In the long run, the chances for AI agents shall be “gigantic,” says NVIDIA Founder and CEO Jensen Huang.
Already, the progress is “spectacular and surprising” as AI development accelerates and the industry enters the “flywheel zone” that the technology must advance, Huang said in a hearth chat at Salesforce's flagship event. Dream Force this week.
“This is a rare time,” Huang said on stage with Marc Benioff. Salesforce Chairman, CEO and co-founder. “At no time in history has technology advanced faster than Moore's Law. We are advancing much faster than Moore's Law, the truth is, one could argue that we’re Moore's Law squared.”
Agents who work with other agents, “work with us”
In the long run, Huang said, there shall be AI agents that may understand subtleties, think logically and collaborate. They will have the ability to seek out other agents to “collaborate and partner with” while also talking to humans and getting feedback to enhance their dialogue and outputs. Some will “excel” in specific areas, while others can have more general skills, Huang said.
“We're going to have agents working with agents and agents working with us,” Huang said. “We're going to place the everlasting sunlight hours of our company into overdrive. We're going to return to work and get numerous work done that we didn't even know needed to be done.”
He and Benioff agreed that the onboarding process needed to be demystified, and Huang noted that “it should be far more like worker onboarding.”
For his part, Benioff stressed the importance of individuals “really understanding” how they work and what their purpose is, and that they “have to get their hands dirty with the thing.”
“Developing an agent mustn’t be a project for a pc science fair,” he said.
Still, Huang identified that we’ve got “many” challenges ahead. Some of those involve fine-tuning and isolation, but scientists are making progress in these areas on daily basis. In an interesting feedback loop, AI is getting used to curate data and create a secure curriculum for teaching AI.
“Now it's about considering whether the reply I generated is sufficiently secure and appropriate and whether it’s the very best possible answer I can provide,” Huang explained.
Nvidia has “done just a few things right”
Early on, Huang explained, Nvidia realized that general-purpose computers were good for some things but not others, and that there have been also “interesting problems” to be solved that required some expansion of computing power.
The company then focused heavily on accelerated computer architectures, enhanced CPUs with GPUs and built its DGX Platform“We knew that if we were going to be a computing platform, we needed to be architecturally compatible,” Huang said. “The company's policy was to pick problems that this computing architecture could solve.”
He noted that “every kind of complex algorithms” have been ported to Nvidia’s computing platform Cudaand the corporate began using deep learning. One of their first observations was that “deep learning was going to completely change the software,” Huang said. “We were convinced that this meant we needed to re-engineer each computing stack.”
Nvidia had the advantage, says Huang, of “working with every researcher on the planet.” They observed early scientific work (in 2011) to coach one in every of the primary major computer vision models.
“The breakthrough got here once we realized that unsupervised learning could be possible,” he said.
Ultimately, humans are limiting digital AI because we’re unable to call at scale, he stressed. Instead, scientists are using language models to create other language models using multimodal data. This feedback loop is advancing at an “incredible speed.”
“We knew all along today was coming,” he joked, joking that “we left it at that for today.” In reality, nevertheless, he acknowledged that “we did just a few things right.”
Benioff agreed, saying, “Never in my wildest dreams did I imagine (accelerated computing) could do what it could possibly today.”
What motivates Huang and Nvidia?
When asked about his personal motivation, Huang spoke of a palpable enthusiasm. “It's close by,” he said. “We can do that. We could make an actual difference.”
He added that he was “modest enough” and knew that he didn’t know every little thing; lifelong learning was essential.
“When you learn something, you get excited,” he said. “When you connect random ideas that no one knew may very well be related, you get excited.”
Nvidia and others would ultimately bring a level of automation capability the world has never seen before, he stressed, saying his company is in a “unique position in a lifetime and in a generation.”
He marveled: “It's just too exciting at once, don't you think that? Nobody should miss the subsequent decade. You won't wish to miss this film.”
