Top Stories
Robotics Funding Enters Its Mega-Round Era
The robots are coming, and venture capital is betting big on who will build them. Physical Intelligence, founded by former Google DeepMind researchers, is in talks to raise another $1 billion at an $11 billion valuation—doubling their worth in just four months.
They’re not alone. In a single week, robotics startups collectively raised over $1.2 billion: Mind Robotics ($500M), Rhoda AI ($450M), Sunday ($165M), and Oxa ($103M). Shield AI, focused on autonomous military drones, closed $1.5 billion at a $12.7 billion valuation after winning a U.S. Air Force contract.
The thesis behind these bets: AI models that can control physical robots are finally becoming capable enough to handle real-world tasks. Physical Intelligence’s π0 foundation model, open-sourced in February, demonstrated robots learning to perform multiple tasks from a single model rather than needing separate programming for each action.
Source: TechCrunch, Bloomberg
AI System Writes Paper That Passes Nature Peer Review
For the first time, a fully AI-generated scientific paper has passed peer review and been published in Nature. The system, called The AI Scientist v2, was developed by researchers at Sakana AI, University of British Columbia, Vector Institute, and University of Oxford.
The AI Scientist autonomously generates research ideas, writes code, runs experiments, analyzes data, produces figures, and writes the complete manuscript—including its own peer review. The accepted paper passed the first round of peer review at a machine learning workshop with a 70% acceptance rate.
The achievement comes with caveats. The system sometimes produces “naive or underdeveloped ideas,” struggles with deep methodological rigor, and occasionally hallucinates citations or duplicates figures. It’s a tool that accelerates research, not a replacement for human scientists—at least not yet.
Source: Sakana AI, Scientific American, Nature
NOAA’s AI Weather Models Now Operational
NOAA has deployed a new generation of AI-driven weather prediction models that dramatically cut computing costs while extending forecast accuracy. The Artificial Intelligence Global Forecast System (AIGFS) produces a 16-day forecast using just 0.3% of the computing resources of traditional models.
The AI ensemble system extends forecast skill by 18-24 hours compared to physics-only approaches. Perhaps most interesting: a hybrid system combining AI and traditional physics models consistently outperforms both pure approaches. NOAA also released HRRRCast Version 3 in March, upgrading regional AI forecasts.
Source: NOAA
Quick Hits
-
Anthropic’s rough month: Five outages in March, including disruptions lasting hours. Rapid growth in Claude’s user base is straining infrastructure. The New Stack
-
Mistral’s Voxtral TTS: New open-source text-to-speech model supports nine languages including English, French, German, and Hindi. TechCrunch
-
White House AI framework: The administration’s legislative recommendations emphasize sector-specific oversight through existing agencies rather than a new federal AI regulator. Cooley
-
Alibaba’s Qwen 3.5: The 9B parameter open-source model outperformed models 13x its size on graduate-level reasoning benchmarks, scoring 81.7 on GPQA Diamond vs. GPT-OSS-120B’s 73.5. VentureBeat
-
Model Context Protocol hits 97M installs: MCP is transitioning from experimental standard to foundational agentic infrastructure, with every major AI provider now shipping MCP-compatible tooling. AI Unfiltered
Worth Watching
Apple’s AI-powered Siri overhaul remains perpetually “coming soon.” Originally targeted for March 2025, then delayed to March 2026, now some features may slip to iOS 27 in September. The company has confirmed plans to open Siri to multiple AI assistants—but first they need to ship the Gemini-powered upgrade that’s been stuck in development for two years.
Anthropic’s “Mythos” model, accidentally revealed through an unsecured data lake, reportedly represents a “step change” in capabilities. The company is preparing for an IPO as early as October at a valuation above $60 billion, putting it in direct competition with OpenAI’s own public listing plans.