Epistemic Strategies of Safety-Capabilities Tradeoffs — LessWrong