Just writing off AI upscaling completely is bs. It's not some magic bullet to use on every video and there is a learning curve on how to apply but there are absolutely scenarios where you can get shockingly good results. I think a lot of people make judgements on it based on super small sample sizes.
On a separate note also not mentioned llm's are really good at generating ffmpeg commands. Just discuss with chatGPT your source file and goals for a video and you can typically oneshot a targeted command even if you aren't familiar with ffmpeg cli.
AI upscaling can also be done during playback, to the benefit of lower file sizes, but at the cost of higher processor utilisation. So it is a trade-off.
On a separate note also not mentioned llm's are really good at generating ffmpeg commands. Just discuss with chatGPT your source file and goals for a video and you can typically oneshot a targeted command even if you aren't familiar with ffmpeg cli.