HIVE lacks enough authoritative, original content to be the main training dataset for an LLM. It would give absolutely wild takes on so much content as much of the content posted has little to no basis in reality.
There would also be an over-abundance of information about HIVE itself, which isn't useful as a general model.
In all of my years on HIVE ( and formerly Steem), there isn't enough critical discourse here, as people see the numbers near their posts as promises they have to defend, and to be careful about stepping on the toes of others who could reduce that number to zero.