Currently, the Proofreading LLM output shows clear patterns of LLM writing. Most notably, the LLM loves, loves, LOVES em dashes, which, despite their clear grammatical utility, have become a notorious marker of LLM writing. I find myself having to, ironically, proofread the output of the Proofread tool to ensure it doesn't LLMify my text in undesirable ways. It has gotten to the point where I put "Do not use em dashes" into the context field every time I use the tool.
The tool should not make changes or suggestions that increase the perceived likelihood of the text being deemed AI writing by outside observers.
One potential way to fix this is to include the contents of the Signs of AI writing page from Wikipedia in the system prompt.
