The design learns by having a piece of textual content from the information (say, the opening sentence of the Wikipedia posting) and trying to forecast the subsequent token from the sequence. It then compares its output with the actual text within the instruction corpus and adjusts its parameters to