Nathan Lambert

@ChorzempaMartin @EpochAIResearch @Afinetheorem This isn't taking into account the massive error bars on measuring capabilities (and cost per task).