Yin & Wang reveal universal patterns of progress from 6.8M solutions

In a landmark study published on arXiv, Yian Yin and Dashun Wang set out to understand the seemingly chaotic nature of scientific and technological breakthroughs. They assembled an unprecedented dataset: 6.8 million solutions to 6,700 tasks spanning materials science, structural biology, AI, computational biomedicine, data science, theoretical computer science, Formula-1 racing, and even physical wheel building. Across these diverse domains, they discovered three universal patterns. First, the waiting times between new frontier records are heavy-tailed—meaning most effort yields no breakthrough, but when one comes, it is often followed by a flurry. Second, frontier records accumulate at a sublinear rate: faster than logarithmic but slower than linear growth, indicating diminishing returns from adding more resources. Third, record-breaking events are temporally correlated, enabling short-term predictability (e.g., knowing a breakthrough is due) but long-term unpredictability (the specific timing of the next big leap remains chance).

The researchers found that existing models from complex systems, record statistics, economics of innovation, and cultural evolution all failed to reproduce these patterns. The missing ingredient: the distinction between radical and incremental innovation. Yin and Wang built a minimal, analytically solvable model that includes both radical resets (which restructure what is achievable) and incremental refinements (which exploit the current frontier). Remarkably, the leading-order predictions are parameter-independent, identifying a new universality class. The model also yields testable predictions about how openness and access to frontier solutions shape the pace of advance. This work suggests that the interplay between radical and incremental progress is a fundamental driver—not just a metaphor—and could inform how we fund research, organize teams, or design AI systems to accelerate discovery.