“Training compute” is constantly evolving, and compute-based AI policies must adapt to remain relevant
Another hit from Séb 👏
💜 And Venkat & Anson in particular!
Future algorithmic discoveries are only going to make compute thresholds (in their current form) less reliable as a proxy for measuring capabilities. Curious to see how regulators respond to this. Good work!
Another hit from Séb 👏
💜 And Venkat & Anson in particular!
Future algorithmic discoveries are only going to make compute thresholds (in their current form) less reliable as a proxy for measuring capabilities. Curious to see how regulators respond to this. Good work!