"In this paper, we show both theoretically and empirically, that these state-of-the-art detectors cannot reliably detect LLM outputs in practical scenarios." T

"In this paper, we show both theoretically and empirically, that these state-of-the-art detectors cannot reliably detect LLM outputs in practical scenarios." This was my finding as well. I abandoned this effort when I found that any created model would be inaccurate on small texts and could be circumvented with a bit of prompt engineering.