As a reviewer and reader, I also feel quite frustrated about this. Using LLMs for (minor) presentation improvements is one thing, but, increasingly, the problem and solution seem to be mainly framed through LLMs, often in rather shallow and recurring ways.
Three papers in my POPL reviews and I have already seen so many occurrences of LLM-generated text that uses vague contrasts to explain the work, e.g., "Our framework is not X; but it is Y", "The problem is not syntactic; it is semantic", "The work is X rather than merely Y" 😢