IRL 2/8: Mitigating degeneracy: multiple experimentation — LessWrong