NerdNews

Nuro has received a permit to test its Lucid Gravity robotaxis on California streets without human operators. The vehicles will feature a multi-pronged sensor system and luxuries like rider-controlled heated seats. Uber plans to deploy 100,000 driverless vehicles in the US, including up to 35,000 powered by Nuro's self-driving technology.

Google's Gemma 4 AI models use speculative decoding to achieve up to 3x faster performance. The models leverage a form of speculative decoding to predict future tokens, speeding up generation without loss of quality. This technology can be used on various hardware configurations, including consumer GPUs and mobile devices.

Palo Alto Networks has warned customers of a critical-severity unpatched vulnerability in the PAN-OS User-ID Authentication Portal that is being exploited in attacks. The zero-day bug, tracked as CVE-2026-0300, allows unauthenticated attackers to execute arbitrary code with root privileges on Internet-exposed PA-Series and VM-Series firewalls.

Elon Musk demanded full control of OpenAI, but co-founders Greg Brockman and Sam Altman refused, leading to Musk's departure from the company's board in 2018. The incident is now part of a lawsuit between Musk and OpenAI's co-founders.

Samsung's valuation surged to $1 trillion driven by the AI boom, with shares rising over 10%. The company's high-bandwidth memory chips are in high demand for AI systems, leading to increased profits. However, Samsung faces competition and potential headwinds, including labor unrest and chip shortages.

Google is updating its AI search results to include expert advice from Reddit and other online platforms. The new feature will display a preview of perspectives from public online discussions, social media, and other firsthand sources. Google will also recommend in-depth articles and link to more sources directly in its generated answers.

Anthropic has struck a deal with SpaceX to use its Colossus 1 data center, resulting in doubled rate limits for paid Claude Code users and loosened other limits. The partnership adds over 300 megawatts of new capacity and may lead to the development of orbital AI compute capacity.

Wes Reisz discusses AI-first software delivery, emphasizing the need for a strategic approach based on code longevity and automated verification. He introduces the RIPER-5 framework to amplify engineering discipline and ensure successful implementation.

Google has unveiled its 8th generation of Tensor Processing Units (TPUs), featuring two specialized chips designed to accelerate model training and agent workflows. The new TPUs deliver better performance, memory, and energy efficiency, with a 3x increase in compute performance over the previous generation.

OpenAI's cofounder and president, Greg Brockman, stated that the company expects to spend $50 billion on computing power by the end of the year. This expenditure is largely dependent on investments from other companies, such as Microsoft, Amazon, and SoftBank, which are contingent on OpenAI using their compute resources. The company's ability to turn a profit remains uncertain, despite its significant investments in AI research and development.

Taiwanese chipmaker TSMC signs a 30-year corporate power purchase agreement for 100% of the power produced by the Hai Long offshore wind project to meet its growing energy needs amid soaring AI chip demand and an energy crunch in Taiwan.

Broadcom has updated VMware Cloud Foundation (VCF) to provide a secure and cost-effective infrastructure platform for production artificial intelligence (AI) workloads. VCF 9.1 delivers an AI and Kubernetes native private cloud platform with integrated security and mixed compute infrastructure support. The update aims to address common challenges faced by enterprises when running AI workloads, including data and IP privacy concerns, surging infrastructure costs, and readiness for agentic AI.

Amazon Web Services has introduced a new service that enables AI agents to drive its virtual cloudy desktops, known as WorkSpaces. This service, currently in preview, allows users to assign agents an identity using Amazon's Identity and Access Management service, giving them access to a WorkSpace at a unique pre-signed URL. Agents can then interact with the desktop using computer vision, taking screenshots or video, interpreting what they 'see', and taking action. However, research by Reflex suggests that using agents can be 45 times more expensive than using APIs, with one benchmark finding that an agent needed half a million tokens to click on a dropdown menu.

Google DeepMind has acquired a minority stake in Fernris Creations, the maker of the popular MMORPG EVE Online. The partnership will explore new AI-enabled gameplay experiences and improve the understanding of intelligence in complex, dynamic systems.

Aurora Innovation will start hauling loads in driverless trucks for distribution giant McLane, operating autonomously between Dallas and Houston with plans to expand to new routes by the end of the year.

LinkedIn has introduced a unified integrations platform to standardize and reconcile hiring data across disparate systems, improving data quality and enabling downstream AI applications. The platform reduces partner onboarding time by 72% and expands data coverage, allowing external partners and internal systems to integrate without custom transformations.

An attacker purchased 30 WordPress plugins and introduced a PHP deserialization backdoor, affecting 400,000 installations. The backdoor was dormant for 8 months before injecting cloaked SEO spam into sites. WordPress.org permanently closed all 31 plugins, and a forced auto-update was pushed to neutralize the phone-home mechanism.

This article discusses a metrics-driven approach to achieving sustained iOS performance on real devices, highlighting the importance of testing on physical devices, understanding cross-metric amplification, and using Xcode Instruments for profiling. It also provides case studies and reference thresholds for production-grade iOS apps.

Proximie utilizes computer vision and generative AI to optimize surgery logistics, reducing 'hanging around' time and improving efficiency. The company's hybrid edge-to-cloud architecture manages 120TB of unstructured video data, ensuring patient data sovereignty and security.

Extreme Networks has launched Agent ONE, a new class of AI agents for enterprise networking. Agent ONE moves beyond generic, prompt-based AI and provides proactive, context-aware intelligence with real-time decision-making and automated execution at machine speed. The AI stack has four layers: infrastructure, core, skills, and agentic, which provide reasoning, context, procedures, and execution. Agent ONE is part of the Extreme Platform ONE system and will be released in two modes: Coworker and Operator.