this is completely wrong.. the entire LLM system was bootstrapped by "self-supervised learning" .. where data sets are divided and then training proceeds on par

this is completely wrong.. the entire LLM system was bootstrapped by "self-supervised learning" .. where data sets are divided and then training proceeds on parts.. it is literally self-training