news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Zhitong Finance APP learned that Daron Acemoglu, a professor at the Massachusetts Institute of Technology (MIT), said that in the next decade, only a quarter of artificial intelligence-related tasks can be cost-effectively automated.

Even if there are major breakthroughs in artificial intelligence, the impact will take years to be felt, Acemoglu said on the Goldman Sachs Exchange Podcast.

Goldman Sachs said this means that over the next 10 years, AI will have an impact on less than 5% of all tasks, will only increase U.S. productivity by 0.5%, and will contribute only 0.9% cumulatively to GDP growth.

“It turns out that current large language model (LLM) architectures are more impressive than many people predicted, but I think it takes a lot of confidence that we can get something as smart as HAL from 2001: A Space Odyssey just based on this architecture predicting the next word,” he said.

“Our current LLM architecture can be very limiting,” Acemoglu said.

He also doubts that AI can get to where it needs to be faster by simply throwing more GPU capacity at it.

He added that increasingly high-quality data, rather than volume, would be needed, and it was unclear where such data would come from.

Jim Covello, head of global equity research at Goldman Sachs, said that in order to get returns from companies such as Nvidia (NVDA.US), Microsoft (MSFT.US), Google (GOOGL.US), Meta (META.US), Amazon (AMZN.US) and Advanced Micro Devices (SMCI.US) that are expected to invest in AI capital expenditures in the coming years, AI must be able to solve complex problems.

“We’ve been working on this for a few years, and at this point, nothing has been cost-effective,” he said. “I think there’s an incredible misunderstanding of what this technology can do. The problems it can solve are not big problems. There’s no cognitive reasoning in this.”