NerdNews

Meta is introducing AI agent tools for businesses on WhatsApp, Instagram, and Messenger, allowing them to delegate tasks such as interacting with customers and booking appointments. The goal is for these agents to eventually help run entire businesses, with advanced capabilities like market research and competitive intelligence.

Google has announced a new feature to detect deepfake calls on Android devices. The feature, which requires Google's Phone, Contacts, and Messages apps, uses a confirmation signal to verify the caller's identity. If the signal is missing, the feature sends an authenticated RCS ping to the supposed caller to verify their identity. The feature is available on devices running Android 12 or higher. Additionally, Google has expanded AirDrop support to more Android devices.

General Motors is utilizing AI and machine learning to significantly accelerate its development process. By leveraging virtualization tools, the company can now perform tasks such as finite element analysis (FEA) runs in just one minute, a process that previously took 15 hours. This shift has enabled GM to explore a broader range of designs and optimize vehicle performance more efficiently. The technology is being applied across various domains, including motorsport, energy, and defense, allowing for faster iteration and innovation.

Microsoft has announced several new AI initiatives, including a super app, in-house reasoning models, and AI agents, as it competes with OpenAI in the AI market. The company's AI chief, Mustafa Suleyman, emphasized the goal of becoming one of the top four AI labs in the world, and Microsoft's CEO, Satya Nadella, highlighted the company's focus on AI cybersecurity and its new AI-powered tools.

Google Labs launches Dreambeans, an AI-fueled app that generates curated lifestyle suggestions based on user's Google data, aiming to provide inspirational ideas and reduce doomscrolling

Nvidia has partnered with Chinese robotics startup Unitree to create a humanoid robot called H2 Plus. The robot combines Unitree's 6-foot, 150-pound body with Nvidia's Thor T5000 chip and an advanced humanoid hand. The partnership aims to make it easier for researchers to build and train cutting-edge humanoids, despite security concerns and US-China techno-competition.

Microsoft's Discovery agentic AI tool has improved the quality of qubits in its next quantum chip, Majorana 2, enabling more reliable computation and putting the team on a path to achieve a scalable quantum computer by 2029.

Nvidia's game-changing DLSS is finally coming to Blender, allowing for higher resolutions and better visual fidelity without a performance hit. DLSS 4.5 Ray Reconstruction will be available in Blender 5.3, making it easier for 3D artists to work on lighting and lookdev interactively.

Google has developed a system for coordinated A/B testing across its global service fleet. The system provides a centralized experimentation framework, ensuring consistent and reliable experimentation while minimizing interference between tests. It uses a unified assignment layer and exposure logging to support correctness in measurement, and integrates guardrails to prevent experiments from exceeding traffic limits or violating safety constraints.

Sepehr Khosravi discusses the evolution of developer productivity tools, evaluating the strengths of tools like Cursor and Claude Code, and shares actionable techniques for senior engineers, including context engineering, custom rules, and Model Context Protocol (MCP) integrations.

Microsoft's Build conference featured announcements on native OpenClaw app for Windows, new Unmetered Agentic AI models, and a preview of Microsoft's newest quantum computer, Majorana 2.

The UK's Competition and Markets Authority has ordered Google to put clearer attributions and links to publishers' content in its AI-generated search features. Google must also give publishers a way to opt out of AI features in search. The changes aim to improve transparency and trust in Google's search results.

Google is introducing a new Android security feature to detect and flag phone calls where scammers use AI to impersonate personal contacts. The feature, called 'fake call detection,' works by sending a silent, encrypted confirmation signal between devices and will be enabled by default on Android 12 and later devices.

Uber plans to deploy 500 data-collection vehicles globally this year, equipped with sensors to collect 2 million miles of high-fidelity data per month for its autonomous vehicle partners, including Avride, Waymo, and WeRide.

Nvidia's RTX Spark laptops are a significant improvement in gaming PC technology, with the ability to run graphically capable games without compromising battery life. The laptops use Nvidia's new RTX Spark chip, which is based on ARM architecture and co-designed with MediaTek. The chip uses 6,144 CUDA cores and has up to 128GB of unified memory. Nvidia claims that the laptops will deliver 'all-day battery' and will support anti-cheat systems for multiplayer titles. The laptops will also have support from publishers like Remedy Entertainment and Riot Games, and will supposedly support Xbox with future titles.

A security researcher has released exploit code for a Visual Studio Code (VS Code) zero-day vulnerability that allows attackers to steal GitHub authentication tokens by tricking users into clicking a link. The vulnerability exploits VS Code's sandboxed webview message-passing system, allowing attackers to install malicious extensions that extract GitHub OAuth tokens.

Nvidia is planning at least two additional generations of RTX Spark chips, dubbed N2X and N3X, with the ultimate goal of creating Star Trek-like computers that can be controlled with voice commands. The company aims to make local AI compute more accessible and affordable, allowing users to interact with their devices in a more natural way.

Amazon will display AI-generated product images in its shopping app to help users find products based on their search queries. The feature uses visual search capabilities to show users various product styles, which can be clicked to direct them to relevant search results.

AethexAI, a startup founded by former Goldman and Meta employees, has raised $3 million to build voice AI for markets in Africa and the Middle East. The company built its own small model and orchestration layer from scratch to handle localized dialects and reduce latency. AethexAI is targeting the customer support and service sectors, and is already handling over 17,000 calls per day.

The article discusses two misconfigurations that caused Spark Out of Memory (OOM) failures on Kubernetes. The first misconfiguration was setting spark.kubernetes.local.dirs.tmpfs to true, which caused all local scratch directories to be backed by RAM instead of disk. The second misconfiguration was a hard rule that forced all executors onto the same node, concentrating shuffle-time memory pressure on a single node. The article provides a checklist to prevent similar failures in Spark-on-Kubernetes deployments.