NerdNews

The Nvidia and OpenAI partnership could dominate the AI industry, similar to how Microsoft and Intel dominated the PC market. The $100 billion deal gives OpenAI access to millions of Nvidia GPUs and capital for next-generation data centers, potentially leaving other AI companies without the necessary resources.

Anthropic launches Claude Sonnet 4.5, its best AI model for coding, offering state-of-the-art performance on coding benchmarks and the ability to build production-ready applications. The model will be available via the Claude API and in the Claude chatbot, with pricing starting at $3 per million input tokens.

OpenAI has introduced a safety routing system and parental controls on ChatGPT to detect and redirect sensitive conversations. The system switches to GPT-5, a model trained with 'safe completions,' to handle high-stakes safety work. Parental controls allow customization of teen experiences, including quiet hours and content protections.

Anthropic's new AI model, Claude Sonnet 4.5, has demonstrated the ability to maintain focus for over 30 hours on complex, multi-step tasks. The model has also shown improved coding and computer use capabilities, beating OpenAI and Google on coding tests.

Google DeepMind has unveiled its first 'thinking' robotics AI, Gemini Robotics 1.5, which uses generative AI to output robot actions. The system consists of two models: one for thinking and one for action, allowing robots to undertake complex tasks without needing to be reprogrammed.

Microsoft has launched a new Agent Mode in Excel and Word, allowing users to generate complex spreadsheets and documents with just a prompt. The feature is powered by OpenAI's GPT-5 model and Anthropic models, and is available in the Frontier program for Microsoft 365 Copilot customers.

DeepSeek is a Chinese AI lab that has gone viral with its chatbot app, which has risen to the top of the Apple App Store charts. The company's AI models, trained using compute-efficient techniques, have led to questions about the US lead in the AI race and the demand for AI chips. DeepSeek's models are available under permissive licenses, allowing for commercial use, and have been adopted by developers. However, the company's success has also raised concerns about data security and propaganda.

Alex, an AI recruiter startup, has raised $17M in Series A funding to automate initial job interviews, helping companies conduct video interviews and phone screens, and freeing up recruiters' time to build relationships with pre-qualified candidates.

Former Microsoft executives have launched Maximor, a startup that uses AI agents to automate financial tasks, aiming to replace Excel-driven finance for mid-market enterprises. The platform connects to ERP, CRM, and billing systems, providing real-time financial visibility and streamlining audits.

Microsoft is introducing 'vibe working' with new AI-powered tools in its Office products. The tools, including Office Agent and Copilot, allow users to work alongside AI to develop documents and presentations. The tools support Excel and Word workflows, with PowerPoint support coming soon.

A self-driving Waymo car was pulled over by police in California for making an illegal turn, but the officers were unable to issue a ticket as there was no human driver. The incident highlights the need for updated laws to handle autonomous vehicles, with changes set to come into effect in July 2026.

OpenAI introduces Instant Checkout, allowing ChatGPT users to make purchases within conversations. This feature, available to ChatGPT Pro, Plus, and Free logged-in users, enables frictionless shopping experiences and could reshape the e-commerce landscape, potentially challenging Google and Amazon's dominance.

ComplexChaos, a startup, is developing an AI tool to help people find common ground and arrive at a consensus. The tool uses large language models like Google's Habermas Machine and OpenAI's ChatGPT to generate questions, come up with goals for conversations, and help summarize long documents. In a trial with delegates from African countries, the tool reduced coordination time by up to 60% and helped 91% of participants see perspectives they would have otherwise missed.

Top AI companies like Google DeepMind, Meta, and Nvidia are shifting focus to world models, which learn from videos and robotic data to understand human environments, as large language model advancements slow. This new approach has potential applications in self-driving cars, robotics, and entertainment.

Apple has released iOS 26, which features a new design called Liquid Glass, improved multitasking, and several other updates. The new operating system is available for download on eligible iPhones. iOS 26 also includes new features such as live translation, polls in group chats, and enhanced audio recording. Additionally, iPadOS 26 has been released with multitasking and real windowing capabilities.

Google is enhancing its Colab Enterprise notebooks to integrate SQL, Python, and Apache Spark, eliminating the need to switch between environments. The update includes native SQL cells, interactive visualization cells, and enhanced Data Science Agent. This move aims to increase productivity in data science and machine learning, and is part of a larger trend of vendors like Snowflake and Databricks trying to bridge the gap between data analytics and machine learning.

Disaggregated serving architectures optimize prefill and decode phases in LLM inference, improving performance, cost, and operational metrics. Frameworks like vLLM, SGLang, and TensorRT-LLM demonstrate up to 6.4x throughput improvements and 20x reduction in latency variance. Organizations can reduce total infrastructure costs by 15-40% through optimized hardware allocation and improved energy efficiency.

Researchers struggle to detect AI models trained for malicious purposes, with current methods being ineffective. Transparency in training data and supply chain regulation may be the key to preventing deceptive AI.

California Governor Gavin Newsom has signed SB 53, a bill that sets new transparency requirements on large AI companies, including OpenAI, Anthropic, Meta, and Google DeepMind. The bill requires these companies to be transparent about safety protocols, ensures whistleblower protections for employees, and creates a mechanism for reporting potential critical safety incidents.

Meta is expanding its facial recognition technology to the UK, EU, and South Korea to combat scammers impersonating public figures. The technology will be used to identify and remove impostor accounts, and will be available on Facebook and Instagram.