●researcherConsider the limitations of current benchmark trends when forecasting model capabilities.
●investorUse these findings to question the reliability of current capability-based valuations for frontier models.
●policyThis provides evidence that current metrics for measuring AI progress may be insufficient for governance.