Chain-of-Thought: Making LLMs 'Show Their Work'
Chain-of-thought prompting makes an LLM 'show its work' by generating intermediate reasoning steps before the final answer. This simple few-shot technique dramatically improves performance on complex tasks like math word problems or commonsense questions, especially for very large models. The common footgun is applying it to smaller models, where it can actually degrade performance instead of helping, as the reasoning ability hasn't yet emerged.
Chain-of-thought (CoT) prompting transforms a large language model from a black-box guesser into a step-by-step reasoner. Instead of asking for just the final answer, you prompt the model to 'show its work' by generating the intermediate steps it would take to solve a problem. This is a powerful few-shot technique for complex reasoning; providing just a handful of examples demonstrating the step-by-step process can unlock significant performance gains in arithmetic and commonsense tasks without any model fine-tuning. The primary footgun is assuming CoT works on all models. It's an emergent ability of scale, and applying it to smaller models often fails to improve and can even hurt performance, as they lack the capacity to generate coherent reasoning chains.
Read the original → arXiv
- #llm
- #prompt engineering
- #reasoning
- #emergent abilities
Get five bites like this every day.
Tezvyn delivers a daily feed of 60-second tech bites with quizzes to lock in what you learn.