Jay van Zyl @ ecosystem.Ai

Jay van Zyl @ ecosystem.Ai

Shifting order in multiple-choice questions massively affects LLM performance

Recent research proposes that Large Language Models (LLMs) may not be as reliable as we think. In fact, the order of options in a multiple-choice question drastically influences the responses from LLMs such as GPT-4 and InstructGPT. If you want t…