⚡️ NEW: Anthropic's latest research uncovers a "global workspace" inside Claude, revealing internal reasoning steps the model performs without writing them down. (Cointelegraph)