GPT-3: Language Models are Few-shot Learners Abstract task-agnostic model 현재 아키텍쳐 자체는 task-agnostic (특별히 어떤 구체적인 태스크를 수행하라고 지시되지 않음) 하지만 fine-tunin
What NLPs are nownarrow expertsbrittle and sensitive to the data distribution or trivial changesSingle task training on single task datasetstill super