NerdNews

Nvidia will spend $26 billion over the next five years to build open-source AI models, positioning itself to compete with OpenAI and DeepSeek. The company released its most capable open-weight AI model, Nemotron 3 Super, with 128 billion parameters, and claims it outperforms other models across several benchmarks.

Andre Ribeiro discusses Healthily's AI symptom checker, which uses Bayesian inference and RAG models to bridge the gap between medical insights and confident patient action, and its evolution into conversational chatbots for safer and more reliable healthcare.

Meta's acquisition of Moltbook points to a future where AI agents interact with each other, potentially changing the advertising landscape. AI agents could negotiate directly with consumer agents, and Meta could capitalize on this market by deciding which agents talk to each other. This could expand Meta's ads business into new territory, but it depends on consumer adoption of the agentic web.

Google has completed its $32 billion acquisition of Israeli cybersecurity firm Wiz, marking its largest acquisition to date. Wiz provides a security platform that protects major cloud environments and will maintain its brand while joining Google Cloud.

Oracle is investing $533bn in AI infrastructure, with a focus on cloud infrastructure and AI acceleration hardware. The company is committed to meeting its remaining performance obligation, but there are concerns about the return on investment and the potential for an AI bubble. Oracle's cloud infrastructure revenue has increased by 84%, but the company is also planning thousands of job cuts as it drives forward its datacentre expansion plans.

OpenAI's ChatGPT now features interactive visuals for over 70 math and science topics, allowing users to engage directly with formulas and variables in real-time. This new feature expands ChatGPT's educational capabilities, making it easier for users to understand complex concepts.

An autonomous AI-powered bot compromised GitHub Actions workflows across major open-source repositories, achieving remote code execution and stealing credentials. The attacks targeted projects from Microsoft, DataDog, and the Cloud Native Computing Foundation, exploiting vulnerabilities such as untrusted data flowing from source to sink without validation.

Oracle claims AI coding tools are helping it avoid the SaaSpocalypse. The company is using these tools to build new SaaS products and embed AI agents into existing applications, allowing smaller engineering teams to deliver more complete solutions to customers. Oracle's revenue for the quarter hit $17.2 billion, up 22 percent year-over-year, with AI infrastructure accounting for $4.9 billion of that.

Ford introduced an AI assistant to help its commercial customers monitor and analyze data, including seatbelt use and vehicle health. The AI assistant, called Ford Pro AI, is built on Google Cloud and uses internal data to reduce errors. It provides detailed information on fuel consumption, idle times, and acceleration events, and is included in the Ford Pro telematics subscription.

Nvidia's Xinzhou Wu discusses the company's approach to autonomous driving, combining end-to-end AI models with traditional engineering protocols. Nvidia's DRIVE Hyperion platform offers multiple configurations, including a base version with cameras and radar, and a higher-end version with lidar. Wu believes that Nvidia's vertically integrated approach can deliver safety performance at a lower cost, and that the company's simulation technology can help it catch up to competitors like Tesla and Waymo.

Meta has developed four new computer chips, MTIA 300, 400, 450, and 500, to power its AI features and content ranking systems. The chips are built on the open-source RISC-V architecture and will be used for training algorithms and inference. Meta partnered with Broadcom to develop the chips, which will be fabricated by Taiwan Semiconductor Manufacturing Corporation.

Nuro, a Silicon Valley-based startup, is testing its autonomous vehicle technology on Tokyo's streets. The company's self-driving software is being used in Toyota Prius vehicles with human safety operators. Nuro's AI approach allows its system to learn as it drives, navigating public roads in Tokyo without prior training on Japanese driving data.

Mind Robotics, a spin-out of Rivian, has raised $500 million in a Series A funding round to develop industrial AI-powered robots. The company aims to address the limitations of current industrial automation solutions by creating robots that can perform tasks requiring human-like dexterity and adaptation.

Researchers at Unit 42 found that large language models (LLMs) used to enforce safety policies can be manipulated into authorizing policy violations through prompt injection attacks. They developed an automated fuzzer called AdvJudge-Zero to identify trigger sequences that exploit an LLM's decision-making logic, achieving a 99% success rate in bypassing controls.

Wayve and Qualcomm collaborate to deliver end-to-end AI for autonomous vehicles, aiming to simplify ADAS/AD implementation and reduce development time and cost for automakers. The partnership combines Wayve's AI Driver with Qualcomm's Snapdragon Ride Platform, enabling scalable and reliable autonomous driving solutions.

A report on AI's potential impact on the economy caused a significant drop in the stock market, highlighting the ongoing anxiety and uncertainty surrounding AI's effects on the job market and economy. Despite criticisms of the report's methodology, the incident demonstrates how sensitive investors are to AI-related news and the potential for AI to disrupt traditional industries.

Uber redesigned its MySQL infrastructure using MySQL Group Replication, reducing failover time from minutes to seconds and improving cluster uptime. The new architecture embeds consensus within the database, ensuring strong consistency and automatic election of a new primary node if needed.

The increasing power demands of AI computing are pushing datacenter power architectures to their limits. Traditional 48V power distribution is no longer sufficient, and the industry is moving towards high-voltage DC (HVDC) power architecture, such as 800V, to reduce current and increase efficiency. This shift requires new designs, components, and standards to support the higher voltage and power requirements of AI workloads.

Zoox, an Amazon-owned company, plans to make its robotaxis available on the Uber app in Las Vegas later this year. The company needs federal approval to commercially deploy its robotaxis, which don't have a steering wheel or pedals. Zoox is seeking exemptions from eight federal safety standards and plans to launch its own commercial robotaxi service before making the vehicles available on Uber.

Anduril, a defense tech company, has acquired ExoAnalytic Solutions, a firm that operates a network of 400 telescopes to track spacecraft in high orbits. The acquisition will help Anduril improve its space domain awareness and missile defense capabilities, and could aid in winning government contracts, including the development of the Golden Dome air and missile defense system.