LibGuides: AI Tools for Academic Work: AI Prompts for Academics

A generative AI tool is more likely to produce the best response when given clear prompts or ask concise questions. Hence, formulating effective prompts is essential and plays a crucial role in getting the desired outputs from AI models. The diversity of AI prompts shows how adaptable AI is to handle different tasks across multiple domains.

The process of crafting prompts that direct the generative AI model to create the desired results is called Prompt Engineering. The approach involves iteratively improving or engineering the prompts until the AI generator produces the desired outcomes. Developing the prompts requires an awareness of the constraints and capabilities of the AI model.

To create effective prompts, it is imperative to take the context, audience, and desired outcomes into account. When incorporating AI tools into academia, the prompts ought to inspire scholars to interact with the tools critically and creatively. Typical tasks performed by large language models with various prompts include;

Text Summarization & Synthesizing
Text Generation
Image Generation
Question and Answering
Code Generation
Data Analyzing
Language Translation

Academic Prompts	Examples
Text generation and creative writing	Stimulate creativity: Generate a creative story about a future where humans coexist with intelligent robots. Technical documentation: Explain the architecture and operation of a neural network for natural language, highlighting its potential applications and disadvantages.
Question and Answer	Domain-Specific Knowledge Extraction: What are the key principles of quantum mechanics? Emphasize its applications in real life. Ambiguous Queries: Given the ambiguous query "How far is it from here?" provide a response that seeks clarification or additional context from the user.
Text Summarization	Research article: Summarize the key findings and methodologies used in the research paper titled "Deep Reinforcement Learning Approach for Trading Automation and Forecasting in the Stock Market.". Conference proceedings: Summarize the key arguments and conclusions from the engineering thesis "Integration of Machine Learning for Atomic Simulation and Activity Prediction in Heterogeneous Catalysis".
Programming Code Generation	Algorithm implementation: Write a Python function to solve 2D Heat Equation using numerical methods. Graph algorithms: Implement Johnson's algorithm in C++ for finding the shortest path in a edge-weighted graph.
Research and Analysis	Industry trend analysis: Analyze recent trends in renewable energy technology. Provide insights into emerging technologies, market dynamics, and potential implications for sustainability. Comparative Analysis: Compare the environmental impact of traditional agriculture and vertical farming methods. Analyze resource efficiency, carbon footprint, and potential scalability. Economic forecasting: Use economic indicators to forecast the growth of the technology sector in the next five years. Analyze key factors influencing the industry.
Interactive Learning	Concept explanation: Explain the concept of artificial neural networks and provide some examples for better understanding. Language correction: Generate a sentence describing a recent technological advancement in artificial intelligence. Provide feedback and corrections for grammar and clarity.
Thesis/Essay Writing	Abstract: Write a concise summary of the thesis "role of computer vision in healthcare", capturing the key elements of each chapter. Highlight the research problem, methodology, major findings, and conclusions. Provide a thesis statement: Support or refuse the statement: 'The use of artificial intelligence in culture raises interesting ethical reflections.' Ethical Dilemmas: Discuss the ethical challenges faced by healthcare professionals in allocating scarce resources during a pandemic. Discuss the principles guiding decision-making. Discussion and Conclusion: Summarize the key arguments and conclusions from the engineering thesis 'Integration of Machine Learning in Predictive Maintenance of Industrial Equipment. Provide recommendations for future research or interventions.
Image Generation	Image Translation: Explore image-to-image translation using CycleGAN. Choose a specific task, such as translating satellite images to maps or black-and-white photos to color. Image Captioning: Combine computer vision and natural language processing to create an image captioning model. Train the model to generate contextually rich captions for given images. Conditional Imaging: Implement a conditional image generation model using a Generative Adversarial Network (GAN). Train the model to generate images based on specific conditions, such as different classes or attributes.

Effective prompt techniques are essential to achieve the best result out of a generative AI tool. Modifying prompts according to a particular use case and intended result can have a great impact on the quality of AI-generated responses. In order to achieve that, bear in mind the following strategies:

Clear and Precise:

Strategy: Define your task or request clearly with specific details. Avoid any ambiguity to guide the AI tools to generate the desired output.
Example: Instead of "Translate this," use "Translate the following French paragraph to both Arabic and English."

Context and Background Information:

Strategy: Provide relevant context and background to enhance understanding. Include pertinent information or citations.
Example: Use "Generate a concise summary of the key findings and conclusions from this scientific research paper on climate change" instead of "Summarize this article,"

Use Iterative Refinement:

Strategy: Iteratively refine your prompt based on initial results. Analyze model responses and adjust prompts or rephrase the request accordingly.
Example: If the generated output is not adequate, give feedback and request the model to improve the response.

Prompt Length and Complexity:

Strategy: experiment with the length of prompts and complexities to find the optimal balance.
Example: Try short and direct prompts as well as longer, more detailed queries to observe variations in responses.

Use Domain-Specific Language:

Strategy: Use domain-specific jargon or keywords relevant to your request to match the model's output with the terminology expected in the topic.
Example: Rather than "Discover information on space," try "Retrieve comprehensive information about recent developments in space exploration."

Specify Output Format or Structure:

Strategy: State clearly the desired format for the model’s response.
Example: Instead of " Tell me about artificial intelligence," try " List the key concepts of artificial intelligence in bullet points."

Multistep Prompts:

Strategy: Split complicated requests into several steps. Provide clear instructions in the prompt, leading the model through each step.
Example: Instead of " How to set up Python” use "Compose a step-by-step guide for setting up a virtual environment in Python, covering installation and basic configuration”

Positive and Negative Reinforcement:

Strategy: Incorporate explicit positive or negative reinforcement into the prompt to direct the model's output towards the desired tone or attitude.
Example: Create a positive review for the given product. Highlight the features that users appreciate.

Analogical Reasoning

Strategy: Frame prompts that uses analogies to guide the model to make logical connections and establish comparisons in its responses.
Example: Analogous to a wave equation in the electromagnetic field, describe a prominent equation in the biophotonics for nuclear magnetism.

Use Case-Specific Prompts:

Strategy: Tailor prompts to the exact use case or field.
Example: Use "Give insights into the economic impact of renewable energy policies in the last decade" instead of "Provide renewable energy policies.".

Multi-turn Interactions

Strategy: Make an interaction with multiple turns by asking the model a series of questions or providing cues. This allows the model to construct context and produce logical responses.
Example: "Inquire about the weather, then ask for recommendations on outdoor activities based on the forecast."

Combine Pre-training and Fine-tuning Prompts

Strategy: Try combining cues from the pre-training and fine-tuning phases if you're using a pre-trained model. This can assist in optimizing the model for certain tasks while utilizing its prior expertise.