@twizzay I'm mainly worried about LLM shenanigans. If they train some spicy autocorrect, i would expect deviously hard to detect errors instead of quirks that one can learn and deal with. Apart from the absurd cpu load that woud probably take.