ai-scaling-laws-amodei

Forecast AI capability trajectories by modeling pretraining versus RL scaling paths and applying a 7-month doubling heuristic. This helps product teams estimate feature timing and evaluate risk across evolving model generations.

The 7-month doubling heuristic is a forecasting rule used to estimate how quickly AI model capabilities double. It serves as a baseline input that you adjust using task-horizon estimation and a self-correction multiplier.

Apply scaling laws to differentiate phases by evaluating scaling behavior during pretraining separately from reinforcement learning. This two-phase distinction guides the timing of strategic decisions and feature releases.

Estimate task horizons by supplying your current model horizon, the expected doubling pace, and your desired target capability. The framework uses these inputs to generate a refined capability forecast.

Use AI scaling forecasts for strategic risk management by comparing capability predictions against scaling law expectations. This framework supports evaluating risks and planning product timing across upcoming model generations.