Table 4. Blinded guess of question writer (i.e. AI vs human).

	AI (total = 50)	Human (total = 50)	Correlation	p
	(Correct guess, %)	(Correct guess, %)
Assessor A	24, 48%	23, 46%	- 0.14–0.26	0.55
Assessor B	14, 28%	41, 82%	- 0.38–0.10	0.24
Assessor C	33, 66%	24, 48%	- 0.35–0.06	0.16
Assessor D	27, 53%	26, 52%	- 0.26–0.14	0.55
Assessor E	26, 52%	32, 64%	- 0.36–0.04	0.11
GPT-2 Output Detector	7, 14%	45, 90%	- 0.40–0.21	0.54