Part 4/10:
- 2022: GPT-4 levels up to advanced high school performance, excelling in complex coding and advanced mathematics.
Within merely four years, AI has progressed from a child's cognitive capacity to that of a gifted adolescent. The speed of this advancement fuels concerns about future capabilities, and the anticipated increases in AI’s proficiency show no signs of abating.
Benchmarking the Unthinkable
As benchmarks once regarded as testing the limits of AI become easily surmountable, it raises the question of how to evaluate AI competence effectively. High-stakes tests like Dan Hendris's “Humanity’s Last Exam,” designed with 3,000 complex questions, may soon be solvable by AI, crossing into territory only graspable by top-tier experts in various fields.