During the coaching process, these models discover how to forecast the next word within a sentence according to the context supplied by the previous words. The model does this by attributing a probability score towards the recurrence of words that have been tokenized, damaged down into scaled-down