Nvidia CEO Jensen Huang stated in an interview that current AI does not provide the most reliable answers, emphasizing the need for greater computational power to build trust in AI responses. "Today, the answers that we have aren't the best that we can provide," Huang explained. He noted that as AI continues to develop, it should move toward offering answers that users can trust without doubt, but that progress is still several years away.
In his remarks, Huang discussed limitations in today's large language models, claiming they need robust improvement regarding accuracy and reliability. He highlighted how hallucinations remain a major concern, and that users shouldn't feel unsure about whether AI outputs are logical or factual. "We have to get to a point where the answer that you get - you largely trust," he emphasized.
Huang also pointed to the broader challenge facing AI companies, who must advance technology without becoming overly dependent on the constantly shrinking pool of available data. He argued that simply pre-training models on large datasets will not suffice, suggesting a more nuanced approach is required to actually increase AI effectiveness and reliability going forward.
Despite the rapid advancements in AI over the past few years, Huang insisted that current systems still fall short of delivering quality responses, reiterating that the industry must keep enhancing computational capabilities. "I think that we're several years away from being able to do that and, in the meantime, we have to keep increasing our computation," he concluded.
Collection
[
|
...
]