Maximizing impact at startup with high EA potential? — LessWrong