arXiv:2309.16797 · 中英对照阅读
Promptbreeder: Self-Referential Self-Improvement via Prompt Evolution
中文速览
手工设计的提示词会明显影响大语言模型的表现,但很难保证每个领域都用得好,反复挑选也容易停在原地。研究者提出 Promptbreeder,让模型在训练题上不断生成、测试和筛选提示词,同时还让模型改进“怎样生成提示词”的指令,形成一边优化、一边改进优化方法的循环。实验显示,它在常见的数学和常识推理基准上超过了思维链、先计划再解题等提示方法,也能为仇恨言论分类演化出复杂提示词。它不需要改动模型参数,就能针对任务自动找到更合适的提示方式,为低成本提升不同领域的大模型表现提供了新思路。
摘要
Popular prompt strategies like Chain-of-Thought Prompting can dramatically improve the reasoning abilities of Large Language Models (LLMs) in various domains. However, such hand-crafted prompt-strategies are often sub-optimal. In this paper, we present Promptbreeder, a general-purpose self-referential self-improvement mechanism that evolves and adapts prompts for a given domain. Driven by an LLM, Promptbreeder mutates a population of task-prompts, evaluates them for fitness on a training set, and repeats this process over multiple generations to evolve task-prompts. Crucially, the mutation of these task-prompts is governed by mutation-prompts that the LLM generates and improves throughout evolution in a self-referential way. That is, Promptbreeder is not just improving task-prompts, but it is also improving the mutation-prompts that improve these task-prompts. Promptbreeder outperforms state-of-the-art prompt strategies such as Chain-of-Thought and Plan-and-Solve Prompting on commonly used arithmetic and commonsense reasoning benchmarks. Furthermore, Promptbreeder is able to evolve intricate task-prompts for the challenging problem of hate speech classification.
术语表
- Promptbreeder
- Promptbreeder
- Large Language Models (LLMs)
- 大型语言模型(LLM)
- self-referential self-improvement
- 自指式自我改进
- prompt strategy
- 提示策略
- prompt engineering
- 提示工程
- task-prompt
- 任务提示
- mutation-prompt
- 变异提示
- thinking-style
- 思维风格
- evolutionary algorithm
- 进化算法
- fitness
- 适应度
- evolutionary unit
- 进化单元
- mutation operator
- 变异算子
- Chain-of-Thought Prompting (CoT)
- 思维链提示(CoT)
- Zero-shot CoT
- 零样本思维链
- Self-Consistency (CoT-SC)
- 自洽性(CoT-SC)
- Tree of Thoughts (ToT)
- 思维树(ToT)
- Graph of Thoughts (GoT)
- 思维图(GoT)
- Plan-and-Solve Prompting (PS)
- 计划与求解提示(PS)
- Least-to-Most Prompting
- 由简到繁提示
- Self-Refine
- 自我精炼
- Soft Prompting
- 软提示
- Automatic Prompt Engineer (APE)
- 自动提示工程师(APE)
- Scratchpad method
- 草稿板方法
- GSM8K
- GSM8K
- commonsense reasoning
- 常识推理
- arithmetic reasoning
- 算术推理
- hate speech classification
- 仇恨言论分类