Brief

Researchers have developed a novel framework integrating electroencephalogram (EEG) with augmented reality (AR) Vestibular/Ocular Motor Screening (VOMS) tasks to estimate ocular response times. The system utilizes a Redundant Discrete Wavelet Transform (RDWT)-driven deep neural network for analyzing EEG signals, which acts as an effective denoising strategy. Dynamic Time Warping (DTW) was then employed to calculate response times, revealing significant inter-subject differences and task-dependent temporal behaviors, suggesting potential for multimodal mild traumatic brain injury (mTBI) assessment. AI

Researchers have developed nASR, a novel trainable neural layer designed to improve Electroencephalogram (EEG) signal processing for Brain-Computer Interfaces (BCIs). This new layer addresses limitations in existing Artifact Subspace Reconstruction (ASR) methods by introducing trainable parameters that allow for more precise artifact detection and selective channel-level reconstruction. An ablation study demonstrated that nASR variants outperform traditional ASR in classification metrics and significantly reduce inference time, making it suitable for real-time BCI applications. AI

Researchers have introduced a new framework called Chain-of-Procedure (CoP) to enhance visual-language models' ability to answer questions about procedural tasks. This framework addresses limitations in current models by improving the retrieval of structured instructions based on visual cues and aligning the granularity of image sequences with textual step decomposition. CoP first retrieves relevant instructions, then refines steps through semantic decomposition, and finally generates the next action, showing up to a 13% improvement on a new benchmark called ProcedureVQA. AI

Researchers have developed a deterministic agentic workflow for Harmonized System (HS) tariff classification, a complex task requiring multi-dimensional rule reasoning. Unlike self-planning agents, this fixed control flow confines language model calls to specific stages, ensuring interpretability through structured outputs and verbatim citations of relevant notes. When evaluated on the HSCodeComp dataset using a Qwen3.6-plus model, the workflow achieved significant accuracy at both four- and six-digit levels. An open-weight Qwen3.6-27B-FP8 backbone also demonstrated strong performance, and a manual audit suggested potential discrepancies in existing ground-truth labels. AI

Researchers have developed RCLAgent, a new framework designed to improve root cause localization in complex microservice systems. This approach utilizes a multi-agent recursion-of-thought strategy with parallel reasoning to overcome limitations of existing LLM-based methods, such as context explosion and serial reasoning. By assigning dedicated agents to different parts of the system and organizing them recursively, RCLAgent aims to enhance both accuracy and efficiency in diagnosing failures. AI

Researchers have developed REALM, a new framework for real-time decoding of local field potentials (LFPs) in brain-computer interfaces. This method uses a retrospective distillation process to transfer knowledge from a powerful offline model to a more efficient causal model. REALM significantly improves decoding accuracy compared to existing LFP-based methods while reducing model size and training time, offering a practical alternative to spike-based decoding for next-generation BCIs. AI

Researchers have introduced Slot-MPC, a novel framework for goal-conditioned model predictive control that utilizes object-centric representations. This approach aims to improve generalization in agents by modeling scene dynamics at an object level, allowing for more adaptable action planning. The system employs vision encoders to create these object-specific representations and a differentiable dynamics model that enables efficient, gradient-based optimization for planning, outperforming non-object-centric baselines in simulated robotic manipulation tasks. AI

Researchers have developed Viverra, a system designed to address the lack of correctness guarantees in AI-generated code. Viverra synthesizes C programs from natural language descriptions and automatically produces candidate assertions about their safety and correctness. These assertions are then verified using a suite of bounded model checkers, aiming to improve developer efficiency by reducing the need for manual code review and testing. AI

Researchers have developed a new framework that uses multiple BERT-based models to analyze digital traces like social media posts and chats for shifts in depression status. The system combines signals for sentiment, emotion, and depression severity, organizing them into temporal trajectories to identify changes over time. An integrated large language model generates human-readable reports detailing these mental health signal evolutions and key transitions, offering a more interpretable view than direct LLM reporting. AI

Researchers explored how AI-use disclosures in news articles affect reader attention and cognitive load. Their study found that brief, one-line disclosures significantly increased eye-tracking metrics like fixation duration and saccade counts, especially for AI-edited content. However, detailed disclosures did not impose an additional burden, and overall cognitive load, measured by NASA-TLX and pupil diameter, remained unaffected by disclosure detail. The findings suggest that brief disclosures might prompt more visual scrutiny due to insufficient information, and interviews indicated a preference for detailed or on-demand disclosure designs. AI

Researchers have developed a new flow-matching planner for autonomous driving that directly generates control trajectories. This model, conditioned on a bird's-eye-view scene representation, outputs acceleration and curvature profiles through Ordinary Differential Equations for low-latency inference. Trained on urban scenarios, the planner demonstrated reliable generalization to out-of-distribution environments like highways, maintaining stable control. AI

Researchers have introduced a new framework called Entity-Rubrics to evaluate how well AI models understand and execute abstract image editing instructions, moving beyond simple literal commands. This framework breaks down complex edits into smaller, entity-level assessments, correlating well with human judgment. A new benchmark, AbstractEdit, was also created to test 11 leading models, revealing that current architectures struggle to balance user intent with image preservation, often leading to over or under-editing. The study suggests that integrating advanced LLM text encoders and iterative reasoning is crucial for improving performance in this area. AI

Researchers have introduced LPH-VTON, a novel framework designed to improve virtual try-on technology by addressing the trade-off between garment structure and texture. The method decomposes the generation process, first using a structure-focused model to create a stable body-garment scaffold. Subsequently, a texture-focused model takes over to render high-fidelity details, achieving a better balance between geometric accuracy and visual realism. Experiments on the VITON-HD dataset demonstrate that LPH-VTON sets new benchmarks for perceptual quality while maintaining strong structural alignment. AI

An AI app designed to analyze users' poop images for gut health insights has been found to be selling a database of this sensitive data. The app, PoopCheck, collected over 150,000 images from approximately 25,000 users, with the developer now seeking to sell this dataset for machine learning training or medical studies. This practice contradicts the app's stated privacy policies, which claim no data collection and emphasize user privacy. AI

Alibaba Health is consolidating its B2C and O2O pharmaceutical business under a unified command structure led by CEO Shen Difan. This move aims to improve synergy and collaboration within the company's healthcare operations. The announcement was made during the Q4 fiscal year 2026 earnings call. AI

JR East is developing an AI robot to automate track inspections, aiming to reduce labor and enhance safety. The robot will autonomously navigate railway lines, using sensors and AI to detect track abnormalities and transmit data to remote operators. This initiative, in partnership with Preferred Robotics, is slated for testing on various lines starting in late 2026, with the goal of mitigating risks associated with manual inspections during natural disasters and improving working conditions. AI

Microsoft has discontinued its AI-powered coding assistant, GitHub Copilot Chat, for its own internal use. The decision comes amid concerns about the direction GitHub has been taking under Microsoft's ownership. This move suggests a potential shift in Microsoft's strategy regarding AI integration within its developer tools. AI

YouTube is integrating AI-powered conversational search and other features into its smart TV and gaming console experience as part of its Q1 2026 update. This update includes AI search, gaming title cards, family groups, and chapter navigation. Meanwhile, SEO expert Lily Ray cautions that the industry is misinterpreting AI search, leading to tactics like GEO spam and listicle manipulation that Google and Microsoft are now identifying as "LLM poisoning." AI

CookSleep's gpt_image_playground has emerged as a rapidly growing open-source AI project, according to a recent report. The tool, which enables users to generate and interact with images, has achieved a growth score of 67.40. Its popularity is further evidenced by its 1,193 GitHub stars. AI

AI is revolutionizing drug discovery, with new multimodal models like MAML capable of understanding complex biological interactions. These advanced AIs can identify novel treatments for diseases such as cancer and repurpose existing drugs for new applications, significantly reducing development time and cost. While the first AI-designed drugs are nearing FDA approval, the long-term impact on humanity hinges on how effectively these powerful tools are managed. AI

Consumers are increasingly accepting AI tools in retail, expecting them to enhance personalization, speed, and convenience, according to new research. A SOTI report indicates a significant portion of shoppers desire AI for improved product recommendations and visual search capabilities. While embracing AI-driven personalization and tailored offers, consumers also express concerns about data privacy and transparency in how their information is utilized by retailers. AI

A user has discovered a free coding platform that leverages Claude to manage its own bugs and deploy code automatically. The platform, referred to as a "Vibe Coding Platform," appears to offer unlimited usage without any apparent drawbacks, according to the user's experience. AI

The travel platform trip1.com is now enabling AI assistants to book hotel stays autonomously. This integration allows AI agents to handle the entire booking process, including payments via cryptocurrency. The system aims to streamline travel arrangements by removing the need for human intervention in booking and payment. AI

Anthropic has publicly reinforced its transfer restrictions for its AI models, specifically naming secondary venues. This action has led to a sharp repricing of these models in the secondary market. The move highlights concerns about the integrity of transfer restrictions and the challenges in enforcing them. AI

A new command-line tool called audit-mcp-cli has been released to help developers identify and manage vulnerabilities in their Node.js project dependencies. The tool provides a more structured and detailed report than standard audit commands, including full dependency chains and CVSS scores. Notably, it can function as an MCP Server, allowing AI coding assistants like Claude and Cursor to directly interact with it to audit projects and suggest fixes. AI

Traditional service discovery methods struggle with the dynamic and descriptive nature of AI agents. Semantic search offers a solution by indexing server capabilities as dense vectors, allowing discovery through natural language queries rather than rigid identifiers. This approach is particularly useful for finding AI agents with specific, nuanced functionalities that might not be captured by conventional tags or DNS records. AI

Ultrahuman has released its Ring Pro, a redesigned smart fitness tracker that aims to compete with market leader Oura. The new ring features a dual-core chip with onboard machine learning and significantly increased internal storage, allowing for more sophisticated AI-driven insights. It also boasts an all-titanium body for improved durability and a new charging system designed to mitigate thermal issues. AI

A Grafana dashboard has been developed to enhance the monitoring of Claude Code applications by visualizing OpenTelemetry metrics. This integration allows developers to track application performance in real-time, identify bottlenecks, and make data-driven decisions. The dashboard, which can leverage Prometheus for data storage, aims to improve efficiency and support troubleshooting efforts. AI

FetchSandbox has launched MCP, a tool designed to enhance API integration workflows within IDEs. This new feature allows AI coding agents to not only read API documentation but also to execute and validate entire integration workflows directly from the development environment. By simulating state transitions, webhook behaviors, and end-to-end execution, FetchSandbox MCP aims to provide developers with immediate feedback and context, accelerating the integration process before final provider sandbox testing. AI

MiniMax has launched Mavis, an agent framework designed with a "three ministries and six boards" structure. This initiative aims to provide advanced and informative technology investment news, focusing on global entrepreneurs and achieving a high financing rate. AI

MiniMax has launched Mavis, an AI agent system designed with a "three provinces and six ministries" framework. This system aims to provide cutting-edge technology and investment news, focusing on global entrepreneurs and financial information services. AI

MiniMax has launched Mavis, an AI agent system described as having "three provinces and six ministries." The company is known for its focus on AI technology and has previously released models like MM1. AI

Microsoft has developed a new cybersecurity system called MDASH, which utilizes over 100 specialized AI agents to identify vulnerabilities in Windows. This system successfully uncovered 16 security flaws, including four critical ones, during a single Patch Tuesday update. The use of multiple AI agents in a competitive or collaborative manner represents a significant advancement in proactive software security. AI

Tmall's Apple Store flagship is offering its first official price reduction on the iPhone 17 Pro series, with discounts up to 1000 yuan. This promotion, running from May 15th to June 21st, also includes deals on iPads, Apple Watches, and AirPods. Separately, the report notes a significant drop in New York silver futures, falling over 5% to $81.02 per ounce. AI

Meta is facing internal backlash from employees in the US and UK over new corporate surveillance software. This AI-powered tool records keystrokes and mouse movements, prompting an internal protest that has gone viral among staff. The situation is further complicated by ongoing federal efforts to expand surveillance capabilities, particularly concerning protesters. AI

A new pair of AI-powered glasses, priced at 356 euros, has been introduced. These glasses feature an integrated video and photo camera positioned on the nose bridge, offering a unique point-of-view perspective for recording. AI

Google Analytics has updated its default channel grouping to specifically identify traffic originating from AI assistants. This new categorization aims to provide marketers with clearer insights into how users are interacting with AI-powered tools like ChatGPT and Gemini, enabling more precise campaign tracking and strategy development. AI

Talairis Law Group, a new Seattle-based law firm, is built entirely around AI agents to perform tasks traditionally handled by junior lawyers. The firm's founders, who previously worked at Perkins Coie, claim their AI-driven model can offer services at approximately half the hourly rate of large traditional law firms. AI

Netflix is reportedly establishing a new animation studio that will heavily utilize generative AI technologies. This initiative is indicated by recent job postings from the streaming giant, signaling a significant investment in AI-driven content creation. The move suggests a broader trend of major entertainment companies exploring AI to streamline and enhance production processes. AI

Raycast has launched a completely rebuilt next-generation version for macOS into public beta. This new version features an integrated AI capability, including OpenAI's GPT-5.4 mini, alongside a new indexer for faster searches, system-wide voice input, and enhanced shortcut support. The update also includes a significant UI redesign aimed at improving user experience and numerous productivity enhancements, with some paid features available for free during the beta period. AI