Technically upscaling with machine learning can be amazing, and it's what most real-time renderers use nowadays. You won't be able to tell, and it can even add detail if the network is good. I'm not an expert though, but I suppose it can only get better?