Anthropic reveals Claude's hidden thinking space

Anthropic has cracked open the hood on its AI model Claude to see what’s really happening inside. Using a new tool called the Jacobian lens, or J-lens, researchers discovered a hidden zone they call J-space. Think of it as Claude’s internal scratchpad—words and concepts it’s juggling before deciding what to say. It’s not always pretty; some findings are mundane and others a touch unnerving. Alongside this, OpenAI dropped its long-awaited ‘super app’ that merges its chatbot, coding tools, and fresh models into one workhorse. On the robotics front, humanoids pulled off world-first teleoperated surgery on live pigs, while SK Hynix snagged the largest US listing ever by a foreign company, raising a cool $26.5 billion. Why it matters: Peeking into an AI’s ‘mind’ isn’t just academic—it’s a crucial step toward understanding model safety and trustworthiness before these systems become even more embedded in our daily lives.