AI Gets Chatty: New Search & Assistant Features

The push to make AI more conversational and proactive is accelerating. YouTube is testing an AI-powered search feature for Premium subscribers that provides guided, chat-like answers by pulling from videos and Shorts. Meanwhile, Otter’s new feature lets users search across their enterprise tools, and the company is releasing a Windows app that can capture meeting notes without joining the call, pushing its assistant further into the workflow. On the operating system front, Canonical has laid out a plan for AI in Ubuntu Linux, signaling a wave of AI integration into core computing platforms. Source: TechCrunch | Source: TechCrunch | Source: The Verge

Building Better Brains: From Encoders to Independent Learning

Two stories highlight foundational shifts in how AI is built. A primer explains the evolution of encoders, the crucial but often overlooked components that translate real-world data into a language AI can understand, tracing their journey to today’s multimodal systems. In a more radical development, DeepMind co-founder David Silver’s new lab, Ineffable Intelligence, raised $1.1 billion to pursue a monumental goal: building AI that learns without human data. This push for “human-independent” learning could redefine the field. Source: AI News | Source: TechCrunch

Policy & Power Plays: Courtrooms and Boardrooms

Legal and regulatory battles are heating up. The blockbuster trial between Elon Musk and OpenAI has begun, with jury selection revealing that “people don’t like him,” setting the stage for a messy public airing of the company’s founding secrets. In the business realm, OpenAI ended a legal standoff with Microsoft, securing the right to sell on AWS while giving Microsoft a larger revenue share. Geopolitics also intervened, as China blocked Meta’s $2B acquisition of Manus, a major setback for Zuckerberg’s AI agent ambitions. Source: The Verge | Source: TechCrunch | Source: TechCrunch

In Other News

• Government AI: Google reportedly signed a classified deal allowing the Pentagon “any lawful” use of its AI, a move that came just a day after hundreds of Google employees, including senior staff, signed a letter demanding the company block such military use. Meanwhile, OpenAI announced its services are now available at FedRAMP Moderate authorization for U.S. federal agencies. Source: The Verge | Source: The Verge | Source: OpenAI
• Security & Infrastructure: A popular open-source npm package, “element-data,” with 1 million monthly downloads was caught stealing user credentials. On the hardware side, Lightelligence’s stock soared 400% on its debut, a massive bet that optical interconnects, not just chips, are AI’s next critical bottleneck. Source: Ars Technica | Source: AI News
• Startup Scene: BCI startup Neurable is looking to license its non-invasive “mind-reading” tech for consumer wearables. Red Hat released Tank OS, a container system to make deployments of OpenClaw AI agents safer and more reliable. Pre-launch, investors backed Skye’s AI home screen app for iPhone, showing high demand for deeper AI integration into mobile. Source: TechCrunch | Source: TechCrunch | Source: TechCrunch

Editorial Take: Today’s news underscores a central tension in AI’s evolution: the race to make it more useful and accessible is running in parallel with profound debates over its control, safety, and fundamental design. We see conversational AI becoming a standard feature, while billion-dollar bets are placed on reinventing how AI learns from the ground up. Simultaneously, the courtroom drama and employee revolts remind us that the trajectory of this technology is being shaped as much by human conflicts and ethical choices as by raw technical innovation. The future of AI is being written in code, in boardrooms, and now, in a San Francisco courtroom.