In an interview that Huang gave in connection with being recently appointed honorary doctor at a technical university in Hong Kong, he says about the answers that AI services provide:
Today, the answers are the best we can get.
Huang means, however, that it is not enough. For users must still decide whether they fully trust the answer given.
Is this hallucinated or not? Does it seem reasonable? Is it reasonable or not reasonable? says Huang and continues:
We must reach a point where you trust the answer you get.
However, it will take time, according to the Nvidia CEO:
I believe we are several years away from being able to do that, and until then, we must continue to increase our computing power.
American Nvidia is the world-leading manufacturer of graphics cards and processors used in AI and machine learning. Things that cost a lot of money, but which according to Huang are still "cheap".
We have taken the cost of calculation and lowered it millions of times. Then behaviors change. It's the biggest thing Nvidia has done, we've made it so that machine learning with enormous amounts of data is something that researchers don't even need to consider whether it's worth it or not.
In a follow-up question where Nvidia's graphics processors were described as expensive, Huang responds:
I gave you a million-fold discount over the past ten years. They are practically free!
