Microsoft CEO Satya Nadella may have just agreed with Google DeepMind CEO Demis Hassabis on ‘next AI breakthroughs’

Two of the most powerful figures in artificial intelligence (AI) industry have signalled the same thing: the current era of AI is not the final one, and that it will take one of two more breakthroughs that can change the world entirely. Both Microsoft CEO Satya Nadella and Google DeepMind CEO Demis Hassabis believe the industry is moving towards something new but neither is sure what form it will take. When asked whether meaningful breakthroughs still lie ahead beyond today's AI models, Nadella responded by saying that there isn’t going to be one model that will change how things are.





“It’s not no longer just one transformer model,” Nadella said in a response while talking on the OMR Podcast.

“If you look at where we’ve been. It was all about pre-training scale. Then it was about post- training, then we came up with reasoning, then we said, ‘Oh, there's (Reinforcement Learning) RL’. And so there's constant innovation that's happening even in what is considered transformers and transformer architectures,” he added.

But his most striking comment was about what might come next.

“I always say that we are one sort of innovation away from the entire regime changing,” Nadella said, pointing to the possibility of a fundamentally new model architecture that could be far more efficient than anything available today.

Google DeepMind CEO Demis Hassabis on AGI

His comments are similar to what Google DeepMind CEO Demis Hassabis has been saying for some time now. Speaking at the Axios AI+ Summit in San Francisco last year, Hassabis discussed his position on AI's current trajectory.

“The scaling of the current systems, we must push that to the maximum because at the minimum, it will be a key component of the final AGI system. It could be the entirety of the AGI system (a hypothetical form of AI capable of reasoning and problem-solving at a human level),” he told the audience.

The theory behind scaling is straightforward: the more data and computing power you feed an AI model, the smarter it becomes.

But Hassabis also discussed the limitations, saying that publicly available training data is finite due to which despite building larger data centres and making sophisticated models, the returns are diminishing. Hassabis also believes AI requires “one or two” additional major breakthroughs beyond scaling alone — potentially in areas like reasoning, memory and what he calls “world model” ideas.

“I still think there'll be one or two more things that are required to really get across the board that you’d expect from [artificial] general intelligence and also the improvement on reasoning and memory,” he said. Hassabis reiterated that true AGI is still five to ten years away. He previously put the odds of reaching AGI by 2030 at around 50%.