🤖 The Unglamorous Grind of Building Robot Smarts

If large language models learned from the internet’s text, how do physical AI systems learn about the real world? They need data—lots of it—and collecting that data is dirty, unglamorous work. Some AI labs are now paying startups like XDOF to handle this crucial but messy task, which involves capturing the nuances of physical interaction that pure simulation can’t replicate. Read more on TechCrunch

This comes as the definition of “humanoid” is being stretched. French startup Genesis AI argues that “humanoid robots don’t need to look human,” unveiling Eno, a robot that might sit on wheels and fold down like a deck chair. Read more on The Verge. Meanwhile, from MIT, a new spatial memory system for robots could one day help them (or you) remember where you left your keys by efficiently cataloging objects in an environment. Read more on MIT News

🔬 Research & Reliability: AI Gets Serious

A major hurdle for AI adoption in high-stakes fields is trust. Pramaana Labs has raised a massive $27 million seed round led by Khosla Ventures to tackle this by bringing formal verification—a method for mathematically proving software correctness—to AI models. The startup is targeting sensitive verticals like law, drug discovery, and tax preparation where errors are costly. Read more on TechCrunch

In healthcare, new research published in Nature shows Google’s conversational AI system, AMIE, matches primary care physicians in complex disease management tasks, a significant step for AI-assisted medicine. Read more on the Google Blog. Google DeepMind is also applying AI to public infrastructure, partnering with the UK government on a prototype to accelerate housing planning decisions. Read more on the DeepMind Blog

🛍️ Products & Platforms: AI in Your Pocket, Home, and Feed

The product rollout continues. Pinterest launched an experimental AI shopping app called ‘Ask Pinterest,’ using a conversational interface for recommendations. Read more on TechCrunch. Google is finally shipping its first new smart speaker in six years, the Google Home Speaker, on June 25th, powered by Gemini. Read more on The Verge. This follows the launch of Android 17, which expands Gemini features across Google’s Pixel devices. Read more on TechCrunch

Translation powerhouse DeepL is expanding its U.S. footprint, acquiring Mixhalo for its live-event audio streaming tech, hinting at real-time translation for concerts and conferences. Read more on TechCrunch

⚖️ Policy, Power, and Public Perception

The intersection of AI, policy, and public trust is growing more complex. A new survey indicates 60% of U.S. consumers find ‘AI’ in brand messaging a turnoff, wary of AI-generated answers. Read more on TechCrunch. This skepticism surfaces as Meta’s new AI Mode in search draws criticism for getting things wrong while trying to answer open-ended questions like “What should I do this weekend?” Read more on The Verge

On the policy front, the DOJ is intervening to allow xAI to continue using unpermitted gas turbines, citing national and economic security needs, highlighting the immense energy demands of AI. Read more on TechCrunch. Meanwhile, Anthropic’s latest regulatory feud with the Trump administration may ironically be boosting its popularity with business users, according to spending data. Read more on TechCrunch

💡 Editorial: The Data Grind Meets the Truth Test

Today’s news underscores a pivotal maturation phase for AI. The field is moving beyond pure software brilliance to confront the hard, physical realities of data collection for robotics and the equally hard problem of trust in critical applications. One startup is getting its hands dirty so robots can learn; another is building a mathematical clean room to verify AI isn’t hallucinating. This dual focus—on grounding AI in the real world and ensuring its outputs are reliable—is what will separate impactful, enduring applications from mere hype. The next breakthrough might not just be a smarter model, but a more verifiable and physically aware one.