Would training LLMs on either smaller batches of text first instead of the (current) large batches of text or on easier subjects/topics first then later train them on subjects/topics with increasing order of difficulty...enable the LLMs to learn better. I ask this because the human mind learns/is taught this way. Also would doing rules-based unsupervised learning enable more efficient learning by the LLMs? By rules I mean teaching LLMs things like grammar and conversation rules first and then doing unsupervised learning?