Not sure the logic follows, if they only just trained and improved the ML model THIS year, then the fact that the new model will work as well on last year's chip is immaterial.