The essay discusses the limitations of using compute thresholds as a governance strategy for managing risks in Generative AI. It argues that current compute thresholds are shortsighted and likely to fail in mitigating risk due to the uncertain and rapidly changing relationship between compute and risk. The essay highlights that while larger models may amplify certain risks, smaller models are increasingly performing better and can outperform larger ones. It also points out that compute thresholds fail to account for optimization techniques, data quality, and architectural improvements that significantly impact model performance. The essay recommends a more nuanced approach to risk assessment, including dynamic thresholds and a combination of performance metrics rather than relying solely on compute. It also emphasizes the need for clear guidelines on how to measure compute and the importance of transparency in governance decisions. The essay concludes that compute thresholds are not a reliable indicator of risk and that a more comprehensive approach is needed to effectively manage the risks associated with Generative AI.The essay discusses the limitations of using compute thresholds as a governance strategy for managing risks in Generative AI. It argues that current compute thresholds are shortsighted and likely to fail in mitigating risk due to the uncertain and rapidly changing relationship between compute and risk. The essay highlights that while larger models may amplify certain risks, smaller models are increasingly performing better and can outperform larger ones. It also points out that compute thresholds fail to account for optimization techniques, data quality, and architectural improvements that significantly impact model performance. The essay recommends a more nuanced approach to risk assessment, including dynamic thresholds and a combination of performance metrics rather than relying solely on compute. It also emphasizes the need for clear guidelines on how to measure compute and the importance of transparency in governance decisions. The essay concludes that compute thresholds are not a reliable indicator of risk and that a more comprehensive approach is needed to effectively manage the risks associated with Generative AI.