Alphabet and Google CEO Sundar Pichai told Semaphore in an exclusive interview that the companies are ready to work on an AI “Manhattan Project” when Donald Trump takes over the White House next year. .
“I think we have an opportunity to work together as a country,” he told Semaphore earlier this week. “These large-scale, physical infrastructure projects that accelerate progress are extremely exciting to us.”
According to Pichai’s mood meter, the intellectual boss known for his Zen-like calm was on Wednesday afternoon after a day of major product announcements following his Nobel Prize win for major advances and achievements in quantum computing. I was totally dazzled by the company’s Mountain View campus. Within Google. He was scheduled to leave the next day to meet President-elect Donald Trump, The Information reported, and Semaphore confirmed.
The head of the $2.4 trillion company appeared to be enjoying a moment of relief after years of intense pressure following the launch of ChatGPT in 2022. Less than a year ago, the conventional wisdom was that Mr. Pichai would not be in office for long. . On Wednesday, he told me he had no doubts about the company’s direction.
“Internally, there was a clear sense that we were making progress,” he said. “It’s definitely very satisfying to see the momentum, but we’re going to do a lot more. We’re just getting started.”
At the core of that innovation is Gemini, the company’s flagship frontier AI model. Google took a different approach than its competitors, focusing its research energy and efforts on building “multimodal” models from scratch rather than large-scale text-based language models.
When Gemini launched a year ago, it didn’t fare very well against its competitors on some key benchmarks, such as coding. However, Gemini 2.0, released on Wednesday, appears to have overtaken Anthropic and moved to the top of SWE-Bench, a key indicator of AI capabilities.
This metric shows that Google is on the right track. And for Pichai, it goes back to a decision he made nearly a decade ago when he became CEO of the search giant.
“In 2015, I took the company in an AI-first direction. As part of that, we built a complete full-stack technology from silicon to AI, from world-class research to building infrastructure. “That’s the basis.”
When ChatGPT arrived (which Pichai calls the “current-generation AI moment”), he first reinvented the company by ultimately consolidating most of its AI forces into one organization, Google DeepMind. He said he decided to invest upfront in construction. We take cutting-edge research and turn it into consumer products.
While that may have made it seem like the company was slow to embrace OpenAI, Pichai said he is playing the long game. “In this age of AI, sometimes you make investments to get things up front. For me, that was setting up Google DeepMind from scratch,” he said.
He added that Gemini 2.0 has finally reached a level of capability that will enable a wave of new consumer products. “We already have a fully-featured model. We can build so many use cases on top of it. The progress is going to be very real,” he said. . “With Gemini 2.0, we are laying the foundations for becoming more agentic.”
Agent AI refers to the ability of models to perform actions on behalf of people. This week, Google showed off a new tool called Project Mariner. Although not yet publicly available, Mariner can take control of your web browser and follow your instructions. For example, “Please fill out the expense report.”
“It would be really great to see models being able to use a browser, but there are some barriers that need to be broken through,” Pichai said. “They say 80% of the effort is in the last 20%. In this case, 90% of the effort can be in the last 10%.”
In addition to breakthroughs in AI research, a small number of companies in the AI race today are fighting for three key resources: computing power, energy, and data. Pichai said Google’s data centers are as powerful, if not more powerful, than those of its competitors. “We’re at the cutting edge and we’re scaling it up. And everything that I see, everything that we benchmark, I think we’re at the frontier there as well,” he said. spoke.
And he said the company is making progress in energy, signing small modular nuclear contracts, exploring geothermal energy and considering a significant increase in solar power. “I’ve always felt that if we really think about it, we should be dealing with an energy surplus,” he says. “Energy should be an accelerator, not a constraint. The only thing standing in our way is our imagination and determination.”
While the company’s AI research is at the core of its product development strategy, its quantum computing division is busy in a quiet satellite office in Santa Barbara, California.
Thanks to a combination of theoretical physics and engineering challenges, the company was able to break through a long-standing barrier: the fact that quantum computers become more prone to errors as they get bigger.
Pichai now said he expects quantum technology could make a meaningful contribution to Google within five years. “For me, Quantum is like the AI of the 2010s. Very few people know about it, but you are working on it methodically,” he said. “This is definitely one of the most positive surprises and a deeper breakthrough addressing error correction during scale-up of quantum computers, which is arguably the most difficult challenge in the field. It was one of the.
“We published state-of-the-art weather prediction models at GenCast, but in the future quantum computing will become available and will underestimate our ability to predict these things at a deeper and better scale. It should not be evaluated,” he added. “These have serious implications.”
Read the rest of our conversation with Pichai here.
