The way the ad is presented makes it look like there's something wrong with the original (❌) and that the mangled version is better (✅), as if it was actually improved.

The tool removed all the subtext from the original by using this very neutral, matter-of-fact language. There is actual information lost there, not just rigmarole. And that's the example they chose to put into the ad.

LLMs will also make shit up or completely misinterpret what's being written, I wouldn't trust it to get through an entire book without grossly misleading the reader or flipping out. They can't parse that much text at once right now so all interpretation of a chunk of text will have only a very broad, short and possibly wrong/irrelevant summary of what came before for context.

I don't even want to know what this would do to something like a Pratchett novel or a textbook.

As far as accessibility tools using machine learning go, wouldn't a text-to-speech reader app be better for dyslexics anyway?