Artificial Intelligence Toolkit for Faculty

Often referred to as "prompt engineering" (but not as technical as it may sound), prompting is the practice of designing effective inputs—or prompts—to elicit desired outputs from large language models like AI. In other words, prompt engineering is the practice of telling AI what you would like it to do. The more strategic and explicitly clear the prompt, the more useful the output.

According to "Prompt Engineering in ChatGPT: A Comprehensive Master Course":

The goal of prompt engineering is to enhance the performance, reliability, and usefulness of AI language models by crafting user inputs that guide the model towards generating accurate, relevant, and coherent responses. This process often involves iterative refinement, experimentation, and deep understanding of the model’s behavior and limitations.

Prompt engineering does not have to use technical language or require arcane knowledge. It is somewhat similar to holding a conversation: when we want to get somewhere in that dialogue, we try things, thinking about how to draw out our interlocutor as we go. As educators, we are used to dialogue and scaffolding, another similar strategy, to assist students.

Tips & Techniques

In her presentation Using ChatGPT in the Workplace, Anna Mills suggests the following:

1. Keep asking yourself whether the generative AI tool is useful (sometimes it’s better to ditch it).
2. Try again, or ask it for multiple answers.
3. Tell it what it gets wrong; ask it to redo.
4. Give it what you have to go on (bullet points, sources, text to respond to, examples).
5. Give it an example of the style you want: “Can you make that more friendly?” or "Can you write that in the voice of a community college student?"
6. Tell it to play a particular role such as a coach, mentor teacher, executive assistant, or editor (see Rob LennonLinks to an external site.’s post about this): “You are a supportive editor.”

A Helpful Formula

Prompt = role + task + instructions. Be specific!

These Large Language Models (LLMs) are useful for text generation, but some may be better than others...

	ChatGPT	Open AI Free, built on GPT-3.5 No web search results. Trained on data only through January 2022. As of April 2024, you can use it without creating an account.
	ChatGPT Plus	Open AI $20/mo, built on GPT-4 Includes: a "browse with Bing" feature that allows it to search for information on websites Advanced Data Analysis" (formerly Code Interpreter) GPT4-Vision (for uploading images and discussing them) Image generation with DALL-E 3. (as of Nov. 6, 2023: all these features combined into "all tools.") A new feature for building chatbots without using code. These are called "GPTs." (available to Plus users only) Enterprise edition for large businesses. Teams account for small business.
	Microsoft Copilot	Microsoft Free, built on GPT-4 (If you use Creative mode, Balanced mode = GPT3.5) Generate images with DALL-E 3 using Bing Image Creator or Microsoft Designer Includes web search results from Bing Search. Copilot Pro Plan with more features for individuals. Copilot for Microsoft 365 with more features for institutions and businesses.
	Gemini (formerly Bard)	Google Free, built on Gemini Pro (as of Dec. 6, 2023) Includes web search results from Google. Gemini Advanced is built on Gemini Ultra ($20/mo)
	Claude	Anthropic Free. Also available in Slack. Built on their own model: Claude 3 Sonnet. Pro version ($20/mo) is based on Claude 3 Opus. Focused on reliability and safety. No web search results. Trained on data through August 2023.
	Perplexity AI	Perplexity Free, built on GPT-3.5.  Includes web search results. Offers "Focus" choices such as YouTube, Reddit, Semantic Scholar ,and Wolfram|Alpha. You can use it without creating an account. Pro account ($20/mo) allows a choice between several models: GPT-4, Claude, Gemini Pro, and more.

Which models are the most capable?

• ChatGPT Plus (with a monthly fee) is based on GPT-4.
• Claude is also very capable, but no web search results. It's free to use and is trained on data up to August 2023.
• Microsoft Copilot (in creative mode) is based on GPT-4. And it's free to use.
• Perplexity AI with their pro account -- you can get access to more capable modesl (like GPT-4 and Claude 3).
• Gemini Advanced (with a monthly fee) based on Google's Gemini Ultra, is now said to be about as capable at GPT-4.

Which models are less capable?

• Google's Gemini (formerly known as Bard)  - the free version, is now about as capable as the free ChatGPT (3.5), since it uses Gemini Pro as the underlying model.
• ChatGPT free version (GPT 3.5)

For a list of more AI tools, see AI Tools Landscape by Carlos Lizarraga-Celaya.

Adapted from AI Literacy in the Age of ChatGPT, University of Arizona Libraries, CC BY 4.0

Text-to-image prompts using tools like DALL-E and Midjourney require different techniques from text-to-text Generative AI, as the output cannot be easily edited. Specifically, when using DALL-E, either in Chat-GPT Pro or for free in Copilot, any changes will mean a completely new image is generated. Getting it right can be time-consuming (and sometimes frustrating). So, the more detailed your prompt at the start, the more likely you will be to get the image you want with fewer tries. 

Here is the basic formula of details you should include in an image prompt: 

Write a clear description of the desired image...

• Subject - person, animal, object or other focal point).
• Setting - environment or background the subject is in (indoor, outdoor, fantasy, etc.)
• Artistic style and mood - realistic, surreal, impressionist, cartoon, paper-art, etc.
• Color scheme (if important)
• Additional details - time of day, specific additional objects on or near subject, weather conditions, etc.
• Diversity elements - If describing a person, include specifics of gender, age, ethnicity, race, etc.. 

Prompt Example:

"Create a detailed description of a photo in __________ art style. The photo should depict a ______________ against/inside/nearby a ______________. The/a _______________ should be prominently featured, showing intricate details such as/of/including _________________. The lighting should be ___________, casting shadows that add depth and dimension to the image. The composition should be ___________, with the main subject positioned ___________. The photo should evoke a sense of ___________ and ___________, capturing a moment of ___________. 

Tools for generating prompts:

You can also use an online tool, such as The Free AI Prompt Generator, to enter a basic idea of the type of image you are looking to create, and it will create additional details based on what image generation tool you want to use (ChatGPT/Bing Copilot, Midjourney, or Stable Diffusion).

Learn More

Watch DALLE-3 Masterclass 2024: Advanced AI Image Generation Tutorial for live examples of strategies to employ when prompting for AI image generation. 

DALL-E (in Bing Copilot or ChatGPT-4): DALL-E 3 is OpenAI’s latest iteration of its image and art generation AI tool. It is easy to use - you can either prompt it to generate a new image, or add an existing image and prompt to edit the image to meet certain specifics. You can access this tool for free in Bing's image creator or in ChatGPT Plus with a paid subscription.

Adobe Firefly: Adobe Firefly is built into a variety of Adobe products, and LAVC faculty have access through Adobe Creative Cloud. Visit firefly.adobe.com and sign in to your institutional Adobe account.

• While the other models such as DALL-E and Midjourney have been unclear about the source images that they used to train their AIs, Adobe has declared that it is only using images it has the right to use.
• Adobe also offers several editing tools to customize their AI-generated images.

Ideogram:  Ideogram is a free text-to-image generator. It supports a diverse set of image style tags and can render coherent text inside images. 

Midjourney: Midjourney is an image generator similar to DALL-E 3 and considered "state of the art." 

• Those comfortable with using Discord state Midjourney has the lowest learning curve. However, the tool may be challenging to use if not familiar with Discord. 
• Midjourney also does not currently offer a free version, with the basic plan starting at $10/month. 

Craiyon: Craiyon is a free AI image generator – previously known as DALL·E mini – that includes upgraded features, such as no watermark, on a subscription basis. 

• Free version includes ads on screen, and takes some time to create an image.

Elicit.org: Can be used for evaluating academic references such as authors, titles, abstracts, and citation data

Explain Paper: Uses machine learning to summarize scientific papers. It uses natural language to processing to extract the key findings and contributions of a paper and presents it in a simplified and easy-to-understand format

Gamma: Creates working presentations, documents, and webpages you can refine. Content created with free accounts will be watermarked.

Harmonize: Multimedia discussion tool available as an app inside Canvas designed to encourage more frequent and thoughtful student engagement.

AI in Education's GenAI Chatbot Prompt Library: A variety of prompts to help with lesson planning and administrative tasks.