The article explores the evolving definition of human-level Artificial General Intelligence (AGI) as advancements in AI technology continue to progress.

A proposed definition suggests that a system achieves human-level AGI when it becomes impossible for humans to create any benchmark where they can reliably outperform the AI.

Data observations indicate that it is increasingly challenging to devise tests where humans can outperform advanced AI systems.

This trend suggests that by around 2025, it may be impossible to create any task where humans perform better than AI, marking a pivotal milestone in AI development.

The definition of human-level AGI is empirical, focusing on test outcomes rather than attempting to define human intelligence, which is subjective and complex.

To track AI benchmarks and the timeline for AI systems reaching human-level performance, the author, Rasmus Lindahl, has launched a website called 'h-matched'.

The author invites readers to share their thoughts on this definition and its implications for the future of AI and human intelligence.