dai-2023-six-classification-datasets

Summary

This pins down exactly which text-classification tasks are being compared, covering sentiment, topic, and entailment, where in-context learning is tested against standard finetuning. It matters because any downstream claim about which method "wins" at classification is only as credible as the specific benchmarks behind it, so knowing the six tasks by name is the baseline for judging whether a conclusion actually applies to the problem you care about.