> is new datasets + post-training shaping the model's behavior (instruction + preference tuning). There is no moat besides that. sure, but acquiring/generatin

> is new datasets + post-training shaping the model's behavior (instruction + preference tuning). There is no moat besides that. sure, but acquiring/generating/creating/curating so much high quality data is still significant moat.