Table 6.

Balanced Accuracy Performance Change with Few-shot Prompting. This table is calculated between the zero-shot and the few-shot sections of Table 4.

Dataset	Dreaddit	DepSeverity
Model	Task #1	Task #2	Task #3	$\overline{\Delta }$–All Three Tasks
$\Delta -{\text{Alpaca}}_{FS\_vs\_ZS}$	↑ +0.039	↑ +0.007	↑ +0.197	↑ +0.081
$\Delta -{\text{FLAN-T5}}_{FS\_vs\_ZS}$	↑ +0.127	↑ +0.014	↑ +0.036	↑ +0.059
$\Delta -\text{GPT-3}{\text{.5}}_{FS\_vs\_ZS}$	↑ +0.036	↑ +0.023	↓ −0.023	↑ +0.012
$\Delta -{\text{GPT-4}}_{FS\_vs\_ZS}$	↓ −0.002	↑ +0.005	↑ +0.025	↑ +0.009
$\overline{\Delta }$–All Four Models	↑ +0.051	↑ +0.012	↑ +0.059	↑ +0.041