Dice Question Streamline Icon: https://streamlinehq.com

Predicting potential plateau in AI agent time horizon growth

Determine whether the exponential growth in AI agent task-completion time horizon will plateau; if so, identify the timing and duration of any plateau.

Information Square Streamline Icon: https://streamlinehq.com

Background

In examining alternative functional forms for horizon growth, the paper notes that early segments of saturating logistic curves resemble exponentials. Given limited data and no present evidence of leveling-off, the authors argue that plateau timing and duration cannot currently be predicted.

Establishing whether and when horizon growth will saturate is essential for credible long-term capability forecasting and risk assessments.

References

Likewise, the initial part of any saturating logistic function $\mathrm{horizon} \sim a \cdot \sigma(\mathrm{release_date} + d) $ looks very similar to an exponential, and without any evidence that AI horizon is leveling off (on our metric), it is essentially impossible to predict when or if it will plateau, or for how long.

— Measuring AI Ability to Complete Long Tasks (2503.14499 - Kwa et al., 18 Mar 2025) in Appendix, Section “Alternative curve fits”