Google DeepMind CEO Demis Hassabis on why AI can win Maths Olympiad but fails to solve basic problems of High School Maths

Google DeepMind

CEO

Demis Hassabis

has said that inconsistency in artificial intelligence (AI) is a major reason why AI can perform exceptionally well in some areas but still fail at simpler tasks. Speaking on the “Google for Developers” podcast, Hassabis said that advanced AI models like Google’s Gemini can win gold medals at the International Mathematical Olympiad but often struggle to solve basic high school maths problems.“The lack of consistency in AI is a major barrier to achieving AGI,” he said, referring to

Artificial General Intelligence

— the stage where AI can reason like humans. This, according to Demis Hassabis, is what is holding back the technology from reaching its full potential.During the interaction, Hassabis also referred to Google CEO Sundar Pichai’s description of the current state of AI as “AJI” — artificial jagged intelligence — a term used for describing systems that excel in certain tasks but fail in others.The DeepMind CEO also stressed that solving AI’s inconsistency problem will take more than just increasing data and computing power. “We need better testing and new, more challenging benchmarks to determine precisely what the models excel at and what they don’t,” Demis Hassabis said.

The debate over AGI continues to divide the tech industry. Hassabis has previously taken a more cautious view of its arrival compared to Google co-founder Sergey Brin, calling for higher standards before declaring that AI has reached that level.

OpenAI CEO takes a u-turn on AGI

OpenAI CEO Sam Altman who earlier suggested that the AI term – AGI aka Artificial General Intelligence is "just around the corner” has taken a u-turn. At CNBC’s “Squawk Box” last week, Altman was asked whether the company’s latest

GPT-5

model moves the world any closer to achieving AGI. Replying to the question, Sam Altman said “I think it’s not a super useful term”. He said the challenge with AGI is that different companies and people define it in different ways. One definition, he explained, is an AI that can do “a significant amount of the work in the world.” But this has problems, as the type of work people do keeps changing.“I think the point of all of this is it doesn’t really matter and it’s just this continuing exponential of model capability that we’ll rely on for more and more things,” he added.