The Digital Press

A new paper proposes a concrete, quantifiable definition of Artificial General Intelligence (AGI) as matching the cognitive versatility and proficiency of a well-educated adult, using established human cognitive models to benchmark AI systems. It reveals current AI models like GPT-4 and GPT-5 scoring 27% and 57% respectively, indicating significant progress but also notable gaps, especially in long-term memory.

Why it matters: Provides a standardized, measurable framework to assess AI progress toward human-level general intelligence.

The big picture: The framework breaks intelligence into ten cognitive domains, adapting human psychometrics to evaluate AI comprehensively.

Stunning stat: GPT-5 achieves 57% AGI score, showing rapid improvements but still far from full human cognitive ability.

Commenters say: They appreciate the thoughtful approach but debate the practicality and shifting goalposts of defining AGI, noting real-world complexity and ongoing hype around the term.