Prompt Engineering

来自ChatGPT的学习路径的建议

如果你想学习和掌握 Prompt Engineering（提示工程）技巧，下面是一个学习路径的建议：

1. 理解GPT模型和Prompt Engineering的基本概念：

   • 学习GPT模型的原理和工作机制，包括生成式预训练和微调等。
   • 了解Prompt Engineering的定义和目标，即如何设计和优化模型输入的提示。

2. 学习模型架构和配置：

   • 研究不同版本的GPT模型（如GPT-3、GPT-4等）的架构和配置，了解它们的特点和应用场景。
   • 学习模型参数和超参数的含义，以及它们对模型性能的影响。

3. 掌握Prompt Engineering的核心技巧：

   • 学习如何设计有效的提示语，包括问题格式、上下文设置、特定关键词的使用等。
   • 理解如何调整提示语的语言风格和语气，使其与特定任务或用户需求相匹配。

4. 学习Prompt Engineering的调试和优化方法：

   • 掌握评估模型输出的技巧，包括人工评估、指标评估和对话交互评估等。
   • 学习使用对抗训练（Adversarial Training）等方法来改进模型的生成质量和鲁棒性。

5. 实践和迭代改进：

   • 在实际应用中使用Prompt Engineering技巧，并通过实践来验证和改进模型的表现。
   • 尝试不同的Prompt Engineering策略和技术，比较它们的效果和适用性。

6. 学习和分享经验：

   • 参与Prompt Engineering社区和论坛，与其他研究者和从业者交流经验和最佳实践。
   • 阅读和研究相关的研究论文和技术博客，关注最新的Prompt Engineering研究动态。

7. 持续学习和关注最新发展：

   • 随着研究和技术的不断进展，保持持续学习的态度，关注最新的Prompt Engineering方法和技术。

请记住，Prompt Engineering是一个不断发展和演进的领域，因此保持对最新研究和实践的关注是非常重要的。通过不断实践和改进，你将逐渐掌握并提升自己在Prompt Engineering方面的技能

2. Prompt Guide

• Brex's Prompt Engineering Guide: Brex's introduction to language models and prompt engineering.
• promptingguide.ai: A prompt engineering guide that demonstrates many techniques.
• OpenAI Cookbook: Techniques to improve reliability: A slightly dated (Sep 2022) review of techniques for prompting language models.
• Lil'Log Prompt Engineering: An OpenAI researcher's review of the prompt engineering literature (as of March 2023).
• learnprompting.org: An introductory course to prompt engineering.

3. Prompt Video Course

• Andrew Ng's DeepLearning.AI: A short course on prompt engineering for developers.
• Andrej Karpathy's Let's build GPT: A detailed dive into the machine learning underlying GPT.
• Prompt Engineering by DAIR.AI: A one-hour video on various prompt engineering techniques.

4. Papers on advanced prompting to improve reasoning

• Chain-of-Thought Prompting Elicits Reasoning in Large Language Models (2022): Using few-shot prompts to ask models to think step by step improves their reasoning. PaLM's score on math word problems (GSM8K) rises from 18% to 57%.
• Self-Consistency Improves Chain of Thought Reasoning in Language Models (2022): Taking votes from multiple outputs improves accuracy even more. Voting across 40 outputs raises PaLM's score on math word problems further, from 57% to 74%, and code-davinci-002's from 60% to 78%.
• Tree of Thoughts: Deliberate Problem Solving with Large Language Models (2023): Searching over trees of step by step reasoning helps even more than voting over chains of thought. It lifts GPT-4's scores on creative writing and crosswords.
• Language Models are Zero-Shot Reasoners (2022): Telling instruction-following models to think step by step improves their reasoning. It lifts text-davinci-002's score on math word problems (GSM8K) from 13% to 41%.
• Large Language Models Are Human-Level Prompt Engineers (2023): Automated searching over possible prompts found a prompt that lifts scores on math word problems (GSM8K) to 43%, 2 percentage points above the human-written prompt in Language Models are Zero-Shot Reasoners.
• Reprompting: Automated Chain-of-Thought Prompt Inference Through Gibbs Sampling (2023): Automated searching over possible chain-of-thought prompts improved ChatGPT's scores on a few benchmarks by 0–20 percentage points.
• Faithful Reasoning Using Large Language Models (2022): Reasoning can be improved by a system that combines: chains of thought generated by alternative selection and inference prompts, a halter model that chooses when to halt selection-inference loops, a value function to search over multiple reasoning paths, and sentence labels that help avoid hallucination.
• STaR: Bootstrapping Reasoning With Reasoning (2022): Chain of thought reasoning can be baked into models via fine-tuning. For tasks with an answer key, example chains of thoughts can be generated by language models.
• ReAct: Synergizing Reasoning and Acting in Language Models (2023): For tasks with tools or an environment, chain of thought works better you prescriptively alternate between Reasoning steps (thinking about what to do) and Acting (getting information from a tool or environment).
• Reflexion: an autonomous agent with dynamic memory and self-reflection (2023): Retrying tasks with memory of prior failures improves subsequent performance.
• Demonstrate-Search-Predict: Composing retrieval and language models for knowledge-intensive NLP (2023): Models augmented with knowledge via a "retrieve-then-read" can be improved with multi-hop chains of searches.
• Improving Factuality and Reasoning in Language Models through Multiagent Debate (2023): Generating debates between a few ChatGPT agents over a few rounds improves scores on various benchmarks. Math word problem scores rise from 77% to 85%.

A. Reference

• Examples and guides for using the OpenAI API, https://github.com/openai/openai-cookbook