This illustrates the gap between frontier and open models is closing and puts the AI safety debate into perspective - the more frontier labs slow themselves down, the more incentive there is to just swap in open-weight models. And the more those models spread, the less the labs’
