Hey all, Joe from Hugging Face here. Even before GPT-3's release last week there's been some interest in the NLP community recently in understanding learning in low-data regimes. For those of you interested in reading about the tasks of zero- and few-shot learning, I wrote up a blog post + demo covering a few recent methods using transformers.
Zero-shot topic classification demo: https://huggingface.co/zero-shot/
I hope some of you find it useful and/or insightful!