It turns out that NVIDIA has offered everyone a discount of a million times. Jensen Huang shared when we can expect trustworthy AI, the company’s role in this process, and how he convinced his wife to marry him.
When Will AI Stop Lying
Jensen Huang believes that the AI industry will struggle with the issue of "hallucinations"—when AI generates information to fill in knowledge gaps—for several more years. He discussed this and other topics during an interview at the Hong Kong University of Science and Technology.
Huang outlined the directions of AI development. The first, pre-training, he likened to college admission, where AI simply gathers data and transforms it into responses. "This is a very important stage, but it's not enough." The second is post-training, involving a deep dive into a specific skill through "reinforcement learning via human feedback, feedback from AI, synthetic data generation, multi-modal learning, etc.—there are many ways to go about it."
The third direction of AI development is real-time testing, which Jensen Huang also refers to as "thinking." This is significantly more complex because AI now breaks down problems step by step to find a solution. It may require repetitions and sometimes simulating different outcomes, especially if the answer is not predictable. "We call this thinking, and the longer you think, the better the answer might be." But for now, despite all these technologies, Huang notes that we still cannot fully trust AI's responses.
“Today we have the best answers we can achieve, but we are [still] approaching the point where the answer will not be the best we can provide, and you will still have to decide whether it’s a 'hallucination' or if it makes sense, whether it’s reasonable or not? We need to reach a stage where you can substantially trust the answer provided. … I think we can achieve this in a few years, and in the meantime, we need to keep increasing [our computing] power.”
A Huge Discount
Jensen Huang emphasized NVIDIA's role in AI development. He pointed out that without the company's efforts, AI processors would be a million times more expensive:
“We have taken the marginal cost of computation and reduced it by a million times. … When the cost of something drops by a million times, your habits change fundamentally… This is the greatest contribution NVIDIA has ever made—to ensure that the use of machines for in-depth study of vast amounts of data is something that researchers need not even think about. That’s how machine learning became popular.”
The Pickup Line
During the event, Jensen Huang was asked for advice for young entrepreneurs on when it's best to start their own business. The interviewer added, “Additionally, you promised your wife Lori that by age 30, you would start a company,” to which Huang responded, “That was my pickup line. I didn’t mean it literally.”
“To make sure she would eventually marry me, I told her that by the time I turned 30… I would be CEO. I had no idea what I was talking about—and then we got married. That’s all the advice I’m going to give to entrepreneurs. That’s it.”
Source: Tom's Hardware
Comments (0)
There are no comments for now