Table 1.

Performance comparison of GPT-4 zero-shot and few-shot on refined categories.

				Discriminative models					GPT-4 (with CoTa)
				LSTMb		BERTc		Zero-shot				Few-shot
Recall score
	Treatment			0.486		0.492				0.724		0.740
	Records			0.715		0.702				0.893		0.902
	Routine			0.764		0.779				0.834		0.899
	Rescheduling			0.769		0.721				0.896		0.837
	Symptoms			0.927		0.878				0.750		0.878
Total
		Accuracy	0.737		0.713		0.828				0.852

^aCoT: chain of thought.

^bLSTM: long short-term memory.

^cBERT: bidirectional encoder representations from transformers.