Table 3. Single-dataset classifiers.

Accuracy scores of the classifiers evaluated on generations from the different language models. Along the diagonal (bold), training and test data belong to the same language model.

Training data	Test data
	s		xl		s-k		xl-k		GPT3		Grover
	Acc.	AUC	Acc.	AUC	Acc.	AUC	Acc.	AUC	Acc.	AUC	Acc.	AUC
s	0.897	0.964	0.728	0.838	0.487	0.302	0.471	0.290	0.475	0.474	0.479	0.454
xl	0.740	0.937	0.759	0.836	0.504	0.434	0.489	0.382	0.468	0.423	0.516	0.485
s-k	0.338	0.247	0.445	0.328	0.927	0.975	0.808	0.924	0.537	0.769	0.502	0.671
xl-k	0.292	0.223	0.382	0.322	0.908	0.967	0.858	0.932	0.535	0.545	0.503	0.514
GPT3	0.436	0.234	0.452	0.316	0.736	0.821	0.658	0.749	0.779	0.859	0.589	0.654
Grover	0.333	0.285	0.439	0.422	0.662	0.785	0.643	0.738	0.537	0.552	0.692	0.767