AI at a PhD level? “That’s nonsense”

Algorithms solve complex mathematical problems, generate academic papers and formulate hypotheses. This has led some people to eagerly label them as “PhD level”. One of the world’s leading AI developers, Nobel Prize winner and head of DeepMind, Demis Hassabis, strongly rejects such claims. “It’s nonsense,” he believes.

Redakcja

Hassabis criticizes comparing AI to a person with a PhD primarily for two reasons. First, such a solution must operate consistently at the same high level: it can’t ace one test brilliantly and then make trivial mistakes in another, as is currently the case. Second, the way a question is posed should not determine the outcome — today’s LLMs are susceptible to “magic phrases” and hallucinate, making them impressive tools, but not so reliable. That’s why Hassabis calls the label “PhD intelligences” nonsense.

As explained in the discussion in the All-in podcast, yes, AI does have abilities similar to a doctor in selected areas, but it still lacks coherence, intuition and creativity. A PhD is a journey, ethos and years of struggle, not just a sum of correct answers – he notes.