That is the beauty of the machine learning paradigm. Whatever we player do to create an advantage/exploit, Sophy will continuously observe, learn and try new things to gain back that advantage -- much like a human.
Like I said earlier, the really interesting part is when you couple this learning behavior with Sophy's propensity to make mistakes, then the racing may take on another dimension.
I think the concept of a player finding a clever exploit to out smart an AI opponent is surely becoming a relic of the past. While this may be absolutely true for all the AI opponents in past games regardless of genre, we are witnessing the beginning of an intelligent agent/opponent that has the capabilities to adapt to our strategies and tactics.
Players' like myself will be thoroughly engrosses in such a scenario, trying to outwit Sophy again and again...but I know my clever methods will be ephemeral...and the challenge of coming up with new ways is where all the fun is for this human player.