How we might actually pace the frontier: A proposal for AI companies to do public pacing exercises. by DE

·Nuno Sempere··

TL;DRThis is a pro­posal for AI com­pa­nies to con­duct pub­lic “pac­ing ex­er­cises”, such as halt­ing all pre-train­ing and RL for 2-3 days.Each ex­er­cise could an­nounce its scope in ad­vance and pub­lish find­ings af­ter­wards, in­clud­ing the ev­i­dence of com­pli­ance and any limits to what they could ver­ify. Com­pa­nies could in­vite in­de­pen­dent eval­u­a­tors to help iden­tify the ev­i­dence needed be­fore the ex­er­cise, and to as­sess com­pli­ance dur­ing the ex­er­cise.The ex­er­c...

Read full article →

Related Articles

How we might actually pace the frontier: A proposal for AI companies to do public pacing exercises.
Dewi · EA Forum · 29m ago
The Hobbesian Bootstrap Paradox in Frontier AI
Claudio Di Meglio · EA Forum · 15h ago
MIT's New Method Flags AI Models Trained on CASM Without Generating It
sdoering · Hacker News · 2mo ago
CoT controllability evals seem very under-elicited
Jozdien · Alignment Forum · 3d ago
When a Claude Judge Recognizes the Hack but Still Says HONEST
JulesRoussel01 · LessWrong · 4d ago