Huang Renxun: 90% of my daily prompts are questions; AI makes the brain smarter
NVIDIA CEO Huang Renxun said in a CNN exclusive interview that he uses AI extensively every day, but never lets it think for him; instead, he treats AI as a private tutor to learn entirely new fields. He stated that 90% of his daily prompts are essentially high-quality questions posed to his team and the model, rather than commands. He believes that asking questions is itself an extremely mentally taxing cognitive skill, and that people who outsource execution to AI while digging deep into real problems themselves will pull ahead of those who outsource their thinking.
Topics
View original postX product head Peter truly deserves to be the world's top product manager right now; you can always trust Musk's eye for talent. In this interview of just over a minute, Peter pierced through the painful truth that all large-model companies know but dare not say publicly: Putting ads in AI chat products is entirely feasible, but it can kill the company's entire life at any moment. You need to know this is not another banner ad hung on the sidebar of a page, nor is it a sponsored tweet in a feed that you just scroll past. This is the first time in the history of the human internet that commercial sponsorship has been stuffed directly into a one-on-one chat window that users treat as a truth machine. Previously, everyone tolerated the rogue logic of traditional search engines and e-commerce platforms: Open Amazon and search for a product, and the first screen is densely packed with a dozen bidding ads. Users may curse, but they know perfectly well which one is an ad and which is an organic result. You can steal my click, but you cannot steal my guard against the interface information. But the bottom line of the AI chat interface is completely different. Users are no longer manually filtering among tens of thousands of links; instead, they hand over their ultimate trust and decision-making power to this one line of natural language on the screen. When a person asks AI about an illness, legal risk, or which infant formula to choose for their child, if even a cent of the sponsor's will is mixed into the single answer behind it, the entire foundation of trust will instantly vanish into thin air. The cruelest part is the dilemma OpenAI and this batch of large-model companies now face. The compute bill burns like astronomical figures every day, and the $20 monthly subscription fee alone cannot plug the huge cash-flow hole; introducing commercial advertising is almost an inevitable fate. But as long as you dare let ads contaminate the objectivity of generated content, users will, within a second, treat this intelligence source they once worshipped as a god as a cheap door-to-door salesman. Here is an extremely cold, counterintuitive truth: Traditional search sells display slots; if AI chat sells ads, what it actually sells is the soul of the answer. When display slots are stuffed with ads, you can still pinch your nose and use them. But once an agent is suspected of taking dirty money, in the user's eyes even a single punctuation mark becomes a poisonous lie. Commercialization must be done, but trust is a one-time fragile item. In the future, whoever is first to make monetization work in the AI era will never be judged by how many sponsors they can stuff in, but by whether they dare, when commercial interests hold a knife to their neck, to firmly hold the red line of never tampering with the facts of an answer.
View original
X 产品负责人Peter不愧是目前世界上对顶级的产品经理，永远可以相信马斯克选人的眼光， 在这段 1分多钟的访谈里，Peter把所有大模型公司最痛但又最不敢公开说的大实话捅穿了： 在 AI 聊天产品里插广告完全行得通，但它随时会直接杀死这家公司的整条命。 要知道这不是又一次在页面边栏挂个横幅广告，也不是在信息流里那种划过去就算了的赞助推文， 这是人类互联网历史上，第一次把商业赞助直接塞进一个被用户当成真理机器的单聊窗口里。 以前大家容忍传统搜索引擎和电商平台的流氓逻辑： 打开亚马逊搜索一个商品，第一屏密密麻麻塞了十几个竞价广告，用户虽然骂娘，但心里明镜似的知道哪条是广告、哪条是自然结果。 你骗得走我的点击，但骗不走我对界面信息的防备心， 但 AI 聊天界面的底牌完全不同， 用户不再是在成千上万个链接里做人肉筛选，而是把终极的信任和决策权，双手奉还给了屏幕上的这一行自然语言。 当一个人向 AI 询问病情、法律风险、或者给孩子选什么奶粉时，这个唯一的答案背后，哪怕只混进了一分钱的金主意志，整个信任基座瞬间就会灰飞烟灭。 最狠的是 OpenAI 这一批大模型公司现在的进退两难， 算力账单每天都在天文数字般地狂烧，光靠每个月 20 刀的订阅费根本堵不住现金流的大窟窿，引入商业广告几乎是不可避免的宿命， 但你只要敢让广告去污染生成内容的客观性，用户就会在一秒钟之内把这个曾经奉若神明的智囊，当成上门推销的廉价销售。 这里有一个极度冷酷的反直觉真相： 传统搜索卖的是展示位，AI 聊天如果卖广告，卖掉的其实是答案的灵魂。 展示位被广告塞满了还可以捏着鼻子用， 但一个一旦被怀疑收了黑钱的智能体，在用户眼里，连一句标点符号都会变成有毒的谎言。 商业化必须要做，但信任是一次性的易碎品， 未来谁能第一个在 AI 时代跑通变现，从来不看他能塞进去多少赞助商， 而看他敢不敢在商业利益把刀架在脖子上时，死死守住那条绝不篡改答案事实的红线。
Body · AI translation
NVIDIA CEO and founder Jensen Huang publicly revealed one of his extremely counterintuitive daily habits in a CNN interview: He uses AI frantically every day, yet he feels his own brain has only gotten smarter.
The way many people use AI is essentially performing a massive mental amputation on themselves. For example, they have it polish an email, have it write a summary in their stead, and the moment they hit a little snag that requires some thinking, they casually shove the keyboard over to the model and let it make the decision for them.
Treating the tool as a crutch for laziness, the result is that after half a year, their skills are all ruined, and they have thoroughly degenerated into overgrown infants who cannot think without a chat box.
But in the interview, Huang directly tore off this fig leaf. He said he never lets AI think for him. He has AI serve as his private tutor, to teach him entirely new fields he knows nothing about, or to gnaw on the hard bones that sheer human experience simply cannot crack.
Don't use it to do things you could already do; use it to push open the cognitive walls you originally could never get past.
The most brutal part was the real bottom line about being a CEO that Huang mentioned when Zakaria pressed him.
He said that as the head of a chip empire worth trillions of dollars, 90% of his instructions every day are essentially not giving orders, but frantically posing high-quality questions to his team and to the models.
Writing prompts for large models has never been some mechanical chore of typing in a few keywords; asking questions is itself a top-tier cognitive skill that consumes an extreme amount of brainpower.
Here is a cruel, counterintuitive truth: AI is widening the wealth gap between two kinds of people at the speed of light.
One kind of person outsources thinking to the machine, saves half an hour of slacking off every day, and sees their cognition shrink across the board.
The other kind of person outsources execution to the machine, forces themselves to dig deep into real problems, and uses the world's top-tier computing power every day to complete a remodeling of their own brain.
So let's stop asking whether AI will make people stupid. Tools have no moral preferences. If you treat it as a wheelchair, you will become paralyzed; if you treat it as a lever, you alone can move an entire brain trust.