Predictability
Will capabilities improve smoothly enough for forecasts and evaluations to give advance warning, or emerge through strategically surprising jumps?
Loss curves are smooth, but downstream capabilities can look step-like. Whether emergence is real or an artifact of discontinuous metrics matters enormously for whether labs and governments get advance warning.
View on the map → · Open in Browse →
What changed
01
The trend lines forecasters rely on turned out to have a break in them: Epoch dated a roughly 90% acceleration in measured frontier capability to the arrival of reasoning models and RL scaling — a regime change that complicates the assumption that capabilities improve smoothly enough to give advance warning.
Recent thinking
Yafah Edelman & Jaeho Lee · Epoch AI · 23 Dec 2025 report
AI capabilities progress has sped upThe best score on the Epoch Capabilities Index grew almost twice as fast over the last two years as it did over the two years before that, with a 90% acceleration in April 2024.
Documents a regime change in the pace of measured frontier capabilities coinciding with reasoning models and RL scaling — evidence that even the trend lines used for forecasting are subject to sudden shifts, undermining advance-warning assumptions.
Additional relevant discussion (1)
Foundational reading (2)
Emergent Abilities of Large Language ModelsWei et al. · 2022Are Emergent Abilities a Mirage?Schaeffer et al., NeurIPS · 2023