Demis Hassabis, CEO of Google Deepmind, dismisses the claim that today's AI systems are PhD Intelligences, calling the labels nonsense, claiming that the current model lacks the consistency and reasoning needed for true general intelligence.
“They're not PhD intelligence,” Hassavis said in a recent All-in-Podcast interview. “They have some PhD-level features, but they are generally not competent, and that's exactly what general intelligence should be.
Hassabis' statement came after Openai called its latest AI model, the GPT-5, PHD level.
Hassavis explained that advanced language models can demonstrate impressive skills, but simple problems can fail. “As we all know, as we interact with today's chatbots, raising questions in a certain way can make simple mistakes even in high school maths and simple counting. That's not possible with a true AGI system,” he said.
Deepmind's chief said that artificial general information (AGI) is still five to ten years away, and points out that it lacks features such as continuous learning and intuitive reasoning. “Inconsistent,” he said. “One of the things that separate great scientists from great scientists is creativity: the ability to find patterns across the subject area. One day, AI may be able to do this, but there is no reason to do that for such a breakthrough.”
On industry benchmarks, Hassavis opposed the idea of stagnant performance. “We haven't seen it internally. We're still seeing great progress,” he said, rebutting reports suggesting convergence and slowing improvements between large-scale language models.
Hassavis said scaling may bring about progress, but one or two breakthroughs will be needed over the next few years.
