It's feasible you could have a personal neural net that fine-tunes itself overnight to make less inference mistakes in the future.