Google hired writers with red pens to fix AI writing. The @GoogleDeepMind team told us professional writers copyedited Gemini 4 Argon’s answers and wrote “golden examples” for it to learn from. Targets included “load-bearing,” “in a world,” and the reflex to make every answer a bullet list. Argon sounds less like a standard chatbot—but its drafts are also denser. In our 20-prompt writing benchmark, its prose averaged a 10.64-grade reading level, versus the seventh- or eighth-grade range we’ve…