This was the case 4 years ago with GPT-2. Can't find the paper now, but the ratio was something like 48% vs 52% of people could tell whether an article was AI g

This was the case 4 years ago with GPT-2. Can't find the paper now, but the ratio was something like 48% vs 52% of people could tell whether an article was AI generated