Are Language Models Worse than Humans at Following Prompts? It's Complicated
This work cautions against idealizing human behavior in AI research, highlighting the need for empirical validation when training or evaluating models.
The study challenges the assumption that humans perform poorly on tasks with irrelevant or misleading prompts, finding that humans ignore irrelevant instructions but follow misleading ones, unlike language models.
Prompts have been the center of progress in advancing language models' zero-shot and few-shot performance. However, recent work finds that models can perform surprisingly well when given intentionally irrelevant or misleading prompts. Such results may be interpreted as evidence that model behavior is not "human like". In this study, we challenge a central assumption in such work: that humans would perform badly when given pathological instructions. We find that humans are able to reliably ignore irrelevant instructions and thus, like models, perform well on the underlying task despite an apparent lack of signal regarding the task they are being asked to do. However, when given deliberately misleading instructions, humans follow the instructions faithfully, whereas models do not. Our findings caution that future research should not idealize human behaviors as a monolith and should not train or evaluate models to mimic assumptions about these behaviors without first validating humans' behaviors empirically.