From Math Olympiads to Navier-Stokes: How Fast Is AI Progressing? with Greg Burnham - #778

AI capabilities are evolving rapidly, shifting from academic exam-style benchmarks to complex, real-world research challenges. Greg Burnham, who leads capabilities research at Epoch AI, tracks this trajectory using unified benchmarking methodologies that reveal a steady, linear progression across model generations. While AI systems have achieved significant milestones—such as solving long-standing mathematical problems like the Navier-Stokes equations—they still struggle with tasks requiring deep, original creative insight or on-the-fly learning. Current research focuses on identifying potential "tripwires" or discontinuities that could signal a shift toward recursive self-improvement or superhuman performance. By analyzing how models handle intricate computations versus novel problem-solving, researchers aim to better predict the impact of AI on scientific and economic progress while navigating the inherent challenges of measuring capabilities that lack clear, pre-defined solutions.