Jensen Huang's AI habit: make the models fight each other

Jensen Huang's rule for using AI is simple: don't use it as a crutch for things you can do. He uses it for two jobs, learning things he doesn't know and cracking problems he couldn't reasonably solve alone. He never hands it his thinking.

His workflow is basically a multi-model debate. He doesn't accept the first reply. He pushes back ("are you sure this is the best answer you can provide?"), gives one AI's output to another to critique, and asks the same question of several models so they can "compare each other's notes" and merge the best parts. His view is that prompting is "a highly cognitive skill" and that questioning the answers is what sharpens critical thinking. "As a CEO, I spend most of my time asking questions."

The quotes are going around again, but they come from Huang's CNN interview with Fareed Zakaria in mid-2025. He was answering MIT Media Lab's "Your Brain on ChatGPT" study. It found that essay writers who leaned on ChatGPT showed lower brain activity, drifted into copy-pasting, and struggled to recall their own work. Huang's answer was that he uses AI every day and his cognitive skills are "actually advancing". In his view the problem is how people use AI, not the tool.

Why it matters: making models cross-check each other is now standard practice for anyone who needs AI answers they can trust.