UPRISE: Universal Prompt Retrieval for Improving Zero-Shot Evaluation

Summary: Researchers have proposed a new method, UPRISE (Universal Prompt Retrieval for Improving zero-Shot Evaluation), for improving the generalization abilities of large language models (LLMs) by automatically retrieving prompts for zero-shot tasks without requiring model-specific fine-tuning or task-specific prompt engineering. The method was demonstrated to be universal in a cross-task and cross-model scenario, with the retriever being tuned on a diverse set of tasks and tested on unseen task types, and was shown to mitigate the hallucination problem in experiments with ChatGPT, suggesting its potential to improve even the strongest LLMs.