Zero-shot adaptive prompting of large language models

Recent advances in large language models (LLMs) are very promising as reflected in their capability for general problem-solving in few-shot and zero-shot setups, even without explicit training on these tasks. This is impressive because in the few-shot setup, LLMs are presented with only a few question-answer demonstrations prior to being given a test question. Even more challenging is the zero-shot setup, where the LLM is directly …

advances capability cloud cloud ai few-shot general language language models large language large language models llms machine learning natural-language understanding problem-solving prompting research researcher research scientist setup tasks team training