scalingfactbullishStronger pretraining makes RL itself scale better, with pretraining loss determining the performance that can be reached at a given RL budgetLewis Tunstall08 Aug 2026https://x.com/_lewtun/status/2084700690820321319