Companies like Open AI and Google are running out of the data used to train artificial intelligence systems. Can new methods continue years of rapid progress?

Demis Hassabis, one of the most influential artificial intelligence experts in the world, has a warning for the rest of the tech industry: Don’t expect chatbots to continue to improve as quickly as they have over the last few years.

A.I. researchers have for some time been relying on a fairly simple concept to improve their systems: the more data culled from the internet that they pumped into large language models — the technology behind chatbots — the better those systems performed.

But Dr. Hassabis, who oversees Google DeepMind, the company’s primary A.I. lab, now says that method is running out of steam simply because tech companies are running out of data.

Interviews with 20 executives and researchers showed a widespread belief that the tech industry is running into a problem many would have thought was unthinkable just a few years ago: They have used up most of the digital text available on the internet.

Not everyone in the A.I. world is concerned. Some, like OpenAI’s chief executive, Sam Altman, say that progress will continue at the same pace, albeit with some twists on old techniques. Dario Amodei, the chief executive of the A.I. start-up Anthropic, and Jensen Huang, Nvidia’s chief executive, are also bullish.

Researchers called Dr. Kaplan’s findings published in 2020 “the Scaling Laws.” Just as students learn more by reading more books, A.I. systems improved as they ingested increasingly large amounts of digital text culled from the internet, including news articles, chat logs and computer programs.

The problem is, neither the Scaling Laws nor Moore’s Law are immutable laws of nature. They’re simply smart observations. One held up for decades. The others may have a much shorter shelf life. Google and Dr. Kaplan’s new employer, Anthropic, cannot just throw more text at their A.I. systems because there is little text left to throw.

Dr. Hassabis said that existing techniques would continue to improve A.I. in some ways. But he said he believed that entirely new ideas were needed to reach the goal that Google and many others were chasing: a machine that could match the power of the human brain.

Ilya Sutskever, who was instrumental in pushing the industry to think big as a researcher at both Google and OpenAI before leaving OpenAI to create a new start-up this spring, made the same point during a speech last week. “We’ve achieved peak data, and there’ll be no more,” he said. “We have to deal with the data that we have. There’s only one internet.”

Dr. Hassabis and others are exploring a different approach. They are developing ways for large language models to learn from their own trial and error. By working through various math problems, for instance, language models can learn which methods lead to the right answer and which do not. In essence, the models train on data that they themselves generate. Researchers call this “synthetic data.”

OpenAI recently released a new system called OpenAI o1 that was built this way. But the method only works in areas like math and computing programming, where there is a firm distinction between right and wrong. “These methods only work in areas where things are empirically true, like math and science, the humanities and the arts, moral and philosophical problems are much more difficult.” said Dylan Patel, chief analyst for the research firm SemiAnalysis.

During a call with analysts last month, Mr. Huang, Nvidia’s chief executive, was asked how the company was helping customers work through a potential slowdown and what the repercussions might be for its business. He said that evidence showed there were still gains being made, but that businesses were also testing new processes and techniques on A.I. chips.

2024.12.19 New York Times