Google’s CEO has warned that the era of ‘low-hanging’ AI development is over

The days of AI developers using large data sets scraped from the Internet and “low-hanging fruit” may be over, Google CEO Sanders Pichai suggested Wednesday at the New York Times' annual Delbook conference.

Pichai said, “In the current LL.M. In a generation of models, a few companies have risen to the top. “I think the growth is going to get stronger.”

Pichai's comments come as researchers say the performance of AI models has slowed compared to two years ago when ChatGPT went public.

Others who raise concerns about diminishing returns of AI models include Ethereum founder Vitalik Buterin, 16z's Marc Andreessen and Ben Horowitz, and former OpenAI co-founder and former chief scientist Ilya Sutskever, who have pointed out that AI models can learn large amounts if not unlabeled. Data, has reached a plateau.

“When I look at 2025, the low-hanging fruit is gone, the curve, the hill is steep,” Pichai added. I think that in 2025 there will be some notable teams, so I think it's an interesting year from that point of view.

Pichai's comments reflect a growing sentiment in the AI ​​world that the days of simply building bigger models and throwing more data at them to get better results may be over.

Adding to these concerns is the risk of the “AI Ouroboros” effect, where models are trained on data generated by other AIs rather than human-generated content.

Named after the ancient image of a snake eating its tail, the Ouroboros effect in AI occurs when one AI uses information generated by another AI, creating a feedback loop. Over time, as the system begins to rely on AI and not human-generated data, this can lead to repetitive or distorted results.

While some developers worry that AI models have reached a plateau, Pichai remains optimistic, noting that he will see significant advances in the coming year.

“I expect a lot of growth in 2025, so I don't fully subscribe to the wall idea,” Pichai said. “But when you start growing fast, you can throw more calculations, and you can make a lot of progress, but when we go to the next level, you definitely need deeper discoveries.”

Edited by Sebastian Sinclair