“…Weak supervision techniques obtain these noisy labels by tapping into heuristics (Ratner et al, 2017;Meng et al, 2018;Awasthi et al, 2020), feature annotation (Mann and McCallum, 2010), external knowledge bases (Hoffmann et al, 2011;Min et al, 2013), pretrained models (Bach et al, 2019;Zhang et al, 2021) and third-party tools (Lison et al, 2020). Moreover, weak supervision can be combined with the active learning framework (Gonsior et al, 2020) to select the most informative data to be annotated by humans and utilize weak supervision to decide noisy labels. Given LLMs' stunning zero-shot capabilities, our work explores the possibility of using them as a more efficient labeling source, thus freeing up resources to be reinvested in the research pipeline.…”