NerdNews

Microsoft introduces Mico, a new virtual assistant for Copilot, designed to make interactions more natural and emotionally engaging. Mico is a modern interpretation of Clippy and features a Learn Live mode and an updated memory system. It's currently available in the US and is part of Microsoft's strategy to develop a distinct personality for Copilot.

Vercel introduced AI SDK 6 beta, Vercel Marketplace updates, use workflow library, Vercel Agent beta, and Vercel Python SDK beta. These tools aim to simplify AI development, deployment, and integration. The updates include agent abstraction, human-in-the-loop approval, and observability features.

OpenAI introduces ChatGPT Atlas, a browser with ChatGPT built-in, allowing for real-time assistance and task automation. Atlas integrates ChatGPT's memory and agent mode, enabling features like summarizing articles and performing multi-step tasks.

OpenAI released data showing over a million people talk to ChatGPT about suicide weekly. The company has updated its model to better respond to mental health issues, with a 65% increase in desirable responses. OpenAI has also added new evaluations to measure mental health challenges and rolled out more controls for parents.

AMD and the US Department of Energy have partnered to develop two AI supercomputers, Lux and Discovery, with a total investment of $1 billion. The supercomputers will be housed at Oak Ridge National Laboratory and are expected to accelerate AI-driven science and innovation. Lux is slated to come online in 2026, while Discovery will follow in 2029.

Researchers at NeuralTrust have discovered a vulnerability in OpenAI's Atlas browser that allows attackers to inject malicious prompts, potentially leading to harmful actions. The attack works by disguising a malicious prompt as a URL, which is then interpreted as trusted user input by the browser. This vulnerability highlights the need for stricter boundaries between trusted user input and untrusted content in agentic browsers.

The article discusses the challenges of vector embeddings in AI-driven microservices and presents 5 essential Vector Sync Patterns to solve these challenges. These patterns include Dependency-Aware Propagator, Semantic Change Detector, Versioned Vector Registry, Business Rule Filter Chain, and Adaptive Sync Orchestrator. The article also highlights the importance of event-driven architecture and technologies like Kafka and Flink in implementing these patterns.

The Spring ecosystem has released first release candidates for Spring Boot, Security, GraphQL, Integration, Modulith, and other projects, featuring bug fixes, new features, and dependency upgrades.

This article demonstrates how to build a Retrieval-Augmented Generation (RAG) application using Spring Boot, Spring AI, MongoDB Atlas Vector Search, and OpenAI. The RAG paradigm combines generation with the retrieval of information from corporate databases, ensuring accurate and transparent responses. The implementation presented creates a sentiment-based music recommendation system, leveraging ingestion, embedding, semantic search, and reranking pipelines.

Skyline Nav AI's Pathfinder software uses AI to recognize scenes and generate real-time navigation, useful in areas with blocked GPS signals and as a backstop against GPS jamming, with potential applications in various industries

Qualcomm has announced AI200 and AI250 chip-based accelerator cards for inferencing workloads, with the AI200 supporting 768 GB of LPDDR memory and the AI250 offering innovative memory architecture. The cards will be shipped in pre-configured racks with direct liquid cooling, PCIe, and Ethernet. Qualcomm claims its products will enable efficient and high-performance AI inference at a low cost.

AWS director Olawale Oladehin discusses the integration of AI agents in SaaS, citing complexity in deployment and commercial models. He highlights the importance of choice and flexibility in pricing, as seen in the examples of Pegasystems and Zendesk. The use of AI agents is expected to simplify application integration with open standards, but commercial arrangements are still evolving.

Researchers found that AI-powered search engines tend to cite less popular websites and ones that wouldn’t appear in the top 100 links listed in an organic Google search. The study compared traditional link results from Google to its AI Overviews and Gemini-2.5-Flash, as well as GPT-4o’s web search mode.

Google's AI health coach is a chatbot that helps users reach fitness and health goals. It starts with a 5-10 minute conversation to assess goals and can be used to review and adjust fitness plans, create workouts, and track sleep metrics. A preview version is launching for some Fitbit Premium users in the US, with an iOS version in the works.

RADiCAIT, an Oxford spinout, uses a generative deep neural network to convert CT scans into PET scans, making diagnostic imaging more affordable and accessible. The company has begun clinical pilots for lung cancer testing and is pursuing an FDA clinical trial.

OpenAI, founded as a nonprofit in 2015, is proposing a shift to a for-profit structure, sparking controversy and criticism from cofounder Elon Musk and nonprofit groups. The company has served subpoenas to advocacy groups opposing the restructure, prompting concerns about its commitment to its charitable mission.

Zoom CEO Eric Yuan believes AI will transform work, enabling a shorter workweek. He discussed AI's potential at TechCrunch Disrupt 2025, including a 'digital twin' feature and AI-assisted email management.

Waymo's co-CEO Tekedra Mawakana emphasized the importance of scaling safely, aiming for 1 million trips per week by 2026. She highlighted the need for transparency in proving autonomous vehicle safety and noted that Waymo's vehicles are already five times safer than human drivers.

Nephrogen, a biotech startup, combines AI and gene therapy to reverse kidney disease. The company's founder, Demetri Maxim, was motivated by his own experience with Polycystic Kidney Disease (PKD). Nephrogen has developed a delivery mechanism that is 100 times more efficient than current FDA-approved methods. The company plans to advance its novel delivery mechanism into clinical studies in 2027 and is raising a $4 million seed round.

The term 'clinical-grade AI' is being used by companies like Lyra Health to describe their AI chatbots, but it has no specific regulatory meaning and is not recognized by the FDA. Experts say this term is an example of marketing puffery designed to borrow authority from medicine without the strings of accountability or regulation.