP bound) to improve the efficiency of solving the maximum-entropy remote sampling problem (MERSP). This problem involves selecting a subset of variables to maximize information about unobservable targets, assuming a joint Gaussian distribution. The hNLP bound offers theoretical advantages, including dominance over previous bounding methods and the ability to handle rank-deficient covariance matrices, which was a limitation of prior approaches. AI

This article explains how to accurately calculate token usage for large language models before sending requests, which is crucial for managing costs. It details three methods using `tiktoken`, `anthropic-tokenizer`, and the Gemini SDK, along with formulas for estimating costs in rubles. The piece highlights that token density varies significantly between languages, with Russian being less dense than English, making Russian prompts more expensive. AI

A user is experiencing significant frustration with Cursor's AI-powered support bot, which they claim is hallucinating and failing to resolve technical issues. The user highlights the irony of an AI coding assistant lacking effective human customer support for paying customers. They are seeking a way to contact a human representative at Cursor to address their blocked workflow. AI

This article explores how Obsidian, a note-taking application, can be enhanced with various plugins to function as a powerful personal knowledge management system. It suggests integrating AI chatbots like Claude to further augment Obsidian's capabilities, transforming it into a more dynamic tool for users. AI

36Kr is hosting its annual WAVES event in Panyu, Guangzhou, focusing on startups and investors under 36. The event will feature discussions on emerging tech sectors like embodied intelligence and AI, aligning with Panyu's strategic focus on these future industries. Panyu offers significant advantages including advanced transportation networks, a high concentration of established companies and startups, and a robust talent pool from its numerous universities. AI

A new macOS menu bar application has been released to help users monitor their Claude Code usage quotas. The tool displays real-time utilization for both 5-hour and weekly limits, with color-coded indicators for different thresholds. It also provides a countdown to quota resets and supports multiple accounts, automatically discovering them or allowing manual configuration. AI

The Model Context Protocol (MCP) is an emerging standard for AI assistants to interact with development tools. A recent article highlights ten MCP servers designed to enhance developer workflows in 2026. These servers provide AI assistants with capabilities ranging from GitHub repository management and local filesystem manipulation to database querying and Docker container orchestration. AI

This article explains how to use Claude AI for data cleaning and exploratory data analysis (EDA). It emphasizes that the quality of analytical results directly depends on the quality of the data used, particularly in handling missing values. AI

Anthropic's Claude AI is enabling dynamic workflows through an orchestration framework that allows multiple agents to work in parallel. This system scales complex tasks by coordinating these agents, moving AI-assisted software development into a new dimension. The framework is designed for deep dives into AI's multi-agent capabilities. AI

Using multiple smaller AI models can be more effective than a single large one for tasks like code review, according to mathematical analysis. The key is that the smaller models should have uncorrelated errors, meaning their mistakes do not overlap. This approach, similar to RAID for disks or ensemble classifiers, can achieve higher accuracy rates than a single, more powerful model, often at a lower cost and with parallel processing benefits. AI

Linux developers are leveraging AI coding assistants like GitHub Copilot to maintain and update drivers for vintage AMD GPUs. This effort focuses on the R600 Gallium3D driver, which supports AMD/ATI HD 2000 through HD 6000 series graphics cards, some of which are nearly two decades old. The use of AI aims to compensate for limited manpower in maintaining these older drivers, ensuring their continued compatibility with modern systems and APIs. Linus Torvalds has approved the use of AI in kernel development, provided proper tagging and rigorous testing by developers. AI

This cluster focuses on Android Auto applications useful for off-roading, highlighting six essential apps, many of which are free. The article emphasizes that Android Auto's utility extends beyond urban environments, providing tools for road trips and outdoor exploration. AI

Hefei Sanzhiyang Big Data Operations Co., Ltd. has undergone a change in leadership, with Wang Fenqing stepping down as legal representative, manager, and director, and Du Gang assuming these roles. The company, established in June 2023 with a registered capital of 5 million RMB, focuses on big data services and is wholly owned by Sanzhiyang (Hefei) Holding Group Co., Ltd. AI

Coinbase has introduced Coinbase for Agents, a new platform designed to allow artificial intelligence agents to manage cryptocurrency trades and payments. This tool enables AI models such as ChatGPT and Claude to execute transactions and handle financial operations on behalf of users, streamlining the process of interacting with the Coinbase ecosystem. AI

The Wittgenstein AI Tournament is a competition focused on developing AI agents for cybersecurity defense. Participants are tasked with building functional AI agents based on open-source models. The tournament includes a semifinal round where the top three teams win $1,000 and advance to the finals, with a grand prize of $15,000 for the ultimate winner. AI

A developer used Anthropic's Claude Fable to create a complex browser game simulating SpaceX rocket landings. The AI independently developed realistic aerodynamic physics, including atmospheric drag and reentry heating, and even wrote its own testing harness with an autopilot to verify the physics. Claude Fable also identified and corrected a flaw in the game's physics simulation related to engine spool-up time, demonstrating advanced debugging capabilities. AI

Groundy is a new open-source Python library designed to simplify the process of building and deploying large language model (LLM) applications. It aims to streamline the development workflow for LLM-based projects, making it easier for developers to integrate and manage LLM functionalities. AI

xAI has launched a new MongoDB plugin for its Grok AI assistant, available in the beta Grok Build Plugin Marketplace. This plugin allows users to interact with MongoDB data, optimize database performance, and build vector search systems directly through prompts. The marketplace also features beta plugins for Vercel, Sentry, Cloudflare, and Chrome DevTools, enabling terminal-based development workflows. AI

Researchers have developed an AI model capable of identifying Bryde's whale vocalizations within seismic survey data. This breakthrough allows for the detection of whale calls that would otherwise be masked by the noise of seismic exploration in the South China Sea. The AI's success offers a new method for monitoring whale populations in areas with significant industrial activity. AI

AI researchers are now prioritizing the study of imperfect feedback in predictive models. This focus is particularly relevant for areas such as imitation learning and bandit problems, where models must learn from incomplete or noisy data. The ongoing work aims to advance the capabilities of AI systems in understanding and utilizing suboptimal information. AI

Alibaba has announced a management change at DingTalk, with Chen Hang stepping down as CEO. Chen Yusen, a tech leader born in 1992, will take over as the new CEO, making him Alibaba's youngest division CEO. This transition marks a new chapter for the enterprise communication and collaboration platform. AI

Zhiyuan has launched the Lingxi X2 EDU, a version of their robot designed for educational and training purposes. This new iteration emphasizes a modular hardware design that allows for easy disassembly, expansion, and secondary development. It is specifically targeted at scenarios such as scientific research education, engineering training, and robotics competition development. AI

Users of Anthropic's Claude Desktop application on Windows have reported a significant bug where the app automatically launches a Hyper-V virtual machine upon startup. This VM, identified as Vmmem, consumes approximately 1.8 GB of RAM, even when the user is only utilizing the chat functionality and not engaging with agent or Cowork features. The issue appears to stem from uncleaned session files and a service that triggers the VM on launch, leading to increased system resource usage and potential sluggishness. AI

This article details the construction of a real-time AI system designed to audit customer calls. The system integrates speech-to-text technology with LLM evaluation to monitor calls for compliance and quality. It aims to provide immediate feedback and analysis on every customer interaction. AI

An AI developer detailed three unexpected failure patterns encountered while running Anthropic's Claude Code unattended for a week. The most critical issues involved API errors due to modified 'thinking' blocks, significant data loss from infrequent progress checkpoints, and stalled queues caused by stale lock files. The developer emphasizes that subtle, progress-masking failures are more damaging than loud crashes in unattended AI operations. AI

Armin Ronacher, the creator of the Python web framework Flask, has developed a new AI agent that integrates four distinct tools. This agent utilizes a 1,000-token system prompt, which is notably shorter than the system prompts offered by some competitors like Claude Code and OpenCode. Ronacher's approach prioritizes a more focused and efficient interaction with the AI. AI

Researchers have developed a new memory pattern for AI agents called "Standing Questions" to address the issue of stale information in long-term projects. Instead of storing answers, which can become outdated, the system stores a set of critical questions that the agent must re-derive answers to at the start of each session. This ensures that the agent's understanding is constantly updated against the current state of the project, preventing the accumulation of incorrect beliefs. AI

Researchers have developed a novel Physics-Informed Neural Network (PINN) framework that integrates auxiliary knowledge from sources beyond experimental data. This new approach enhances parameter discovery by incorporating information from peer-reviewed literature and network structures, specifically applied to modeling microbial interactions. The framework demonstrated significant improvements in accuracy and predictive power for microbial community modeling, outperforming existing methods and revealing ecological insights. AI

Researchers have developed a novel structured pruning framework for deep neural networks that utilizes multi-armed bandit (MAB) algorithms to remove entire neurons. This method treats each neuron as an 'arm' in a bandit problem, temporarily masking it to measure the impact on the loss function before updating its removal reward estimate. Evaluations across image, text, and reasoning tasks demonstrated that MAB-based pruning, particularly with UCB1 and Thompson Sampling policies, effectively reduces model size while often outperforming unpruned models and other pruning techniques. AI

Researchers have developed EssentialGIN, a novel approach for predicting essential genes using graph isomorphism neural networks. This method integrates biological data like gene expression and orthology information with network topology to enhance prediction accuracy. Experiments show EssentialGIN outperforms existing centrality-based and machine learning methods, particularly in complex organisms like humans. AI

Researchers have developed a new framework using a multi-planar 2D-U-Net architecture to segment five abdominal organs in 3D CT scans. This method enhances segmentation accuracy by incorporating fuzzy 3D spatial maps that provide anatomical location cues. Evaluations on 80 CT scans demonstrated a Dice improvement of approximately 4% compared to models trained without these spatial occurrence maps. AI

Researchers have developed a new algorithm called TAMUNA designed to improve the efficiency of distributed optimization and federated learning. TAMUNA addresses the communication bottleneck by combining local training and data compression techniques, while also uniquely supporting partial client participation. This approach allows for doubly-accelerated convergence rates, outperforming previous methods that required all clients to be active. AI

Researchers have developed a new semi-supervised learning (SSL) framework for indoor localization using Wi-Fi RSSI fingerprinting. This framework, based on the Mean Teacher model, efficiently utilizes both labeled and unlabeled data for improved accuracy and generalization. It addresses challenges like time-consuming data collection and performance degradation in dynamic environments. The proposed method demonstrated significant reductions in localization errors compared to traditional supervised learning approaches. AI

Researchers have developed a novel approach for load forecasting in data-scarce environments by leveraging a large language model called Chronos. This LLM framework utilizes its extensive pre-trained knowledge to achieve accurate predictions without requiring extensive fine-tuning on specific datasets. Experiments across five real-world datasets demonstrated that Chronos significantly outperforms nine traditional baseline models in both deterministic and probabilistic forecasting, showing substantial reductions in error metrics. AI

Researchers have developed BLINQ, a novel model-based algorithm designed to learn Whittle indices for Markov Decision Processes. This new approach constructs an empirical estimate of the MDP and then computes the indices, offering a proven convergence guarantee and a bound on learning time. Numerical experiments indicate BLINQ requires fewer samples than existing Q-learning methods for accurate approximations and has a lower overall computational cost. AI

Researchers have evaluated seven uncertainty quantification (UQ) methods on the AION-1 astronomical foundation model for predicting galaxy properties. Conformal prediction methods, particularly the Locally Valid and Discriminative (LVD) framework, demonstrated superior calibration and local validity compared to non-conformal baselines. The study suggests LVD is the preferred UQ approach for foundation model embeddings in astrophysics, offering more reliable uncertainty estimates for scientific inference. AI

Researchers have uncovered a new relationship between the noise introduced by Stochastic Gradient Descent (SGD) and the curvature of the loss landscape in deep learning models. Their findings indicate that this noise is not directly proportional to the Hessian of the loss, as previously assumed under specific conditions. Instead, the study reveals a more general connection where the SGD noise covariance is related to the expected value of per-sample Hessians, suggesting these two factors approximately commute rather than coincide. AI

Researchers explored whether large language models (LLMs) could simulate the collective experience of watching a K-pop concert by generating real-time fan chat. In a pilot study with 11 K-pop fans, LLM agents with assigned fan personas produced more natural-sounding chat compared to a baseline without personas. However, this persona conditioning did not significantly improve social connectedness, engagement, or emotional response among participants. Interviews suggested that online concert chat functions more as a collective monologue, and genuine participation relies on shared fandom identity rather than simulated crowd behavior. AI

Researchers have developed a new framework to improve the user experience of online shopping by personalizing review ranking and summarization. This system integrates user preference modeling, sentiment analysis, and Large Language Models (LLMs) to tailor review content to individual needs. By analyzing historical reviews and user-selected product aspects, the framework ranks and summarizes reviews to reduce information overload and enhance decision-making confidence. Evaluations showed this personalized approach significantly outperformed traditional ranking methods and improved user satisfaction and efficiency. AI

Researchers have developed MPC-Flow, a novel framework for solving inverse problems using flow-based generative models. This method employs model predictive control to guide the model's dynamics, making conditional generation more practical. MPC-Flow offers a spectrum of guidance algorithms, some of which bypass the need for backpropagation through the generative model's trajectory. The framework has demonstrated strong performance and scalability on image restoration tasks, including in-painting, deblurring, and super-resolution, even with large-scale models like FLUX.2 on consumer hardware. AI

Researchers have developed AMS-HD, a novel framework utilizing hyperdimensional computing (HDC) for real-time detection of acute mountain sickness (AMS) from wearable physiological signals. This approach significantly reduces energy consumption and computational resources compared to traditional machine learning methods. AMS-HD achieves high accuracy, comparable to or exceeding SVM and MLP baselines, while requiring minimal battery, memory, and processing time, making it suitable for resource-constrained health monitoring devices. AI

Researchers have developed a novel framework for infrared small target detection (IRSTD) called STGBD-Net, which utilizes Basis Decomposition Theory to improve feature fusion. This approach reformulates the process into an adaptive decomposition-and-reconstruction paradigm, employing Gradient Decomposition Modules (GDMs) to treat normalized gradient features as basis vectors. The resulting networks, including spatial and spatio-temporal variants, demonstrate state-of-the-art performance on multiple benchmarks with enhanced accuracy and computational efficiency. AI

Researchers have developed GVC-Seg, a new method for 3D instance segmentation in point cloud data that does not require training. This approach leverages geometric visual correspondence to overcome biases caused by varying confidence levels in multiple pre-trained foundation models. By integrating 3D geometric cues with 2D visual cues, GVC-Seg improves proposal quality assessment and enables unbiased ensemble learning, achieving state-of-the-art results on benchmarks and showing promise for open-vocabulary semantic segmentation. AI

Researchers have developed Muses, a novel method for generating 3D fantasy creatures without requiring any training data. This approach utilizes a 3D skeleton to guide the composition and generation of diverse elements, ensuring a coherent structure and appearance. Muses integrates design, composition, and generation into a unified pipeline, starting with a graph-constrained reasoning process to create a well-structured skeleton, followed by a voxel-based assembly within a latent space, and concluding with appearance modeling for style-consistent texturing. The method demonstrates state-of-the-art performance in visual fidelity and alignment with textual descriptions. AI

Researchers have identified a consistent geometric structure, termed the "cross-architecture substrate," within modern vision encoders, regardless of their specific training objective or domain. This substrate, a 16-dimensional object, remains stable across diverse visual domains and survives calibration tests. The findings suggest a fundamental invariant in how these networks process visual information, leading to practical applications in areas like model transferability and domain detection. AI

Researchers have developed a new deep learning framework for 3D intraoral reconstruction, aiming to improve vertex distribution in predicted point clouds. While the previous model achieved 77.49% accuracy, it suffered from vertex clustering. The updated model introduces Hungarian matching and Repulsion Loss to create a more uniform vertex distribution, though this resulted in a lower accuracy of 68.02%. Despite the numerical decrease, the new approach significantly alleviates the vertex clustering issue, leading to more evenly spread vertices across the reconstructed surface. AI

Researchers have developed CRAG, a novel approach to 3D assembly that integrates generative modeling with pose estimation. Unlike previous methods that solely focus on rigid transformations, CRAG treats assembly and shape generation as mutually reinforcing processes. This allows CRAG to synthesize plausible complete shapes and predict part poses, even when some pieces are missing, achieving state-of-the-art performance on in-the-wild objects. AI

Researchers have developed an agent-guided multi-fidelity framework to improve the accuracy of simulating electronic and optical properties in nanomaterials. This new approach addresses computational challenges like numerical instabilities and convergence failures inherent in demanding calculations. By assigning confidence weights and using high-accuracy reference points, the framework corrects artifacts and enhances agreement with experimental data, proving transferable to various optoelectronic nanomaterials. AI

Researchers have developed strategies to improve the detection of tiny objects in aerial images, a task that challenges standard object detection models like YOLOv8. Their approach involves enhancing input resolution, employing data augmentation, and integrating attention mechanisms within a novel pipeline called MoonNet. This pipeline, which incorporates modules like SE Block and CBAM, demonstrated superior accuracy over existing methods on a specific tiny-object benchmark. AI