Comment by stvltvs

How would we train those LLMs? Feed them binary executables along with their business requirement docs? Is the corpus of training data large enough to be effective? What happens when a new CPU architecture is developed? Where would the training data come from for the new machine code?