Mothman at SemEval-2024 Course of 9: An Iterative System for Chain-of-Thought Prompt Optimization
Authors: Alvin Po-Chun Chen, Ray Groshan, Sean von Bayern
Summary: Intensive analysis exists on the effectivity of large language fashions on logic-based duties, whereas comparatively little has been completed on their potential to generate creative decisions on lateral pondering duties. The BrainTeaser shared train assessments lateral pondering and makes use of adversarial datasets to forestall memorization, leading to poor effectivity for out-of-the-box fashions. We suggest a system for iterative, chain-of-thought quick engineering which optimizes prompts utilizing human analysis. Utilizing this shared train, we exhibit our system’s potential to considerably enhance mannequin effectivity by optimizing prompts and take note of the enter dataset
