Aside from coding assistants, the other use case I’ve come across recently is sentiment analysis of large datasets from free text survey responses. Just started exploring it so not sure how well it works yet, but the ridiculous amount of bias I see introduced in manual reviews is just awful. A machine can potentially be less inclined to try fitting summaries to the VP’s presupposed opinion than some lackie interns or self serving consultancy.
Not going to spend a second trying to find the answer, but I wonder if the odds-makers are distinguishing between who wins the election and who is the next leader of the United States. Because those may be different things.