Skip to content

Changes

Page history
Create Prompt Engineering authored by Sarawoot Kongyoung's avatar Sarawoot Kongyoung
Hide whitespace changes
Inline Side-by-side
Prompt-Engineering.md 0 → 100644
View page @ b3a98234
**Prompt engineering** is an emerging field that focuses on the development and optimization of prompts to maximize the efficiency of language models (LMs) across various applications and research areas. It entails honing the skills necessary to understand the capabilities and limitations of large language models (LLMs). Through prompt engineering, researchers aim to enhance the performance of LLMs in a wide array of tasks, ranging from basic question answering to complex arithmetic reasoning. Developers also employ prompt engineering to devise robust and effective techniques for interacting with LLMs and other accompanying tools.
## Basic Prompting
Two fundamental approaches for prompting language models (LMs) are zero-shot learning and few-shot learning, which have been extensively explored in LLM research and commonly employed for evaluating LM performance.
### Zero-Shot
Zero-shot learning involves directly providing the task description to the model and requesting the corresponding results.
```
Example: Task - Topic Classification
Text: "Please classify the following text: 'The importance of renewable energy for a sustainable future.'"
Classification:
```
### Few-Shot
In few-shot learning, a set of high-quality demonstrations is presented to the model, comprising input examples and their desired outputs for the target task. By exposing the model to these informative examples, it gains a better understanding of human intentions and the criteria for desired answers. As a result, few-shot learning often leads to improved performance compared to zero-shot learning. However, it requires more token consumption and may encounter constraints on context length when the input and output text are lengthy.
```
Example: Task - Sentiment Analysis
Demonstration 1:
Input: "I absolutely loved the movie!"
Desired Output: Positive sentiment
Demonstration 2:
Input: "The book was incredibly boring."
Desired Output: Negative sentiment
Demonstration 3:
Input: "The restaurant served delicious food."
Desired Output: Positive sentiment
Prompt: Based on the given examples, classify the sentiment of the following text: "I had a terrible experience at the hotel."
Classification:
```
Numerous studies have explored techniques for constructing in-context examples to maximize performance. It has been observed that the choice of prompt format, training examples, and the order in which examples are presented can lead to significant variations in performance, ranging from nearly random guesses to near state-of-the-art results.
## Instruction Prompting
The purpose of incorporating few-shot examples in the prompt is to communicate our intent to the model effectively. It involves describing the task instructions to the model through demonstrations. However, few-shot learning can be costly in terms of token usage and imposes limitations on input length due to context constraints. So, why not directly provide the instructions?
Instructed LM (such as InstructGPT or natural instruction) involves fine-tuning a pretrained model using high-quality tuples of (task instruction, input, ground truth output) to enhance the model's understanding of user intent and its ability to follow instructions. Reinforcement Learning from Human Feedback (RLHF) is a common approach used for this purpose. The advantage of instruction-following style fine-tuning is that it aligns the model more closely with human intention, significantly reducing the communication cost.
When interacting with instruction models, it is crucial to provide detailed and specific task requirements. It is important to be precise in the instructions and avoid using negations (such as "not do something") but rather clearly specify what actions should be taken.
To label the sentiment towards the movie in the provided movie review, please indicate whether the sentiment is "positive" or "negative."
```
Review Text: "I'll bet the video game is a lot more fun than the film."
Sentiment Label:
```
In order to provide clear instructions, it is essential to consider the intended audience. For instance, when creating educational materials for children, you might describe "quantum physics" in a way that a 6-year-old can understand. Similarly, for content intended for professional settings, it is important to ensure the language used is appropriate for the workplace.
In-context instruction learning, as demonstrated by Ye et al. in 2023, combines the advantages of few-shot learning with instruction prompting. This approach incorporates multiple demonstration examples within the prompt, encompassing instructions, task inputs, and outputs. It is worth noting that their experiments focused solely on classification tasks, and the instruction prompt includes all possible label options.
Examples:
```
Definition: Determine the speaker of the dialogue as either "agent" or "customer."
Input: "I have successfully booked your tickets."
Output:
Definition: Determine the category of the question, either "Quantity" or "Location."
Input: "What's the oldest building in the US?"
Output:
Definition: Classify the sentiment of the given movie review as "positive" or "negative."
Input: "I'll bet the video game is a lot more fun than the film."
Output:
```
## Prompt Engineering Paper
- **Prompt Engineering Techniques**:
- [A Prompt Pattern Catalog to Enhance Prompt Engineering with ChatGPT](https://arxiv.org/abs/2302.11382) [2023] (Arxiv)
- [Hard Prompts Made Easy: Gradient-Based Discrete Optimization for Prompt Tuning and Discovery](https://arxiv.org/abs/2302.03668) [2023] (Arxiv)
- [Synthetic Prompting: Generating Chain-of-Thought Demonstrations for Large Language Models](https://arxiv.org/abs/2302.00618) [2023] (Arxiv)
- [Progressive Prompts: Continual Learning for Language Models](https://arxiv.org/abs/2301.12314) [2023] (Arxiv)
- [Batch Prompting: Efficient Inference with LLM APIs](https://arxiv.org/abs/2301.08721) [2023] (Arxiv)
- [Successive Prompting for Decompleting Complex Questions](https://arxiv.org/abs/2212.04092) [2022] (Arxiv)
- [Structured Prompting: Scaling In-Context Learning to 1,000 Examples](https://arxiv.org/abs/2212.06713) [2022] (Arxiv)
- [Large Language Models Are Human-Level Prompt Engineers](https://arxiv.org/abs/2211.01910) [2022] (Arxiv)
- [Ask Me Anything: A simple strategy for prompting language models](https://paperswithcode.com/paper/ask-me-anything-a-simple-strategy-for) [2022] (Arxiv)
- [Prompting GPT-3 To Be Reliable](https://arxiv.org/abs/2210.09150) [2022](Arxiv)
- [Decomposed Prompting: A Modular Approach for Solving Complex Tasks](https://arxiv.org/abs/2210.02406) [2022] (Arxiv)
- [PromptChainer: Chaining Large Language Model Prompts through Visual Programming](https://arxiv.org/abs/2203.06566) [2022] (Arxiv)
- [Investigating Prompt Engineering in Diffusion Models](https://arxiv.org/abs/2211.15462) [2022] (Arxiv)
- [Show Your Work: Scratchpads for Intermediate Computation with Language Models](https://arxiv.org/abs/2112.00114) [2021] (Arxiv)
- [Reframing Instructional Prompts to GPTk's Language](https://arxiv.org/abs/2109.07830) [2021] (Arxiv)
- [Fantastically Ordered Prompts and Where to Find Them: Overcoming Few-Shot Prompt Order Sensitivity](https://arxiv.org/abs/2104.08786) [2021] (Arxiv)
- [The Power of Scale for Parameter-Efficient Prompt Tuning](https://arxiv.org/abs/2104.08691) [2021] (Arxiv)
- [Prompt Programming for Large Language Models: Beyond the Few-Shot Paradigm](https://arxiv.org/abs/2102.07350) [2021] (Arxiv)
- [Prefix-Tuning: Optimizing Continuous Prompts for Generation](https://arxiv.org/abs/2101.00190) [2021] (Arxiv)
- **Reasoning and In-Context Learning**:
- [Multimodal Chain-of-Thought Reasoning in Language Models](https://arxiv.org/abs/2302.00923) [2023] (Arxiv)
- [On Second Thought, Let's Not Think Step by Step! Bias and Toxicity in Zero-Shot Reasoning](https://arxiv.org/abs/2212.08061) [2022] (Arxiv)
- [ReAct: Synergizing Reasoning and Acting in Language Models](https://arxiv.org/abs/2210.03629) [2022] (Arxiv)
- [Language Models Are Greedy Reasoners: A Systematic Formal Analysis of Chain-of-Thought](https://arxiv.org/abs/2210.01240v3) [2022] (Arxiv)
- [On the Advance of Making Language Models Better Reasoners](https://arxiv.org/abs/2206.02336) [2022] (Arxiv)
- [Large Language Models are Zero-Shot Reasoners](https://arxiv.org/abs/2205.11916) [2022] (Arxiv)
- [Reasoning Like Program Executors](https://arxiv.org/abs/2201.11473) [2022] (Arxiv)
- [Self-Consistency Improves Chain of Thought Reasoning in Language Models](https://arxiv.org/abs/2203.11171) [2022] (Arxiv)
- [Rethinking the Role of Demonstrations: What Makes In-Context Learning Work?](https://arxiv.org/abs/2202.12837) [2022] (Arxiv)
- [Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answering](https://arxiv.org/abs/2209.09513v2) [2022] (Arxiv)
- [Chain of Thought Prompting Elicits Reasoning in Large Language Models](https://arxiv.org/abs/2201.11903) [2021] (Arxiv)
- [Generated Knowledge Prompting for Commonsense Reasoning](https://arxiv.org/abs/2110.08387) [2021] (Arxiv)
- [BERTese: Learning to Speak to BERT](https://aclanthology.org/2021.eacl-main.316) [2021] (Acl)
- **Evaluating and Improving Language Models**:
- [Large Language Models Can Be Easily Distracted by Irrelevant Context](https://arxiv.org/abs/2302.00093) [2023] (Arxiv)
- [Crawling the Internal Knowledge-Base of Language Models](https://arxiv.org/abs/2301.12810) [2023] (Arxiv)
- [Discovering Language Model Behaviors with Model-Written Evaluations](https://arxiv.org/abs/2212.09251) [2022] (Arxiv)
- [Calibrate Before Use: Improving Few-Shot Performance of Language Models](https://arxiv.org/abs/2102.09690) [2021] (Arxiv)
- **Applications of Language Models**:
- [Prompting for Multimodal Hateful Meme Classification](https://arxiv.org/abs/2302.04156) [2023] (Arxiv)
- [PLACES: Prompting Language Models for Social Conversation Synthesis](https://arxiv.org/abs/2302.03269) [2023] (Arxiv)
- [Commonsense-Aware Prompting for Controllable Empathetic Dialogue Generation](https://arxiv.org/abs/2302.01441) [2023] (Arxiv)
- [PAL: Program-aided Language Models](https://arxiv.org/abs/2211.10435) [2023](Arxiv)
- [Legal Prompt Engineering for Multilingual Legal Judgement Prediction](https://arxiv.org/abs/2212.02199) [2023] (Arxiv)
- [Conversing with Copilot: Exploring Prompt Engineering for Solving CS1 Problems Using Natural Language](https://arxiv.org/abs/2210.15157) [2022] (Arxiv)
- [Plot Writing From Scratch Pre-Trained Language Models](https://aclanthology.org/2022.inlg-main.5) [2022] (Acl)
- [AutoPrompt: Eliciting Knowledge from Language Models with Automatically Generated Prompts](https://arxiv.org/abs/2010.15980) [2020] (Arxiv)
- **Threat Detection and Adversarial Examples**:
- [Constitutional AI: Harmlessness from AI Feedback](https://arxiv.org/abs/2212.08073) [2022] (Arxiv)
- [Ignore Previous Prompt: Attack Techniques For Language Models](https://arxiv.org/abs/2211.09527) [2022] (Arxiv)
- [Machine Generated Text: A Comprehensive Survey of Threat Models and Detection Methods](https://arxiv.org/abs/2210.07321) [2022] (Arxiv)
- [Evaluating the Susceptibility of Pre-Trained Language Models via Handcrafted Adversarial Examples](https://arxiv.org/abs/2209.02128) [2022] (Arxiv)
- [Toxicity Detection with Generative Prompt-based Inference](https://arxiv.org/abs/2205.12390) [2022] (Arxiv)
- [How Can We Know What Language Models Know?](https://direct.mit.edu/tacl/article/doi/10.1162/tacl_a_00324/96460/How-Can-We-Know-What-Language-Models-Know) [2020] (Mit)
- **Few-shot Learning and Performance Optimization**:
- [Promptagator: Few-shot Dense Retrieval From 8 Examples](https://arxiv.org/abs/2209.11755) [2022] (Arxiv)
- [The Unreliability of Explanations in Few-shot Prompting for Textual Reasoning](https://arxiv.org/abs/2205.03401) [2022] (Arxiv)
- [Making Pre-trained Language Models Better Few-shot Learners](https://aclanthology.org/2021.acl-long.295) [2021] (Acl)
- [Language Models are Few-Shot Learners](https://arxiv.org/abs/2005.14165) [2020] (Arxiv)
- **Text to Image Generation**:
- [A Taxonomy of Prompt Modifiers for Text-To-Image Generation](https://arxiv.org/abs/2204.13988) [2022] (Arxiv)
- [Design Guidelines for Prompt Engineering Text-to-Image Generative Models](https://arxiv.org/abs/2109.06977) [2021] (Arxiv)
- [High-Resolution Image Synthesis with Latent Diffusion Models](https://arxiv.org/abs/2112.10752) [2021] (Arxiv)
- [DALL·E: Creating Images from Text](https://arxiv.org/abs/2102.12092) [2021] (Arxiv)
- **Text to Music/Sound Generation**:
- [MusicLM: Generating Music From Text](https://arxiv.org/abs/2301.11325) [2023] (Arxiv)
- [ERNIE-Music: Text-to-Waveform Music Generation with Diffusion Models](https://arxiv.org/pdf/2302.04456) [2023] (Arxiv)
- [Noise2Music: Text-conditioned Music Generation with Diffusion Models](https://arxiv.org/abs/2301.11325) [2023) (Arxiv)
- [AudioLM: a Language Modeling Approach to Audio Generation](https://arxiv.org/pdf/2209.03143) [2023] (Arxiv)
- [Make-An-Audio: Text-To-Audio Generation with Prompt-Enhanced Diffusion Models](https://arxiv.org/pdf/2301.12661.pdf) [2023] (Arxiv)
- **Text to Video Generation**:
- [Dreamix: Video Diffusion Models are General Video Editors](https://arxiv.org/pdf/2302.01329.pdf) [2023] (Arxiv)
- [Tune-A-Video: One-Shot Tuning of Image Diffusion Models for Text-to-Video Generation](https://arxiv.org/pdf/2212.11565.pdf) [2022] (Arxiv)
- [Noise2Music: Text-conditioned Music Generation with Diffusion Models](https://arxiv.org/abs/2301.11325) [2023) (Arxiv)
- [AudioLM: a Language Modeling Approach to Audio Generation](https://arxiv.org/pdf/2209.03143) [2023] (Arxiv)
- **Overviews**:
- [Piloting Copilot and Codex: Hot Temperature, Cold Prompts, or Black Magic?](https://arxiv.org/abs/2210.14699) [2022] (Arxiv)
## Prompt Engineering Tools & Code
| Name | Description | Url |
| :-------------------- | :----------: | :----------: |
| **GPT Index** | GPT Index is a project consisting of a set of data structures designed to make it easier to use large external knowledge bases with LLMs. | [[Github]](https://github.com/jerryjliu/gpt_index) |
| **Promptify** | Solve NLP Problems with LLM's & Easily generate different NLP Task prompts for popular generative models like GPT, PaLM, and more with Promptify | [[Github]](https://github.com/promptslab/Promptify) |
| **Arize-Phoenix** | Open-source tool for ML observability that runs in your notebook environment. Monitor and fine tune LLM, CV and Tabular Models. | [[Github]](https://github.com/Arize-ai/phoenix) |
| **Better Prompt** | Test suite for LLM prompts before pushing them to PROD | [[Github]](https://github.com/krrishdholakia/betterprompt) |
| **Interactive Composition Explorerx** | ICE is a Python library and trace visualizer for language model programs. | [[Github]](https://github.com/oughtinc/ice) |
| **Haystack** | Open source NLP framework to interact with your data using LLMs and Transformers. | [[Github]](https://github.com/deepset-ai/haystack) |
| **LangChainx** | Building applications with LLMs through composability | [[Github]](https://github.com/hwchase17/langchain) |
| **OpenPrompt** | An Open-Source Framework for Prompt-learning | [[Github]](https://github.com/thunlp/OpenPrompt) |
| **Prompt Engine** | This repo contains an NPM utility library for creating and maintaining prompts for Large Language Models (LLMs). | [[Github]](https://github.com/microsoft/prompt-engine) |
| **PromptInject** | PromptInject is a framework that assembles prompts in a modular fashion to provide a quantitative analysis of the robustness of LLMs to adversarial prompt attacks. | [[Github]](https://github.com/agencyenterprise/PromptInject) |
| **Prompts AI** | Advanced playground for GPT-3 | [[Github]](https://github.com/sevazhidkov/prompts-ai) |
| **Prompt Source** | PromptSource is a toolkit for creating, sharing and using natural language prompts. | [[Github]](https://github.com/bigscience-workshop/promptsource) |
| **ThoughtSource** | A framework for the science of machine thinking | [[Github]](https://github.com/OpenBioLink/ThoughtSource) |
| **PROMPTMETHEUS** | One-shot Prompt Engineering Toolkit | [[Tool]](https://promptmetheus.com) |
## Tutorials
- **Introduction to Prompt Engineering**
- [Prompt Engineering 101 - Introduction and resources](https://www.linkedin.com/pulse/prompt-engineering-101-introduction-resources-amatriain)
- [Prompt Engineering 101](https://humanloop.com/blog/prompt-engineering-101)
- [Prompt Engineering Guide by SudalaiRajkumar](https://github.com/SudalaiRajkumar/Talks_Webinars/blob/master/Slides/PromptEngineering_20230208.pdf)
- **Beginner's Guide to Generative Language Models**
- [A beginner-friendly guide to generative language models - LaMBDA guide](https://aitestkitchen.withgoogle.com/how-lamda-works)
- [Generative AI with Cohere: Part 1 - Model Prompting](https://txt.cohere.ai/generative-ai-part-1)
- **Best Practices for Prompt Engineering**
- [Best practices for prompt engineering with OpenAI API](https://help.openai.com/en/articles/6654000-best-practices-for-prompt-engineering-with-openai-api)
- [How to write good prompts](https://andymatuschak.org/prompts)
- **Complete Guide to Prompt Engineering**
- [A Complete Introduction to Prompt Engineering for Large Language Models](https://www.mihaileric.com/posts/a-complete-introduction-to-prompt-engineering)
- [Prompt Engineering Guide: How to Engineer the Perfect Prompts](https://richardbatt.co.uk/prompt-engineering-guide-how-to-engineer-the-perfect-prompts)
- **Technical Aspects of Prompt Engineering**
- [3 Principles for prompt engineering with GPT-3](https://www.linkedin.com/pulse/3-principles-prompt-engineering-gpt-3-ben-whately)
- [A Generic Framework for ChatGPT Prompt Engineering](https://medium.com/@thorbjoern.heise/a-generic-framework-for-chatgpt-prompt-engineering-7097f6513a0b)
- [Methods of prompt programming](https://generative.ink/posts/methods-of-prompt-programming)
- **Resources for Prompt Engineering**
- [Awesome ChatGPT Prompts](https://github.com/f/awesome-chatgpt-prompts)
- [Best 100+ Stable Diffusion Prompts](https://mpost.io/best-100-stable-diffusion-prompts-the-most-beautiful-ai-text-to-image-prompts)
- [DALLE Prompt Book](https://dallery.gallery/the-dalle-2-prompt-book)
- [OpenAI Cookbook](https://github.com/openai/openai-cookbook)
- [Prompt Engineering by Microsoft](https://microsoft.github.io/prompt-engineering)
## Videos
- [Advanced ChatGPT Prompt Engineering](https://www.youtube.com/watch?v=bBiTR_1sEmI)
- [ChatGPT: 5 Prompt Engineering Secrets For Beginners](https://www.youtube.com/watch?v=2zg3V66-Fzs)
- [CMU Advanced NLP 2022: Prompting](https://youtube.com/watch?v=5ef83Wljm-M&feature=shares)
- [Prompt Engineering - A new profession ?](https://www.youtube.com/watch?v=w102J3_9Bcs&ab_channel=PatrickDebois)
- [ChatGPT Guide: 10x Your Results with Better Prompts](https://www.youtube.com/watch?v=os-JX1ZQwIA)
- [Language Models and Prompt Engineering: Systematic Survey of Prompting Methods in NLP](https://youtube.com/watch?v=OsbUfL8w-mo&feature=shares)
- [Prompt Engineering 101: Autocomplete, Zero-shot, One-shot, and Few-shot prompting](https://youtube.com/watch?v=v2gD8BHOaX4&feature=shares)
## References
- [Awesome-Prompt-Engineering](https://github.com/promptslab/Awesome-Prompt-Engineering)
- [Prompt Engineering](https://lilianweng.github.io/posts/2023-03-15-prompt-engineering/)
- [https://github.com/dair-ai/Prompt-Engineering-Guide](https://github.com/dair-ai/Prompt-Engineering-Guide)